University of Nebraska - Lincoln DigitalCommons@University of Nebraska - Lincoln

Biological Systems Engineering: Papers and Publications Biological Systems Engineering

2020

Ground and Aerial for Agricultural Production: Opportunities and Challenges

Santosh Pitla University of Nebraska-Lincoln, [email protected]

Sreekala Bajwa Montana State University, [email protected]

Santosh Bhusal Washington State University

Tom Brumm Iowa State University, [email protected]

Tami M. Brown-Brandl University of Nebraska-Lincoln, [email protected]

FSeeollow next this page and for additional additional works authors at: https:/ /digitalcommons.unl.edu/biosysengfacpub

Part of the Bioresource and Commons, Environmental Engineering Commons, and the Other Civil and Environmental Engineering Commons

Pitla, Santosh; Bajwa, Sreekala; Bhusal, Santosh; Brumm, Tom; Brown-Brandl, Tami M.; Buckmaster, Dennis R.; Condotta, Isabella; Fulton, John; Janzen, Todd J.; Karkee, Manoj; Lopez, Mario; Moorhead, Robert; Sama, Michael; Schumacher, Leon; Shearer, Scott; and Thomasson, Alex, "Ground and Aerial Robots for Agricultural Production: Opportunities and Challenges" (2020). Biological Systems Engineering: Papers and Publications. 727. https://digitalcommons.unl.edu/biosysengfacpub/727

This Article is brought to you for free and open access by the Biological Systems Engineering at DigitalCommons@University of Nebraska - Lincoln. It has been accepted for inclusion in Biological Systems Engineering: Papers and Publications by an authorized administrator of DigitalCommons@University of Nebraska - Lincoln. Authors Santosh Pitla, Sreekala Bajwa, Santosh Bhusal, Tom Brumm, Tami M. Brown-Brandl, Dennis R. Buckmaster, Isabella Condotta, John Fulton, Todd J. Janzen, Manoj Karkee, Mario Lopez, Robert Moorhead, Michael Sama, Leon Schumacher, Scott Shearer, and Alex Thomasson

This article is available at DigitalCommons@University of Nebraska - Lincoln: https://digitalcommons.unl.edu/ biosysengfacpub/727 Number 70 Issue Paper November 2020 Ground and Aerial Robots for Agricultural Production: Opportunities and Challenges

Ground and aerial robots combined with (AI) techniques have potential to tackle the rising food, fiber, and fuel demands of the rapidly growing population that is slated to be around 10 billion by the year 2050. (Photo illustration by MeganW ickham with images from Santosh Pitla and Shutterstock.)

ABSTRACT which subsequently laid foundation for in and animal to in- mechanized agriculture. While preci- crease input efficiency and agricultural Crop and animal production tech- sion agriculture has enabled site-specific output with reduced adverse impact on niques have changed significantly over management of crop inputs for improved the environment. Ground and aerial ro- the last century. In the early 1900s, yields and quality, precision bots combined with artificial intelligence animal power was replaced by trac- farming has boosted efficiencies in (AI) techniques have potential to tackle tor power that resulted in tremendous animal and dairy industries. By 2020, the rising food,fi ber, and fuel demands improvements in field productivity, highly automated systems are employed of the rapidly growing population that is

This publication was made possible through funding provided by the United Soybean Board (“USB”). As stipulated in the Soybean Promotion, Research, and Consumer Information Act, USDA’s Agricultural Marketing Service (“AMS”) has oversight responsibilities for USB. AMS prohibits the use of USB’s funds to influence legislation and/or to influence governmental policy or action. Any opinions,fi ndings, conclusions, or recommendations expressed in this publication are those of the author(s) and do not necessarily reflect the views of USB, USDA, and/or AMS. CAST Issue Paper 70T ask Force Members

Authors Dr. John Fulton, Associate Professor, Department Head, Mississippi State Ohio State University, Columbus, Ohio University, Mississippi State, Missis- Dr. Santosh Pitla, Chair, Associate Todd J. Janzen, President, Janzen Agri- sippi Professor, University of Nebraska, cultural Law, LLC, Indianapolis, Indiana Lincoln, Nebraska Dr. Manoj Karkee, Associate Professor, Reviewers Dr. Sreekala Bajwa, Vice President, Washington State University, Prosser, Dean & Director, Montana State Uni- Dr. Srinivasulu Ale, Associate Profes- Washington versity, Bozeman, Montana sor, Texas A&M AgriLife Research Dr. Mario Lopez, Technical Develop- Santosh Bhusal, Ph.D. Candidate, Dr. Ed Barnes, Senior Director of ment Manager MQ&OFSS, DeLaval, Washington State University, Pullman, Agricultural and Environmental Kansas City, Missouri Washington Research, Cotton Incorporated, Cary, Dr. Robert Moorhead, Director, Geo- Dr. Tom Brumm, Associate Profes- North Carolina systems Research Institute, Mississippi sor, Iowa State University, Ames, Dr. Brian Steward, Professor, Iowa State University, Mississippi State, Iowa State University, Ames, Iowa Mississippi Dr. Tami Brown-Brandl, Profes- Dr. Michael Sama, Associate Profes- sor, University of Nebraska-Lincoln, CAST Liaison sor, University of Kentucky, Lexington, Lincoln, Nebraska Kentucky Dr. Carroll Moseley, Syngenta Dr. Dennis R. Buckmaster, Profes- Dr. Leon Schumacher, Professor, Uni- Crop Protection, Greensboro, North sor, Purdue University, West versity of Missouri, Columbia, Missouri Carolina Lafayette, Indiana Dr. Scott Shearer, Professor and Chair, Dr. Juan Ticarico, Innovation Center Dr. Isabella Condotta, Assistant Ohio State University, Columbus, Ohio for U.S. Dairy, Rosemont, Illinois Professor, University of Illinois, Champaign, Illinois Dr. Alex Thomasson, Professor and

slated to be around 10 billion by the year 2050. This Issue Paper presents oppor- tunities provided by ground and aerial robots for improved crop and animal production, and the challenges that could potentially limit their progress and adop- tion. A summary of enabling factors that could drive the deployment and adoption of robots in agriculture is also presented along with some insights into the training Figure 1. Classification and Important Application Domains of Agricultural Robots. needs of the workforce who will be in- volved in the next-generation agriculture. address these impending local and global cient crop varieties in research plots, and INTRODUCTION demands (Tilman et al. 2001). Efficiently be more efficient in managing livestock managing crop inputs such as seeds, (Chen 2018). Motivated by these op- AND BACKGROUND chemicals, , and labor to boost portunities, numerous teams around the The increase in demand for food, feed, yields and quality while minimizing world have been working on research and fiber, and energy is inevitable, given environmental impacts is important to development of various types of robotic the projected population increase in the support the on-going biotechnology ef- solutions for crop and animal farming. coming decades as well as increasing forts. Agricultural robots combined with Figure 1 provides a broad classifica- affluence in developing economies of the techniques are seen tion of such agricultural robots. Ground world. Food production has to increase as an effective method for monitoring robots or unmanned ground vehicles by approximately 70% to feed 9.7 bil- large acreages to assess crop conditions, (UGVs) and aerial robots or Unmanned lion people by the year 2050 (Bruinsma increase precise management of crop Aerial Systems (UASs) are used in both 2009). Advancements in biotechnology inputs in row and specialty , allow row-crops and specialty crops. Robotic are seen as one of the potential solu- efficient harvesting in specialty crops, manipulators (robotic arms) on the other tions for increasing food production to evaluate high yielding and resource effi- hand are primarily used in dairy, specialty

2 COUNCIL FOR AND TECHNOLOGY crop, and applications. Each physically entering the field, or provides Producers interested in using aerial of these categories along with some a more efficient way to count or analyze robots on their will need to decide examples are discussed in detail in the livestock. UASs can quickly determine whether to operate their own UAS or hire next section. soil moisture and crop health over the a UAS service provider. Service provid- entirety of the field with great preci- ers can standardize operational variables Ground Robots sion. Similarly, UAS-based systems can to meet data quality and application needs be used to measure animal temperature and amortize the cost of the technology Changing weather patterns and the quickly and non-invasively, and thereby over more users per unit time, whereas need to boost agricultural productiv- identify which animal might be in heat or self-operation provides greaterfl exibility ity require a paradigm shift in how have an illness. As miniaturized sensors, in use and access. Similarly, there are field operations are accomplished. For such as microwave radiometers, short- trade-offs in running analyses locally or example, UGVs could perform repetitive wave infrared sensors, and (light using cloud computing. operations with great precision for up to detection and ranging), are integrated 22 hours per day (with a small portion of with UASs, opportunities to measure soil Robotic Manipulators that time for service and maintenance) moisture and crop health precisely, map Robotic manipulators are mechanisms which could substantially increase the field topography, and locate irrigation that manipulate objects in a given work- daily work output compared to human- system leaks on afi eld-by-field scale, space and are typically found in industrial operated machinery. Deploying UGVs will become affordable to the producer. applications for performing for crop and livestock production is a LiDARs can measure height, often automated movements Robotic manipula- tremendous opportunity to boost pro- an indicator of crop health. data tors are now becoming indispensable for ductivity. However, deploying UGVs can also be used to measure livestock automating precise repetitive motions in that are capable of decision making and growth and health. As researchers de- both animal and crop production. Robotic working in remote harshfi eld conditions velop algorithms to exploit the increased milking stations used in dairy production intelligently and safely comes with both spatial resolution of low-flying UAS, are good examples of type complex engineering and socio-economic increased opportunities will occur to that are commercially avail- issues. Some important engineering reduce inputs to crops by applying inputs able. These stations challenges include: (1) designing effec- to only areas where and when the input is use proximity sensors, robotic arms and tive control architectures and artificial needed. Crop problems such as diseases suction cups for automating the milking intelligence (AI) algorithms for operat- could also be detected earlier before process (Broucek and Tongel 2015). Ro- ing individual and afl eet of UGVs with they become widespread. Better spatial botic arms are a type of manipulators that and without humans in the loop and (2) resolution of livestock images leads to mimic hand movements of a human for designing and deploying robots to work enhanced ability to detect needs of each picking and placing of objects. In crop faster and more gently than human labor animal individually. production, robotic arms are predomi- when interacting with and ani- UASs, however, also suffer from some nantly deployed in and green- mals. Some important socio-economic important limitations. One of the chal- house applications. In , picking challenges include: (1) Fear associated lenges to exploiting UASs in agriculture of fruit is accomplished by mounting with robots in terms of replacing human production will be weather related—most robotic arms on either a ground robot or labor (2) Economic feasibility of robots small UASs cannot operate in high other vehicles. The ground robot for both large and small acreage farms winds or medium precipitation. Another navigates to the fruit tree and the robotic (3) Liability, security, and data privacy challenge will be the acquisition and arm performs the operation of picking the concerns associated with UGVs. maintenance of computing resources to fruit. Robotic arms have been developed rapidly process images into information. for picking vegetables such as toma- Aerial Robots Determining the optimal suite of UASs toes, mushrooms, and cucumbers (Van UASs have been explored as technol- and payloads is a challenge. Calibration Henten et al. 2002; Reed et al. 2001). ogy tools of great value for agricultural of sensors and data to ensure the quality Apple and cherry harvesting robot arms production. One of the challenges to the of the data collected, and useful metadata can be found in the literature (Davidson greater exploitation of UAS in agricul- generation are some of the other chal- et al. 2015; Burks et al. 2005; Silwal et tural production is the computational lenges. The opportunity to add another al. 2014). Very recently, Atefi and col- resources required to exploit all the data data collection system—an UAS—to an leagues (2019) have shown robotic arms and information they can obtain. The existing suite of data collection systems have also been used for leaf grasping of ability to simply rise above the crop or (e.g., yield monitors and spray control- corn and sorghum plants for phenotyp- livestock and see what cannot be seen lers) creates a challenge in data inte- ing applications. While robotic milking from the ground provides opportunity to gration. UAS allows collection of data stations are commercially available, determine crop and animal health more over the entirety of a field, as opposed robotic manipulators for specialty crop easily. For example, the ability to exam- to a point sample collection with a soil and greenhouse applications are still in ine the ground from the air above enables moisture probe or a monitor on a feed or research and development phase. Robotic detection of plant emergence without water trough. arms used in orchards are challenged by

COUNCIL FOR AGRICULTURAL SCIENCE AND TECHNOLOGY 3 occlusion of fruit (one fruit in front of to solve the issue. Seamless information in yield are commonly achieved through another) and varying lighting conditions exchange happens among ground and advances in genetics, improvement of which adversely effects the time required aerial robots with a cloud-based brain soil health, and optimal management of for picking the fruit (Davidson et al. or decision support. Computationally inputs (i.e., timing, rate and placement 2015). The speed of picking a fruit from intensive operations could be done on of nutrients, crop protectants and wa- a tree is still relatively slow compared to the cloud for minimizing computational ter). Mechanization, the historical trend manual picking. Improved motion plan- loads on the robotic vehicles. In an inte- towards larger and more advanced equip- ning and algorithms, and grated animal and crop production con- ment and consolidation of smaller farms, AI techniques are needed to improve the nected farming system such as the one reduces labor costs through economies of fruit picking efficiency of robotic arms. shown in Figure 2, sensor and process scale. Automation has the potential to not In crop production (both row-crop and data from both thefi eld and animal barn only augment current production prac- specialty), robotic manipulators are typi- is sent to the cloud for further analytics tices through optimization of crop inputs, cally mounted on ground robots, hence and improved decision-making, and these but also serves as a disruptive technol- further discussion on the opportunities decisions are sent back to the robotic ogy that has the potential to change the and challenges of this category of robots applicators for implementation in the farming paradigm by reverting to more for specific applications is presented in field. Robotic manipulators housed in the scale neutral technologies and perhaps the section on crop producton. Opportu- barn could be milking the cows, provid- Farming-as-a-Service (FaaS). nities and challenges of robotic manipula- ing right amount of feed, monitoring the The demand for automation in the tors in animal agriculture is presented in environmental conditions, and measuring rapidly evolving agriculture equipment the section on animal agriculture.. physical attributes of the livestock. industry is moving towards the deploy- Ground and aerial robots could be ment of large autonomous vehicles. With working as a team to identify and solve GROUND AND AERIAL the continual increase in engine power problems in thefi eld, whereas robotic and the size of manned equipment, there manipulators could be milking the cows AGRICULTURAL ROBOTS is concern associated with removing the in the barn in a connected farm setting human operator from thefi eld environ- (see Figure 2). UASs could identify larg- Crop Production ment—safety and liability. Liability of er issues atfi eld scale, whereas, UGVs Row crops large autonomous machinery is a driving could be navigating autonomously to Row crop production practices aim to factor for thinking small. In the event of specific locations in the field for high-res- optimize productivity through maximiz- a malfunction or collision, small UGVs olution or application ing yields and minimizing cost. Gains could cause less damage compared to an UGV with high gross vehicle weight. The concept of replacing conventional large agricultural machines with multiple small-to-medium-sized UGVs is fea- sible given the advances in autonomous technologies (Blackmore et al. 2004; Higuti et al. 2019; Troyer, Pitla, and Nut- ter 2016). UGVs will operate 22 hours per day to compensate for the reduced work rate of smaller equipment, and to minimize environmental impact through precise management of crop inputs. Soil compaction that is associated with the use of larger manned machinery, will be po- tentially alleviated with the use of small autonomous equipment (Shearer and Pitla 2013). Low gross vehicle weights and optimized field travel paths of UGVs will contribute to reduced soil compac- tion. Additionally, in a system where one large piece of equipment is used forfi eld operations and if that equipment is down for repair and maintenance, the complete operation is stopped. However, in the case where multiple UGVs are used, even if one or two UGVs are down for Figure 2. UGVs, UASs, and Robotic Manipulators in a Connected Farm. repair and maintenance, other UGVs can

4 COUNCIL FOR AGRICULTURAL SCIENCE AND TECHNOLOGY still continue to work without halting the entire operation and minimize risk (Anil et al. 2015). UGVs could be operating in large numbers in a swarm configura- tion in the future (Project Xaver: https:// www.fendt.com/int/xaver; Swarm Farm: https://www.swarmfarm.com). In addi- tion to minimizing the risk, one of the other major goals of using these UGVs is to reduce the scale of management to a small land unit or potentially to an individual plant. Concept driverless have been showcased by both established equipment manufacturers (CNH Industrial 2017; Farm Equipment 2019, 2020; Raven Au- tonomy 2020) and start-up companies in last few years. A list of commercial robot prototypes offered by start-up companies can be found here: https://builtin.com/ robotics/farming-agricultural-robots. Figure 3. Flex-Ro, a 60 HP UGV, collecting soybean plot information. (Photo credit: Some companies are offering advanced Santosh Pitla, University of Nebraska-Lincoln.) row-crop robotic solutions in the form of Farming-as-a-Service (FaaS: Sabantaoag. com). Researchers are developing robotic prototypes for row cropfi eld applica- tions such as autonomous weed manage- ment, seeding, and plant phenotyping (Murman 2019; Young, Kayacan, and Peschel 2019; Yuan et al. 2019; Zhang et al. 2016). Multi-purpose robotic plat- forms that could be used throughout the growing season are being explored. This is analogous to a being used for multiple field operations such planting, tillage, and grain haulage. A UGV that could be utilized throughout the growing season for multiple operations could be cost effective, however, there are com- plexities in terms of hardware and soft- ware modularity and system performance that need to be resolved (Fountas et al. 2020). Figure 3 shows a 60 horsepower multi-purpose UGV called “Flex-Ro” 3 phenotyping a soybeanfi eld. For this application, Flex-Ro was equipped with cameras, portable spectrometers, and ultrasonic height sensors to characterize Figure 4. phenotyping a corn plant in a green house. (Photo credit: physical traits of different soybean variet- Abbas Atefi, University of Nebraska-Lincoln.) ies (Murman 2019). Robotic manipulators for corn and sor- ghum plant phenotyping are also under development (Atefi et al. 2019; Bawden ture. In the future, this type of robotic leaf and soil samples for both in-field et al. 2016). Figure 4 shows a phenotyp- arms could be mounted on UGVs to im- and laboratory analysis. However, chal- ing robotic arm that is capable of grasp- prove the overall capabilities of the robot- lenges exist for the deployment of UGVs ing an individual leaf to measure leaf ic platforms. For example, a robotic arm in current row-crop production settings. chlorophyll, leaf temperature, and mois- mounted on an UGV could be collecting The economics of autonomous operation

COUNCIL FOR AGRICULTURAL SCIENCE AND TECHNOLOGY 5 play a big role in the adoption of UGVs. standards, particularly for larger equip- other data layers in geographical informa- Cost comparisons between conventional ment that may need to be transported on tion systems (GIS) or farm management agricultural equipment and UGVs are public roads (ANSI 2020). software (FMS). Data are interpreted and beginning to emerge in literature (Shock- While UGVs are aimed at automating used to prescribe management operations ley, Dillion, and Shearer 2019), however both production activities and sensing and/or predict/explain observed spatial detailed economic and life cycle analyses processes, UASs are so far used primarily variability in productivity. of UGV systems or swarms are required. for sensing applications. The predomi- Research applications expand upon Control architectures for stand-alone and nant use of UASs in row crop produc- image-based remote sensing by incor- fleet level operation of UGVs are still tion has been for general scouting (Stehr porating more expensive sensing and evolving (Posselius et al. 2016). These 2015). Weed, pest, and disease infesta- data analysis methods, including spectral control architectures are important for in- tion resulting in observable changes in sensing (Hamidisepehr and Sama 2019; telligent operation of machinery in a field row crop color or physical structure can Varela et al. 2018) and three-dimensional environment whether they are working by readily be observed with consumer-grade modeling (Dvorak et al. 2020). However, themselves or working in afl eet configu- UASs equipped with standard cameras UASs equipped with research-grade ration. Effective control architectures and (Peña et al. 2015; Vanegas et al. 2018; sensors can frequently cost more than AI algorithms will provide the ability Wang et al. 2020). UASs are frequently a well-equipped 100 HP tractor, which for the UGVs to differentiate between used to quantify and document crop dam- has limited use in production agriculture obstacles, crop, and collaborative entities age for making insurance claims (Luciani given the financial risk associated with (e.g., humans or other robots) to accom- et al. 2020). deployment. UAS technology is advanc- plishfi eld tasks (Jasiński et al. 2016). AI A more advanced, but less common, ing at fast pace and companies are just enabled UGVs will allow the machine to use for UASs is to estimate production beginning to offer UAS that can perform diagnose itself when the operator is not in parameters using image-based remote material application (e.g., spot spraying the loop. For example, when crop residue sensing. Visible, multispectral, or thermal of chemicals: https://www.dji.com/t16). plugs a tillage tool and when a nozzle cameras deployed on multi-rotor or fixed- Numerous challenges prevent wide- plugs on a spray boom, or if a bearing wing UASs are used to collect a series of spread adoption of UASs in row crop goes out on a planter row unit, the UGV images along prescribedfl ight paths at production. Prior to early 2010s, platform has to diagnose and resolve the problem high spatial resolutions (1–10 cm per pix- technology was the critical barrier that on its own (Shearer and Pitla 2013). Key el). Figure 5 shows an example of a UAS limited the use of UASs to research appli- sensing components of these control system collecting multispectral images cations. Significant advances in battery architectures are sensors such as LiDARs of soybean plots. Images are generally technology and autopilot control have and multi-spectral cameras were histori- stitched into orthomosaic maps cover- since made consumer UAS technology cally expensive and required significant ing entire fields that can be overlaid with feasible, although cost still presents an is- computing resources. However, with the proliferation of sensors in the driverless on-road vehicle sector, the cost of sens- ing and computing devices (e.g., GPUs) is projected to decrease in the coming years. AI technologies are advancing at a faster pace; however, the availability of agricultural training datasets is limited. More publicly available training datasets of agricultural environments and open source tools are needed. Internet connec- tivity will become increasingly important to the logistics of servicing this equip- ment in the field (e.g., refilling with seed, nutrients, crop protectants and fuel) as well as coordinating the activity of fleets or swarms of UGVs. Safety standards for the operation of autonomous agricul- tural field machinery are still in nascent stages and require effective public-private partnerships to develop benchmarks and regulations to complement the advance- ments in UGV systems. The standardiza- tion process will likely be informed and Figure 5.A Multirotor UAS collecting multispectral images of soybean plots. influenced by the on-road transportation (Photo Credit: Michael Sama, University of Kentucky.)

6 COUNCIL FOR AGRICULTURAL SCIENCE AND TECHNOLOGY sue. Between 2012 and 2016, researchers duction practices. Regional and national private companies/startups around the and producers commonly referenced con- scale research is needed to demonstrate world have been aggressively pursuing straints posed by Federal regulations on the applicability of a given UAS- development of robotic systems forfi eld UASs operating in the National Airspace deployed technology towards current operations in specialty crops including System were commonly referenced by production practices. Very little data are fruit orchards. researchers and producers as a bottleneck available showing the economic benefit In tree fruit crops and many vegetable for integration into standard production that UASs provide to row crop producers. crops, harvesting is the most critical practices. Regulations have since evolved Coordination between UAS and sensor hurdle as it requires the largest seasonal but are still in their infancy and continue manufacturers and independent research- labor and highest cost among allfi eld op- to restrict usefulness in large-scale row ers will be critical for gaining public’s erations (Silwal et al. 2017; Washington crop production. For example, deploy- confidence in the economics and practi- State Employment Security Department ment of fully autonomous UASs remains cality of this emerging technology. 2016). Because of the criticality of this prohibited, and operating beyond line of field operation, there have been focused site, above the stipulated altitude, or at Orchard and Specialty crops efforts around the world to improve night requires waivers that producers or Agricultural mechanization and auto- harvesting of various fruit and vegetable crop consultants with limited experience mation have made a tremendous positive crops including apples (Hohimer et al. may not be able to readily obtain. impact around the world in reducing farm 2019; Silwal et al. 2017), kiwifruit (Wil- Logistical challenges include a labor requirements, optimizing input liams et al. 2019), strawberries (Xiong et wide range of aspects such as platform utilization, and increasing crop yield al. 2019), and asparagus (Leu et a. 2017). endurance and durability, availability of and quality. However, the adoption of Other efforts in developing robotic solu- ground reference data, and data through- advanced technologies and machines in tions for specialty crops include weeding put. Multi-rotor UASs tend to be easier farming specialty crops including tree and in vegetable crops,fi eld to operate and deploy sensor payloads fruit orchards have been minimal. Many logistics (e.g., strawberry tray moving than fixed-wing UASs but cannot cover field operations such as weeding, tree robots [Hayashi et al. 2010], bin dog [Ye areas as large or as quickly asfi xed-wing training, ,fl ower and fruit thin- et al. 2017]), precision chemical applica- UASs. Both types of platforms cannot ning, and fruit and vegetable harvesting tion based on the canopy needs (Oberti et typically be used when precipitation or are still entirely done manually. As the al. 2016), crop canopy management and high winds are present. When used to farming industry faces continually de- crop load management including pollina- collect remotely sensed data, calibrating creasing availability and increasing cost tion, training, thinning, and pruning. To sensors to parameters of interest remains of farm labor around the world (Gallardo take the full benefit of automation and a challenge to turn-key implementation— and Brady 2015), specialty crop farming robotic technologies in specialty crop partly attributed to varying environmental may not be sustainable if labor saving farming, all labor-intensive operations conditions and the need for adequate technologies are not adopted in the near need to be automated so that there is a ground reference data atfi eld scale. future. uniform, long-term employment in farm- While the recent trend towards cloud- Research and development efforts for ing rather than seasonal peaks of hard- based data aggregation and processing addressing the above challenges have working laborers. has improved data throughput, remote been in place for several decades (e.g., A large amount of federal, state, com- sensing data collected with UASs tends weeding [Slaughter, Giles, and Downey modity group, and venture capital fund- to produce large datafi les that can be dif- 2008], harvesting [Silwal et al. 2017], ing has been invested in recent years for ficult to process and manage. and pruning work [He and Schupp advancing automation and robotic solu- The limited number of UAS manufac- 2018]) with only marginal success for tions for specialty crops, creating a highly turers may also present future challenges commercial adoption to date. Some of conducive environment for research and to increased adoption of UASs. Nearly the challenges to successful adoption development. With these resources and three-quarters of consumer UASs are include lack of desired speed, accuracy, efforts, we are closer than ever in making supplied by a single rotor-craft manufac- and robustness as well as high cost of robotic systems a reality for these crops turer and no major consumer UAS manu- the technology. Recent innovations and as evident by the commercial vegetable facturer exists within the United States, advancement in robotic systems and thinning efforts by Blueriver Technolo- which interestingly is in sharp contrast to technologies such as soft robotic hands, gies (acquired by in 2017, military UASs. Agricultural equipment AI tools such as deep learning, and in- [John Deere 2017]), thefi rst commercial manufacturers have been predictably (and creasingly affordable parallel computing of apples by Abundant Robotics justifiably) slow in incorporating UAS- systems such as graphic processing unit in New Zealand in March 2019 (Benson based products into their offerings as they (GPU) have provided new opportunities 2019), and a full-scale robotic apple transition from selling only conventional to overcome some of these challenges harvester is being evaluated by Fresh equipment to also providing technology and develop practical automation and ro- Fruit Robotics (FFRobotics 2016). All of and data solutions. botic solutions for specialty crop farming. these examples illustrate that automated Economics are the ultimate challenge In recent years, utilizing these advance- systems are beginning to be used on com- of incorporating UASs into standard pro- ments, various research institutions and mercial farms. Figure 6 shows a robotic

COUNCIL FOR AGRICULTURAL SCIENCE AND TECHNOLOGY 7 arm equipped with machine vision and AI area include monitoring and assessment al. in press), and monitoring the progress algorithms picking apples. of crop status (Baluja et al. 2012; Poblete of variousfi eld management practices. UASs use in specialty crops such as et al. 2017), identifying/detecting various Use of UASs in these activities can be orchards and vineyards has been in- issues in thefi eld including new pest referred to as a passive application as the vestigated broadly. Some of the recent infestation (Albetis et al. 2017), opera- platform is used just to collect data from research and development efforts in this tion of irrigation system (Chakraborty et the fields but not interact directly with the crop. Alternatively, active application is where tasks are performed in thefi eld including precision spraying of chemi- cals (Faiçal et al. 2014), deterring birds from fruit crops such as wine and blueberries (Bhusal et al. 2019), and crop pollination (Chechetka et al. 2017; Wood et al. 2013). Figure 7 presents an example of active application with UASs for deterring bird pests. These UAS-based techniques show good promise for spe- cialty crop production as the industry is generally output-driven (meaning higher output with higher inputs, if needed), rather than input driven. Minimization approaches are generally employed in some other crops such as corn and wheat. Some major challenges exist in using ground and aerial robots in orchard envi- ronments including lack of desired speed, accuracy, and robustness in outdoor environment. Variability and complexity of crop canopies, variable and uncertain weather, and variability (in shape, size, color, and so on) and delicacy of produce are some of the major challenges impact- Figure 6. Apple Harvesting Robot. (Photo Credit: Manoj Karkee, Washington State ing the performance of robotic machines. University.) One specific case related to complexity of crop canopies would be the occlusion of objects that are meant to be manipulated (e.g., harvested, pruned, or sprayed). For example, a harvesting robot faces the challenges of harvesting a fruit (e.g., apples, strawberries) being occluded and/ or obstructed by leaves, branches, trellis wires, and trunks. Similarly, pruning in tree fruit crops is challenged by the com- plex orientation and position of branches needing to be removed in a tree canopy, whereas weeding in vegetable crops is challenged by the similarities between crop plants and the weed to be removed. In recent years, there has been con- tinuous improvement and adoption of tree fruit canopies to create structured, narrow crop canopies where all or most of the canopy parts (e.g., fruit and branches) are visible and accessible for robotic operation. However, even with the best Figure 7. UAS used for automated bird pest deterrence in vineyard. (Photo credit: possible architecture of apple trees avail- Manoj Karkee, Washington State University.) able in the industry around the world, it

8 COUNCIL FOR AGRICULTURAL SCIENCE AND TECHNOLOGY has been difficult for the latest prototype acquire just a few machines while larger reduce the physically demanding, risky, robots to access and pick more than 80% can acquire dozens of them or and hazardous tasks and create a better of the fruits. Robots could be designed even hundreds of them as desired tofit working environment for people involved to be more dexterous to access and pick the needs of their operations. For the suc- in farming (e.g., remote operation and a larger percentage of fruits. However, cessful implementation of swarm robot- supervision using augmented reality). farming requires simpler, but effective ics, faster broadband connectivity in the As we move into smart farming of machines compared to other industries to fields would be important. Thesefi elds future, where farming decisions are keep the cost at affordable level and to are generally located in the areas that are made increasingly with the AI-powered be able to operate and supervise them in currently less connected and where mo- systems and implemented in the field by farms by workers with limited technical bile service providers have less incentive robotic systems, integration of ground skills. Additionally, farmers should be to expand because of the remoteness of and aerial robots for decision support and able to fix the machines/robots quickly the location and lack of consumer vol- collaborative field operations would be an in remote field locations as they will be ume. Efforts like Farmbeats (Microsoft important direction. For example, there dealing with highly time sensitive opera- 2019) provide a potential opportunity for have been new research efforts for inte- tions like harvesting asparagus where a improving connectivity on farms. Federal grating sensing andfi eld operations using delay of a day might mean a huge loss initiatives such as the Rural Broadband UASs and ground robots in understand- in yield and/or quality of the produce. Connectivity Act of 2018 (USDA 2018) ing disease pressure and nutrient needs Further advancement and adoption of and a sizable growth in connectivity con- and applying certain chemicals to achieve fruit cultivars and canopy architectures sumption by farming industry to support desired outcomes. Similar technolo- friendlier for robotic operation will play a the future of farming may considerably gies could be used to improve irrigation key role for the wider success of robotic expand future broadband connectivity in management in various specialty crops farming in these crops. Genetics, breed- farming areas. A detailed discussion on including nuts, grapes, and tree fruits. ing, and crop physiological studies, in farm connectivity is presented in the sec- close collaboration with engineering tion on farm connectivity. Animal Agriculture studies, are crucial to make this happen. The socio-economic implications of Multi-purpose robotic machines could adopting new technologies are another Dairy be an important opportunity for future crucial aspect to consider in develop- Robots or automated systems cur- research and development. Modular ing and adopting robotic technologies in rently used in the dairy industry include end-effectors could be designed such that farming. In the public, and primarily in automatic milking systems (AMS), feed- different end-effectors could be installed the communities dependent on the farm- ing systems, teat sprayers, calf feeders, in and taken out of a robotic machine to ing jobs being automated, there can be a hoofbaths, and manure handling among perform multiple field operations over the fear of losing jobs when automation/ro- others. Clearly, the offerings and mar- entire farming season using the same ma- botics is introduced to specific industries/ ket competition of several commercial chine. For example, a robot for tree fruit operations. These concerns exist in row manufacturers driving these technologies orchards could be designed so the same and specialty crop production, and other have only helped to improve the technol- machine could be used year-round for non-farming activities as well. There is ogy and offerings to the market. Thefi rst training trees, pruning branches, thinning a huge shortage of labor in agriculture commercially available milking robots flowers, pollinatingfl owers, thinning and it is expected to continue to worsen were available in the early 1990s. Early fruitlets, applying chemicals precisely, in the years to come (Gallardo and Brady adopters of AMS technology initially and harvesting fruit. Other crops such as 2015). If/when we are able to move to a carried the burden of the steep learn- vegetables (particularly organic) need completely automated or robotic farming ing curve associated with the automated efforts in developing robust solutions in specialty crops, some people would process of milking a cow. Presently, for weeding, thinning, and harvesting. In obviously lose their jobs. At the same advances in milking systems place the case of UASs, it is important that active time, new jobs will be created in manu- milking machine at the center of data operations like deterring birds or surgical facturing, sales, supervising, servicing, capture related to milk production, nutri- application of chemicals be combined and maintenance of these machines, tion, reproduction, and animal health. Ex- with regular crop monitoring operations. which will require different set of skills pected performance of these AMS aims In both row-crop and specialty crop than seasonal labor in farming, but will to resolve current demands (and possible production, system modularity is another provide year-around quality jobs with future ones) that dairy producers and big benefit that can be introduced into better pay. markets impose on the industry. Although farming with automated/robotic systems. Many farm labor tasks are difficult, AMS made milking cows faster and more Farming with small robots that can work requiring heavy repetitive motions that efficient while maintaining optimal milk together in close collaboration (as a pose substantial risks to the wellbeing quality and minimizing disease, there are swarm) has been a new direction (Anil et of farm laborers. The advancements we several well-known challenges in dairy al. 2015). With this model, all sizes and have witnessed over the last decade in production such as increased demands on types of farms would be able to adopt AI, robotics, and other relevant disci- traceability and transparency throughout the technology as smaller farmers could plines suggest the these technologies will the food chain (Lopez Benavides and

COUNCIL FOR AGRICULTURAL SCIENCE AND TECHNOLOGY 9 Paulrud 2018). Despite these challenges, industry usually define animal welfare rails that distribute straw across the pens. there are an estimated 35,000 robotics by good health, hence installations have This type of robotic automation is needed milking systems currently in use globally been designed to maximize animal health mainly in Europe, where providing mate- (Salfer et al. 2019). and productivity. Monitoring of behavior rial for enrichment is obligatory by the Initial attraction to AMS resulted and productivity has proven to be useful animal well-being legislation since 2003 from the lifestyle associated with it (i.e., for the early detection of health problems, (Bracke et al. 2006). While not interact- milkers not bound to strict milking times) which helps to improve milk production ing with the animals directly, autonomous and giving producers the opportunity and minimize animal welfare concerns robots for cleaning stalls prior to animals to focus attention on other farm-related (King et al. 2018). entering the building to ensure biosecu- processes or other income opportunities. The concept of animal welfare of cows rity between groups of animals are under The mindset of relegating all milking housed indoors (e.g., free stall systems) development. routine steps to the machine has in some will continue to be debated, mainly In poultry systems, robots are being cases also translated into neglect of some because of the divergence in understand- investigated in the broiler industry and basic management practices associated ing of what natural behavior and animal the buildings for layer hens housed on with the hygiene of the barn environment welfare means for a domesticated animal. beddedfl oor. Robots are designed to pick and maintenance of the milking machine, For this reason, newer developments in up any eggs laid on thefl oor, monitor which generally result in substandard AMS-type environments and housing environmental conditions at the bird milk quality parameters. Nevertheless, must account for societal concerns and level, turn over the litter, and spray dis- the majority of dairy producers using animal motivations, among others, to infectants. TIBOT Technologies suggests AMS manage these systems effectively, ensure they become common practice and that robots placed on thefl oor encour- maximizing cow productivity without accepted by consumers (Beaver, Ritter, ages hens to use nesting boxes, therefore affecting animal welfare or milk quality. and von Keyserlingk 2019). decreasing the number of eggs laid on Small and large dairy herds have now the the floor (TIBOT 2019). Octopus Poultry capacity to use AMS and manage them Precision Livestock Farming (PLF) Safe is a larger robot that is designed quite proficiently, as the knowledge and Robotics in poultry and swine produc- to turn the litter, sanitize the environ- field technical support has grown consid- tion is currently under development, and ment, disinfect the building, and collect erably over the last few decades. there is currently a fair amount of poten- data on temperature, humidity, carbon The obvious trend in interconnected- tial for automation within these animal dioxide, sound, and light (Poultry World ness of cow information and devices in systems, which is mainly focused on the 2019; Octopus Robotics 2019). Besides the farm environment (known as preci- feeding of animals, and the processing of commercially available robots, there is sion ) will help drive the products (meat and eggs). However, this research documenting the development dairies of the future. Research is ongoing section focuses on the automation and of these types of robots. For example, on devices that monitor gait and locomo- robotics that interact with the animals. Georgia Tech University has designed tion, rumination, and breathing, among In the swine industry, automation and a robot that can autonomously pick-up others, which will overall help dairy robotics the focus has been on animal eggs (Usher et al. 2017). Feed delivery farmers understand the complex dynam- breeding and gestation areas. A group systems are a commonplace in both poul- ics of the milking environment and how it in Germany has successfully tested an try and swine facilities. There are new au- can be improved. As AMS use continues for dairy manure tomations that allow for new production to rise and will likely become the norm in scraping in a loose housing gestating sow systems or improved performance. One many parts of the world, some challenges system (Ebertz, Krommweh, and Büscher example of such a system is by Metabolic still need to be overcome. Because of 2019). While the researchers suggest Robotics in broiler facilities that control the need for cows to be milked volun- some modifications, the robot cleaned the time of the feeding, sounds and light. tarily, there will be more attention paid the area and was able to successfully Feed efficiency is increased by 4% under to managing lameness (King et al. 2017), interact with the animals within the pen. this system (The Poultry Site 2018). the physical conformation and placement Other robotic applications are either not While robotics and automation of the udder and teats, and cow behavior autonomous or operate in the absence of improve the efficiency and the effective- (Hallen-Sandgren and Emanuelsen 2017). animals. A “boar bot” is a widely used ness of labor, these changes are faced Initial concerns about cow behavior when robotic application in swine industry. with some challenges such as initial cost, switching from group to individual milk- This is either a cart or a small vehicle that animal-robot interactions, and changes ing—such as decreased milk yield, urina- can hold a boar or its leash. The boar bot associated with production. Changes tion, defecation, increased vocalization drives around animals for nose-to-nose in the production can be minor, such as (Rushen, de Passillé, and Munksgaard contact during heat checking and breed- using a cleaning robot in between groups 1999)—have most likely been resolved, ing of sows. While this is not an autono- of sows in a farrowing house, or a major as demonstrated by Jacobs and Siegford mous vehicle, it shows the need for robot- rethinking of housing changes associated (2012), but other societal concerns and ics within the swine production systems. with electronic sow feeders. perceptions will still need to be clarified Another type of robot that is commercial- Research is being conducted to de- (Beaver et al. 2019). People in the dairy ly available includes a robot that runs on termine the reaction of the birds to both

10 COUNCIL FOR AGRICULTURAL SCIENCE AND TECHNOLOGY aerial and ground robotics systems in Recent publications suggest that ture is a dynamic area which is advancing the broiler industry (Parajuli et al. 2018; new concepts for automation in swine at a very fast pace. These recent publica- Parajuli et al. 2019). While these studies industry focus on the ability to monitor tions provide a comprehensive list of showed an increased reaction distance to pigs to provide information about their additional robotic technologies under aerial and ground robot systems com- growth and well-being. For example, development (Fountas et al. 2020; Fue pared to humans, these tests compared a several researchers have developed and et al.2020; Martinez-Guanter et al. 2020; novel experience of a mechanical object tested methods for estimating weights Ren et al. 2020; Tsolakis, Bechtsis, and with a habitual experience of the human of pigs (Condotta et. al 2018; Kashiha et Bochtis. 2019.; Vougioukas et al. 2020). presence. Once birds become accustomed al. 2014; Kongsro 2014; Pezzuolo et al. to the mechanical objects, the reaction 2018; Shi, Teng, and Li 2016; Stajnko, ENABLING FACTORS FOR distances are likely to change. Further de- Brus, and Hočevar 2008; Wang et al. velopment of robotic and aerial systems 2008; Wongsriworaphon, Arnonkijpan- THE DEPLOYMENT AND in poultry production may include op- ich, and Pathumnakul 2015). This line ADOPTION OF ROBOTS IN tions for estimating bird weight, identify- of research suggests that estimating ing and reporting non-ambulatory birds, weight from depth or digital cameras is AGRICULTURE and possibly removing dead birds. reasonable. However, the difficulty is Use of robots in pig production is the inability to track individual pigs over MachineV ision and AI difficult because of the curious nature of time. Therefore, currently these systems Machine vision and AI technolo- pigs. However, the need for automation would either need to use an RFID system gies are key enabling technologies for in the swine industry has become increas- to track individual pigs or simply record a the advancement of ground and aerial ingly important with the public’s pres- pen average. robots. When a ground or aerial robot sure to eliminate gestation stalls. Several Another example is use of automation passes through thefi eld, or interacts with European countries and some states have systems to identify feeding behavior of livestock, it uses machine vision and AI banned extended confinement of pregnant individuals within a group pen (Adrion et to identify key attributes of the crop and sows in stalls (Anil et al. 2003). One al. 2018; Brown-Brandl, Jones, and Ei- the livestock, respectively. AI is defined long-term solution includes the use of genberg 2016; Kapun, Adrion, and Gall- as the intelligence demonstrated by ma- electronic sow feeding systems. These mann. 2016; Maselyne et al. 2017). These chines to understand their environments systems use radio frequency identifica- researchers showed that feeding behavior and take actions to maximize its success tion (RFID) tags to identify animals so at the feed trough can be monitored suc- of achieving its goals (Russell and Norvig that the amount of feed can be controlled cessfully and that this information may 2003). One of the pressing needs in plant for individual animal. Electronic sow be useful in determining illness in pigs. based agriculture (row crops, vegetables feeding (ESF) stations have been around Increasing the timely identification of and orchards) is to manage weeds (West- since 1982 (Olsson et al. 1986) and are individual pigs with potential illnesses wood et al. 2018). Machine vision and AI currently manufactured by several com- within a commercial size pen of pigs technologies have the potential to manage panies. The challenge with these systems would help producers pinpoint treatment resistant weeds using a suite of is the complete change in production sys- of sick animals and prevent spread to the management strategies (e.g., mechanical tems. However, with the current legisla- rest of the herd, improving animal well- weeding and targeted spot spraying) as tion and increasing public pressure, these being and ensuring the judicious use of there is a lack of new herbicide models. systems are becoming more mainstream. antibiotics. The current downside of this Autonomous using ground In addition, as ESF systems become more system is the need for a RFID tag. The and aerial robots could be the gateway to common, additional automation is needed cost, labor to install, and labor to remove increased use of robotic systems on the to ensure proper care of the animals. This the tags at the slaughterhouse are three farm (Westwood et al. 2018). includes the need to detect non-pregnant apparent issues. The less apparent issue Early commercial applications of sows, weight or condition score changes, is the decrease in value of the ear. Ears machine vision and AI technologies could and the need to detect illness. Boar sta- are currently marketed as dog toys in the be found (FarmWave https://wAww.farm- tions have been integrated into some of United States, and by placing a hole on wave.io/; BlueRiver: http://www.blueri- the ESF layouts. This allows for detection the ear its value is significantly de- vertechnology.com/) and more companies of open or non-pregnant sows. Scales creased. The development of robotics and are beginning to enter this cutting edge are available in some ESF systems, and automation products for housed livestock, domain of AI based agriculture. A signifi- estimation of condition score from depth specifically pigs and poultry, continues to cant limitation to several machine vision video/images has been reported by Con- grow. This growth is being spurred by the and AI tools is the need for publicly dotta and colleagues (2019). Stavrakakis public pressure for changes to improve available robust image training data sets. and colleagues (2015) and Condotta and animal well-being and decrease use of Some of thefi rst few publicly avail- colleages (2019) suggest that determining antibiotics, and the increasing difficulty able image training datasets are Deep- walking patterns of normal and abnor- in finding labor. Weeds (https://github.com/AlexOlsen/ mal sows could be detected with a depth The field of ground and aerial robots DeepWeeds), Date fruit harvest (https:// video. for crop production and animal agricul- ieee-dataport.org/open-access/date-fruit-

COUNCIL FOR AGRICULTURAL SCIENCE AND TECHNOLOGY 11 dataset-automated-harvesting-and-visual- tion areas are most effective when data limiting, because control files (down- yield-estimation), and Sugar Beets (http:// sets are an agglomeration across a region. load) are relatively small; but, the inputs www.dis.uniroma1.it/~labrococo/fds/ A challenge in these (sharing) cases is needed by cloud computing algorithms syntheticdatasets.html). Having access to convincingly communicating of the value are sometimes very large—particularly more publicly available training image (of sharing) to those who own the data to if those inputs are images or videos. For datasets will lead to advancing the intel- share. example, a typical UAS scoutingfl ight ligence and capabilities of ground and For true mining of insights, the data for 20 minutes over a 100-acrefi eld aerial robots. must have the full backstory or context would generate a folder 4 GB in size, Open source technologies play a key (metadata) associated with it (Evans, containing more than 400 color images role in advancing machine vision and AI Terhorst, and Kang 2017; VanEs and plus video (400 ftfl ight elevation, images models. Google’s TensorFlow (https:// Woodard 2017). At times, the definition at 30 mm resolution, and with 75% over- www.tensorflow.org/) is an excellent of metadata and data gets blurred. For lap; J. Scott, personal communication). example of an open source tool that of- example, the speed of the planter might With only digital subscriber line (DSL) fers some of the most advanced machine be “the data” to be interpreted since the connectivity, it will take 20 to 30 times learning and vision tools. Machine decision and actions are control, path longer to upload the data than it took to learning is a subset of AI which involves planning, or logistics of keeping the generate it. “Real time” is a term that machines learning from experiences and planter boxesfi lled; later, when trying to does need a bit of explanation in agricul- data. Another important open source ascertain impacts of practices on pro- tural contexts. Some data (especially for technology is the Robot Operating Sys- duction, speed of the planter would be control) must be within in milliseconds tem (ROS) which is a software environ- metadata or context (as one of many vari- for safety and operations sake. Some ment aimed at creating applications for ables which might have influenced yield). data (such as bin or hopper status) could robots (https://www.ros.org/). A ROS There is a wide variety of data that could have a few seconds of latency without environment dedicated to developing be collected in cropping systems (Table any significant impact. Some data (soil robotic applications for agriculture is now 1). Those data elements have assorted organic matter, soil temperature) is more available (http://wiki.ros.org/agriculture). resolutions in both space and time, highly for model and planning purposes, so Open source technologies are contribut- varied accuracy levels and different origi- latency of even a day(s) might be accept- ing to the advances in robotics research nation points (several machines or sensor able. Most importantly, the data regarding and enabling increased number of start- systems); this offers a wild frontier of op- control must be local or have extremely up companies to offer robotic solutions in portunity for improvement in the design high reliability and low latency. The as- agriculture. of data collection, storage, transfer, and sorted communication pathways (3G, 4G, access systems as well as improvement 5G, Wi-Fi, LoRa, TV whitespace) can Big Data and Cyber in decisions and actions on actual farms each play a role in data flow from remote (Evans, Terhorst, and Kang 2017). devices, but no single pathway will serve Infrastructure As explained in detail in the next sec- as the universal economical solution. Cyber infrastructure and improved tion, connectivity is a key aspect of cyber The value of functional cyber in- data flow in agriculture offer tremen- infrastructure in automated and non-auto- frastructure which raises the level of dous opportunities that are diverse and mated processes. In agricultural contexts, data-enabled decisions and actions comes extensive (National Academies 2018); it is often the upload speed that is more from a blend of scouting and anomaly yet, there are challenges from both social (sharing) as well as technical (size, capacity, interoperability) perspectives. Table 1. Some typical data elements for cropping systems as well as examples of Many of the issues are similar for opera- value. tions completed by autonomous robots and for operations that include humans Data Element Examples of Potential Value as control agents. Much of the enabling Records of who/what/where/when Critical context for later analytics on of improvements in logistics, efficacy, ef- production and logistics ficiency, and sustainable production stem from using the data itself, sometimes in Imagery (RGB, NDVI, LiDAR, Scouting, anomaly detection, support hyperspectral, thermal) analytics for treatment comparisons or near real time, sometimes in retrospective correlation to other sensor data analytics (Buckmaster et al. 2018). Appli- cations for big, shared, public data (often Machine sensors (location, fuel Improved control of the machine anonymized, but not always) include consumption, battery status, slip, crop functions, improved system logistics, or soil moisture, speed, spray rate, metadata for analytics regarding seed marketing, analytics for prescriptions rate) practices and strategic yearly decisions, model building for predictive and preventative Soil sensors (moisture, temperature, Improved logistics planning, influence maintenance, and even tactical within- chemical/biological attributes) on production activities, context for deeper analytics on other data season decisions. Some of these applica-

12 COUNCIL FOR AGRICULTURAL SCIENCE AND TECHNOLOGY detection, estimation of differences due layers of sharing and security protocols, simple example of such a system involves to treatment(s), estimation of yield, and data models, and file formats. Even when wireless moisture-sensor networks, which improvement of logistics (including data are in a common cloud platform, the include sparsely installed soil-moisture both autonomous and supporting non- file interoperability issues exist. Interop- sensors with wireless-transmission autonomous operations (Evans, Terhorst, erability is improving with published capability. Data can be transmitted by the and Kang 2017). Much of the benefit can application program interfaces (APIs), sensors in either a passive or an active be obtained without sharing data, and that but many of the APIs are non-standard context. In an active system, the sensor reduces risk (Department of Homeland and specific to the software, which is nodes “self-report” the data they have Security 2018); however, some of the unwieldy due to the number of players. collected to a remote computer system. benefits of truly big data analytics and Other new developments in data models In passive systems, the sensor nodes models (whether descriptive based on including geohash indexing and graph da- are queried through the network by the biological, chemical, or physical science tabases will also facilitate data movement remote computer system, and they only fundamentals, or AI) do require data and mining; the collaborative develop- report when queried. sharing to build datasets large enough to ment of APIs and ontologies (relation- Large commercial farms commonly robustly capture effects. ships between knowledge representation) apply water, fertilizer and crop protec- Regardless of whether there is “pub- should yield huge benefits of improved tants across wholefi elds in a non-differ- lic” sharing of data, there is always a compatibility. That collaborative develop- entiated way. If they use some form of need for interoperability. This is currently ment should and will be via open source , they typically use a dire need in both cropping and livestock efforts. Open source development has management zones that are multiple acres systems. For example, sharing of simple been the boon in internet development in size, because data are often spatially information such asfi eld boundaries and adoption (Androutsellis-Thotokis et sparse and application equipment tends can be problematic because of assorted al. 2010). Open source is a paradigm and to be large, making more precise applica- formats, naming conventions, and com- agreement that permits users to modify tion impossible. If growers could obtain patibilities (but it should be seamless to existing code or designs with enhance- data at a higher resolution, square-meter share some information aggregated at this ments or extensions. As it relates to data areas in a field or potentially even about level by others such as the Farm Service movement and compatibility (hence individual plants, then in the future they Agency). Even with work such as that of interoperability), open source drastically might be able to apply inputs in a highly AgGateway (www.aggateway.org) in the reduces the need for duplicative work. targeted manner that is more economi- standardized precision ag data exchange Solutions that work will get used, refined, cally and environmentally beneficial. (SPADE) project, we find that many ag- and propagated. A prime example is the Farm machines are increasingly collect- ricultural firms are left with a data-driven open publishing of APIs. Firms, of any ing image and video data that require dream rather than data-enabled reality. In size, that want their data to work with high transmission bandwidth, and new the case of livestock systems, it is typical other data, will save tremendous amounts methods of data collection like remote that feed characteristics and feeding re- of time by using, studying, modifying, sensing with drones require a large cords be in one system while data regard- and distribution solutions that meet cri- amount of storage, computation, and ing production (milk, gain, etc.), activity teria from multiple perspectives. While transmission bandwidth. Furthermore, loggers, and health (and treatments) are open source software has been critical in cases exist in which post-harvest process- in other systems. Data regarding weather countless disciplines and industries, it has ing facilities could benefit from access to or other external factors (input and prod- been largely overlooked in agriculture to farm-collected data. At a cotton ginning uct pricing) are often public but require date. Fortunately, this is changing and facility, for example, the cost, timing, programming to automate consumption the move toward more open collaboration efficiency, and quality of processing seed into decision tools. will democratize data pipeline elements cotton from farmers can be significantly Despite many diagrams to the con- and result in platforms and data-centric enhanced if the facility has data on the trary, data do not need to reside in a solutions that work together better and incoming crop with sufficient lead time common cloud platform for systems are more widely available. so that it can prepare to optimize onsite to work well—in reality, that common batching of varying seed cotton modules, repository will be rare. The reality is that Farm Connectivity staging and logistics, and charge appro- machine data may be in an equipment Connectivity on farm can be thought priate processing fees. As a case in point, company’s cloud: weather may come of as enabling data generated on vari- a cotton gin in southern Texas recently from a public source, genetic information ous machines and sensors to be shared reported that their processing rate in the may reside in the seed company’s service, among the sensors and machines, a farm 2018–2019 ginning season dropped from and chemical application data may come network, and cloud-based storage and 70 to 23 bales per hour when overly from a service provider. Interoperability analysis systems. Wireless connectivity moist seed cotton modules were brought and substantial compatibility is needed enables this data sharing to occur over into the gin, costing more than $2,500 per so that these layers of information can be significant distances at relatively low hour in revenue (USDA-NASS 2019). merged for insight generation (Buckmas- cost, because it does not requirefi xed Connectivity is thus critical for dissemi- ter et al. 2018). This interoperability has hard-wired infrastructure. A common and nation of farm data.

COUNCIL FOR AGRICULTURAL SCIENCE AND TECHNOLOGY 13 The type of wireless connectivity ticularly remote farm fields do not have These IoT systems will be an integral used depends on the needs of the par- wireless infrastructure, so data transmis- part of the ground / aerial robots that are ticular device in the context of the given sion is limited to small data packets at moving into the agricultural production farm. The needs can be categorized in times when machines are in range of cell sector. These IoT systems will imme- terms of bandwidth, range, and power towers or farm-based networks. While diately require technical assistance and requirements, not to mention cost. While 5G wireless technology is expanding, the typical IT technician will neither be wireless moisture-sensor networks even to rural towns because, in part, of ready nor want to work in a pig barn try- typically have a low-bandwidth require- the USDA’s Rural Broadband Initia- ing to determine why the IoT system is ment, devices that collect images and tive, the remoteness of a vast number of not performing. Hence, a new profession video generate large volumes of data and farm fields suggests that real-time data will emerge; individuals with a strong require high bandwidth if near-term data transmission as well as transmission of propensity toward quantitative systems delivery is needed. For example, Wi-Fi large data files like images collected with thinking will be needed to perform the operates at 2.4 to 5.0 GHz and while drones will be impractical for the foresee- updates needed to IoT related systems bandwidth is large, it only has maximum able future (Wiegman, Pitla, and Shearer that are used in the agricultural environ- range of roughly 100 m (Tzounis et al. 2019). On the other hand, industry is ment. There are examples of equipment 2017) in the best of circumstances, so it developing standards to enable seamless manufacturers (e.g., Case IH, AGCO, is largely not useful in thefi eld. Many de- utilization of farm data across vendors. and John Deere) offering autonomous vices use cellular service, which operates concepts forfi eld operations such as at 3.0 to 5.0 GHz with high bandwidth of Interdisciplinary Agricultural planting and spraying. The activities in 2.0 Mbps to 1.0 Gbps, respectively. The this direction will necessitate a workforce range of cellular service is usually several Workforce inclusive of technicians who are adept at km from the nearest tower, but many re- Various agricultural industries have autonomous and electronic technologies. mote farm fields across the United States installed local area networks (LAN) to Local implement dealerships have hired are well outside the range of cell towers. monitor critical specialized systems. a team of IT IoT professionals. Many Satellite connections can be useful at lo- Equipment manufacturers such as John have apps on their phone and the local cations where cellular service is unavail- Deere, Case-IH, and AGCO would not is expecting them to monitor their able, but the bandwidth is typically much even consider running their facilities equipment. The expectation stems from lower and may not be suitable for rapid without the IoT (— the annual fees charged by companies for transmission of large datafi les. The long- many devices connected together with software and cloud services. range connectivity, known as LoRa, has sensors) technology that allows them Moving forward, technicians to repair range of many miles, but the bandwidth to monitor, analyze and provide reports ground and aerial robots, the systems is low and it is suitable only for periodic of their systems. Technicians assess used to connect these robots to the cloud transmission of small data files. A few equipment repair needs from the local for real time (dynamic) review of the other technologies are available but tend dealerships and then make needed repairs data, software development, and manage to have low bandwidth and low range. based on electronic diagnosis of the is- the remote updates will be needed. These Aside from bandwidth and range is the sues on the farm. Grain companies such technicians will not only need an under- issue of standardization of data protocols as ADM, Bartlett, Zen-Noh, and Cargill standing of the technologies used but so that devices and computing systems also use IoT technology to manage their also the environments in which these from different vendors are compatible. grain handling systems. Ironically, the technologies are placed. The advent and The machinery industry has worked one entity that essentially feeds these popularity of video conferencing systems together to develop the ISO 11783 (ISO- corporations—the U.S. farmer—has used such as Zoom, Webex, GoTo meeting, BUS) standard that enables, for example, LAN and IoT technology at a much lower and Skype systems will foster the adop- a tractor from one manufacturer to com- rate than a typical consumer to facilitate tion of IT personnel who spend more municate with a planter from another food production and food safety on their time working from home rather than manufacturer. However, compatibility farms. Some IoT systems are deployed working entirely from a business office. among data-collection devices and cloud- for irrigation systems, milking systems, As society adopts these business and based systems from different vendors is swine facilities, and more, but these sys- technology practices, the support system under development. Various standard- tems still require some technical expertise for IoT systems will both grow and be- ization efforts, for example AgriRouter to operate. Many of these systems are come a leaned upon and needed support and DataConnect, are underway for this such that the farm manager must travel system. There are additional topics that purpose. to the location to make an adjustment to the Farm IT professionals will need to The current state of wireless connec- the system after an educated guess from address: tivity on the farm is that it is typically the data allowed him / her to identify ▪ Firewalls / security / privacy; inadequate to move large datafi les in the issue(s) before the data was lost or ▪ The increase in data driven decision- a timely manner, and even small data animals / crops were negatively impacted making; files cannot be transferred reliably with to drought, disease or a number of other ▪ The IoT system itself as the “man- consistency. Many rural areas and par- conditions. ager”; and

14 COUNCIL FOR AGRICULTURAL SCIENCE AND TECHNOLOGY ▪ BYODx – the increase in Bring Your during autonomous robotic operation, agreements and policies between technol- Own Digital eXperience; lawyers for tort victims will cast blame ogy providers and farmers. The question then becomes how many in three places—the software developer people will be needed for managing IoT (or manufacturer), the retailer, and the CONCLUSIONS systems on the farm? What knowledge, owner. These parties will blame each skills, and abilities will they need to be other in turn. This tension will make the ▪ While labor challenges are driving the effective? Moreover, how soon will they contract license or lease for agricultural demand for automation and robotics be needed? In the next decade, it is an- robots crucial to apportioning liability, in specialty crop production, optimal ticipated that there will be a 35% deficit since these agreements could exculpate management of production inputs is a of graduates with relevant education and the software developer even if a coding primary motivation for using ground experiential opportunities to fill science, error caused the injury. Software license and aerial robots in row-crop produc- technology, engineering, and mathemat- agreements may also include indemnity tion. ics (STEM) related jobs in agriculture provisions that require the robot owner ▪ Use of UGVs in row-crop produc- and natural resources industry (Goeker (farmer) to defend the robot manufacturer tion has the potential to provide scale et al. 2015). Engineering and technology in the event of a negligence claim while neutral technologies to producers with play key roles in developing scientific under the farmer’s control. Farming-as-a-Service (FaaS) model. In tools of discovery and in developing Security is another legal concern. The a system where multiple small UGVs methods to assist scientists working Department of Homeland Security has could be used in place of one high in agricultural sciences domain (NAE warned of potential problems arising horsepower mannedfi eld equipment, Grand Challenges for Engineering 2008). from coding errors, delayed software adverse effects of soil compaction Providing interdisciplinary experiential patches, and hackers (Department of could be minimized. When multiple learning opportunities to students in agri- Homeland Security 2018). Disgruntled UGVs are used, even if one UGV is cultural majors will be critical to address employees who work for robotic manu- down for repair and maintenance, oth- the demands of the future farm where use facturers are also a threat to security ers in thefl eet can still continue to do of robots could be a new normal.Al- of these devices. An isolated incident the operation thereby minimizing the though projects at the graduate program may not be much cause for alarm, but risk of halting the entire operation. level provide opportunities for interdis- concern multiplies when thousands of ▪ UASs are effective technology tools ciplinary training, current undergraduate devices running the same software are that are used in both row and specialty agricultural science curriculums do not all compromised at once. Designers and crop production for general scouting support intentional interdisciplinary train- manufacturers must make sure robotic and special applications, respectively. ing. Changes are required to traditional technology is built with the latest security High initial capital cost, limited flight agricultural education curriculums to measures, and that timely updates are time, and low payload capacity are prepare workforce who will be working provided to address new risks. limiting their widespread use. The at the intersection of engineering/technol- Finally, robotics manufacturers should UASs industry is advancing at a fast ogy, and agricultural and food sciences not forget data privacy and ownership pace and is starting to offer systems in both academia and industry. Elements concerns. Robotic technology has the that can handle heavy payloads with of robotics and automation, automated potential to generate enormous streams the ability perform material application sensor data collection, cloud comput- of data often linked to the manufac- (e.g., spraying). ing, wireless connectivity and IoT, and turer’s cloud. When polled, farmers have ▪ In animal agriculture, robotic milking data science have to be embedded into expressed concern about what happens to stations are matured and commercially undergraduate education curriculums of their agricultural data privacy with new available, whereas, robotics in poultry agriculture majors. technology (Wall 2018). Similarly, many and swine industry is still evolving. farmers have indicated that they do not In addition to automating tasks in the Liability, Security, and Data understand the licensing agreements and animal barns, robots collect environ- privacy policies they are asked to sign mental and animal data that has the Privacy when purchasing or leasing new ag tech- potential for early disease detection Agricultural robots will bring new li- nologies. Legally, the United States lacks and improved animal welfare. ability concerns to farmers. Who is liable a comprehensive data protection statute ▪ There is a need for further research if a farmer’s robot causes injury to people like the European Union’s General Data in assessing economic benefits of us- or property? Liability and legitimate Protection Regulation (GDPR), however, ing ground and aerial robots in crop safety concerns are why small UAS (<50 the industry has developed a private data production. Interdisciplinary research lb) and UGVs (<500 lbs) will likely be transparency certification (www.agdata- efforts with teams consisting of en- the first autonomous systems widely transparent.com) that awards companies gineers, agricultural economists, and commercialized. Traditional negligence that clearly answer certain questions technology companies are needed to cases require an injured party to prove about storage, handling, sharing and dis- fully understand the economic benefits causation,linking the careless act to the posal of the agriculture data they collect. of using agricultural robots. injury. When injury or damage results There remains a need for clear, concise ▪ Autonomous farming using multi-pur-

COUNCIL FOR AGRICULTURAL SCIENCE AND TECHNOLOGY 15 pose robotic machines presents better TASK FORCE reliance on human labor. economic viability than highly special- ▪ Robust educational programs are ized machines, because then a large RECOMMENDATIONS needed for updating policy makers on capital investment would not have to ▪ Farmers have always been reticent technology expansion in agriculture. It stay idle for a long time. Developing to adopt new technologies void of a is essential for stakeholders, industry multi-purpose robots would, therefore, proven ROI. Low adoption rates and professionals and government agen- be another important aspect in the fu- narrow markets for these products cies to remain abreast of recent devel- ture to improve the viability of ground require technology start-ups to rely on opments in the private sector. and aerial robots. wider margin to remain economically ▪ High-speed network connectivity viable. Researchers and technology LITERATURE CITED across rural America and data compat- companies need to work closely with Adrion, F., A. Kapun, F. Eckert, E.-M. Hol- ibility and inter-operability between the early adopters to quickly evaluate land. M. Staiger, S. Götz, and E. Gall- systems are critical for adopting ROI of automated/robotic to support mann. 2017. Monitoring trough visits of robotic and AI-based technologies to increased adoption of appropriate growing-finishing pigs with UHF-RFID. improve production efficiency. technologies. Comput Electron Agr144. 10.1016/j.com- ▪ Converting data to actionable insights ▪ Rural high-speed broadband connec- pag.2017.11.036. require full context or backstory of tivity is essential for the successful Albetis, J., S. Duthoit, F. Guttler, A. Jacquin, M. Goulard, H. Poilvé, J-B, Féret, G. Dedieu. the data and the collection methods. deployment ground and aerial robots. 2017. Detection ofFlavescence dorée Interoperability of datasets is key for Investments to support public-private grapevine disease using Unmanned Aerial seamless sharing of data among mul- partnerships for developing and Vehicle (UAV) multispectral imagery. tiple stakeholders to generate useful building out innovative wireless IoT Remote Sensing9 (4): 308, https://doi. information for decision making. solutions are needed. Robust connec- org/10.3390/rs9040308. ▪ Experiential learning opportunities in tivity among autonomous systems will American National Standards Institute (ANSI). the areas of ground and aerial robots, simplify exchange of information for 2020. U.S. Transportation Plan Supports IoT, and AI are currently missing efficient machine coordination while Voluntary Consensus Standards for Auton- omous Vehicle Technology, https://www. in traditional agriculture educa- enabling data transfer to Edge and ansi.org/news-and-events/standards-news/ tion curriculums. Curriculums need FOG computing hardware and remote all-news/2020/01/us-transportation-plan- to introduce these interdisciplinary computing infrastructure. supports-voluntary-consensus-standards- experiential learning to train the next- ▪ Safety, operational, and performance for-autonomous-vehicle-technology-16 generation agricultural work force. standards must be developed in Androutsellis-Theotokis, S, Spinellis, D. Ke- This will ensure that the work-force is support of ground and aerial robot chagia, M., and Gousios, G. 2010. Open prepared for the agricultural land- deployment in agricultural production source software: A survey from 10,000 feet. Foundations and Trends in Technology, scape of the future where it might be environments. Standards are neces- Information and Operations Management. required for personnel to be competent sary to address product liability for 4:187–347, doi:10.1561/0200000026. in traditional agricultural sciences both original equipment manufactur- Anil, H., K. S. Nikhil, V. Chaitra, and B.S. along with an understanding of how ers (OEMs) and start-up companies. Gurusharan. 2015. Revolutionizing farm- elements of mechatronics, data sci- Standards also shorten product design ing using .International ence, and AI interacts with crop, soils, cycles while ensuring the safe integra- Journal of Simulation: Systems, Science and animals. tion of ground and aerial robots to and Technology. https://doi.org/10.5013/ IJSSST.a.16.06.04 ▪ Ground and aerial robots also provide production environments. Atefi, A., Y. Ge, S. K. Pitla, and J. Schnable. an opportunity to attract new genera- ▪ Educational institutions, government 2019. In vivo human-like robotic phenotyp- tion into farming, which, otherwise, is agencies and the private sector must ing of leaf traits in maize and sorghum in expected to face a crucial challenge in initiate a dialog with policy makers greenhouse.Computers Electronics Agric. the future as the current generation of regarding data privacy, security, own- 163 (19): 104854. farmers retires. ership, and transparency in agriculture. Baluja, J., M. P. Diago, P. Balda, R. Zorer, F. ▪ Deployment of ground and aerial These efforts will ensure open dia- Meggio, F. Morales, and J. Tardaguila. 2012. Assessment of vineyard water status robots in production agriculture could logue on ethical data use, and return variability by thermal and multispectral result in liability, security, and data of value for data generators and users. imagery using an privacy concerns. Industry is address- Perhaps it is time to consider HIPAA- (UAV).Irrigation Science30 (6): 511–522, ing data privacy concerns through like privacy rules for agriculture? https://doi.org/10.1007/s00271-012-0382-9. programs such as data transparency ▪ The regulatory environment must Bawden, O., J. Kulk, R. Russell, C. McCool, A. certification, however more efforts keep pace with the changing land- English, F. Dayoub, C. Lehnert, T. Perez. are needed to address the liability and scape of agriculture to fully realize 2017. Robot for weed species plant-specific management. J Field Robotics 34:1179– security concerns of ground and aerial the expected benefits stemming from 1199 robots. the deployment of ground and aerial Beaver, A., C. Ritter, and M. A. G. von Key- robotic systems in crop and livestock serlingk. 2019. The dairy housing production environments with reduced dilemma: Natural behavior versus animal

16 COUNCIL FOR AGRICULTURAL SCIENCE AND TECHNOLOGY care.Vet Clin North Am Food Anim Pract Artificial Pollinators.Chem 2 (2): órioa, P.A. Vargas, T. Braun. 2014. The use 35:11–27. 162–164, ttps://doi.org/10.1016/j. of unmanned aerial vehicles and wireless Benson, T. 2019. Robot Apple Pickers Could chempr.2017.01.008. sensor networks for spraying .J Change Farming as We Know It.Daily Chen, G. (ed). 2018. Agricultural Robotics. Syst Architect60 (4): 393–404, https://doi. Beast. https://www.thedailybeast.com/ Advances in and org/10.1016/j.sysarc.2014.01.004 robot-apple-pickers-could-change-farming- TechnologiesCRC Press, Boca Raton, Farm Equipment. 2019. “John Deere Reveals as-we-know-it (3 June 2019). Florida. New ‘Driverless’ Tractor Concept.” https:// Bhusal, S., K. Khanal, S. Goel, M. Karkee, M. CNH Industrial. 2017. “Autonomous tractor www.farm-equipment.com/articles/17489- E. Taylor. 2019. Bird Deterrence in a Vine- technology shows way forward for farm- john-deere-reveals-new-driverless-tractor- yard Using an Unmanned Aerial System ing: enhancing efficiency and working concept (UAS).T ASABE62 (2): 561–569, https:// conditions in agriculture,” https://media. Farm Equipment. 2020. “Kubota Unveils doi.org/10.13031/trans.12923. cnhindustrial.com/EUROPE/CASE-IH/ Prototype of Electric Autonomous Trac- Blackmore, B.S., S. Fountas, S. Vougioukas, L. autonomous-tractor-technology-shows- tor.” https://www.farm-equipment.com/ Tang, C.G. Sorensen, and R. Jorgensen. way-forward-for-farming--enhancing- articles/17854-kubota-unveils-prototype- 2004. A method to define efficiency-and-working-condit/s/d9d11785- of-electric-autonomous-tractor behaviors. pp. 1197–1200. In Proceedings 2881-4577-afc2-23e6dadbfc91. (Accessed FFRobotics. 2016. “FFRobotics The Future of of the Mechatronics & Robotics Confer- 29 July 2020). Fresh Fruit Harvest”. https://www.ffrobot- ence (MECHROB), Kyoto, , 7–8 Condotta, I. C. F. S., T. M. Brown-Brandl, L. ics.com/ (3 June 2019). October. A. Rempel, J. R. Miles, A. J. Cross, G. A. Fountas, S. N. Mylonas, I. Malounas, E. Rodias, Bracke, M. B. M., J. J. Zonderland, P. Lenskens, Rohrer, and K. O. Silva-Miranda. 2019. C. H. Santosh, and E. Pekkeriet. 2020. W. G. P. Schouten, H. Vermeer, H. A.M. Assessing feasibility of using depth images Agricultural Robotics for Field Operations. Spoolder, H. J. M. Hendriks, H. Hopster. to acquire body condition score of sows. Sensors20:2672; doi: 10.3390/s20092672 206. Formalised review of environmental Pp. 308–314. In B. O’Brien, D. Hennessy, Fue, K. G., W. M. Porter, E. M. Barnes, G. C. enrichment for pigs in relation to political and L. Shalloo (eds.),Precision Livestock Rains. (2020). An Extensive Review of decision making. Appl Anim Behav Sci98 Farming 2019 - Papers Presented at the Mobile Agricultural Robotics for Field 3-4: 165–182, https://doi.org/10.1016/j. 9th European Conference on Precision Operations: Focus on Cotton Harvesting. applanim.2005.08.021 Livestock Farming, ECPLF 2019, Cork, AgriEngineering 2020, 2(1), 150-174; Broucek, J. and P. Tongel. 2015. Adaptability of Ireland, 26–29 August. https://doi.org/10.3390/agriengineering Dairy Cows to Robotic Milking: A Review. Condotta, I., T. Brown-Brandl, K. Silva-Mi- 2010010 Slovak J Anim Sci 48:86–95 randa, and J. Stinn. 2018. Evaluation of a Gallardo, R. K., and M. P. Brady. 2015. Adop- Brown-Brandl, T. M., D. D. Jones and R. A. depth sensor for mass estimation of grow- tion of labor-enhancing technologies by Eigenberg, R. A. 2016. Modelling Feeding ing andfi nishing pigs.Biosyst Eng173. specialty crop producers: The case of the Behavior of Swine to Detect Illness. InInt 10.1016/j.biosystemseng.2018.03. Washington apple industry.Agricultural Fi- Conf on Agricultural Engineering, CIGR 002. nance Review75 (4): 514–532, https://doi. – AgEng 2016, Aarhus, Denmark, 26–29 Davidson, J. R., A. Silwal, M. Karkee, J. Li, org/10.1108/AFR-05-2015-0022. June. K. Xia, Q. Zhang, and K. Lewis. 2015. Goeker A, Smith E, Fernandez JM, Ali R, and Bruinsma, J. 2009. The Resource Outlook to Extended Abstract: Human-Machine Theller R. (2015). Employment Oppor- 2050: By How Much Do Land, Water and Collaboration for the Robotic Harvest- tunities for College Graduates in Food, Crop Yields Need to Increase By 2050?. ing of Fresh Market Apples.IEEE ICRA Agriculture, Renewable Natural Resources, Paper presented at the FAO Expert Meet- 2015 Workshop on Robotics in Agricul- and the Environment: United States, ing, Rome, Italy, on “How to Feed the ture.doi:10.1155/2012/368503 10.13140/ 2015-2010. United States Department of World in 2050”, Rome, 24-26 June. RG.2.1.3457.2880. Agriculture. Available online at: https:// Buckmaster, D., J. Krogmeier, A. Ault, S. Noel, Department of Homeland Security. 2018. www.purdue.edu/usda/employment/ Y. Wang, Y. Zhang, A. Layton, and A. Threats to Precision Agriculture, Hallen-Sandgren, C. and U. Emanuelsen. 2017. Balmos. 2018. Use cases for real time data https://www.dhs.gov/sites/default/files/ Is there an ideal automatic milking system in agriculture. InProceedings of the 14th publications/2018%20AEP_Threats_to_ cow and how is she different from an ideal International Conference on Precision Precision_Agriculture.pdf (6 March 2019) parlor-milked cow? Pp. 61–68. InProceed- Agriculture, Montreal, Quebec, Canada, Dvorak, J. S., L. F. Pampolini, J. J. Jackson, H. ings of the 56th Annual National Mastitis 24–27 June 2018, https://www.ispag.org/pr Seyyedhasani, M. P. Sama, and B. Goff. Council Meeting, St. Pete’s Beach, Florida, oceedings/?action=download&item=5394 2020. Predicting Quality and Yield of 28–31 January 2017. Burks, T., F. Villegas, M. Hannan, S. F.lood, Growing Alfalfa from a UAV.T ASABEIn Hamidisepehr, A. and M. P. Sama. 2019. B. Sivaraman, V. Subramanian and J. Press. doi: 10.13031/trans.13769. Moisture Content Classification of Soil and Sikes. 2005. Engineering and Horticultural Ebertz, P., M. S. Krommweh, and W. Büscher. Stalk Residue Samples from Spectral Data Aspects of Robotic Fruit Harvesting: Op- 2019. Feasibility Study: Improving Floor Using Machine Learning.T ASABE62 (1): portunities and Constraints.Hor Technol- Cleanliness by Using a Robot Scraper 1–8. ogy15 (1), 79–87. in Group-Housed Pregnant Sows and Hayashi, S., K. Shigematsu, S. Yamamoto, K. Chakraborty, M., L. R. Khot, and Peters, R. T. Their Reactions on the New Device. Kobayashi, Y. Kohno, J. Kamata, and M. 2019. Assessing suitability of modified Animals (Basel)9 (4): 185, doi: 10.3390/ Kurita. 2010. Evaluation of a strawberry- center pivot irrigation systems in corn ani9040185. harvesting robot in afi eld test. Biosyst Eng production using low altitude aerial imag- Evans, K. J., A. Terhorst, and B. H. Kang. 2017. 105 (2): 160–171, https://doi.org/10.1016/j. ingtechniques Information Processing From data to decisions: Helping crop pro- biosystemseng. 2009.09.011. in Agriculture, https://doi.org/10.1016/j. ducers build their actionable knowledge. He, L. and J. Schupp. 2018. Sensing and inpa.2019.06.001. Crit Rev Plant Sci36 (7): 71–88. Automation in Pruning of Apple Trees: Chechetka, S. A., Y. Yu, M. Tange, and E. Faiçal, B. S., F. G. Costa, G. Pessin, J. Ueyama, A Review. 8:211, https://doi. Miyako. 2017. Materially Engineered H. Freitas, A. Colombo, L. Villas, F. S. Os- org/10.3390/agronomy8100211.

COUNCIL FOR AGRICULTURAL SCIENCE AND TECHNOLOGY 17 Higuti, Vitor A. H., A. E. B. Velasquez, D. Va- 2018. A hitchhiker’s guide to the milk- September. rela Magalhaes, M. Becker, and G. Chow- ing systems of the future. InNMC Milano Parajuli, P., Y. Zhao, T. Tabler, and. G. D. dhary. 2019. Under canopy light detection 2018—International Bovine Mastitis Con- Chesser, Jr. 2019. Evaluating avoidance and ranging-based autonomous navigation. ference, Milano, Italy, 11–13 June. distance of broilers exposed to aerial J Field Robotics, https://doi.org/10.1002/ Luciani, T. C., B. A. Distasio, J. Bungert, M. automated systems. In 2019 ASABE Annual rob.21852. Sumner, and T. L. Bozzo 2020. Use of International Meeting, Boston, Massachu- Hohimer, C. J., H. Wang, S. Bhusal, J. Miller, drones to assist with insurance,fi nancial setts, 7–10 July. C. Mo, and M. Karkee. 2019. Design and underwriting related activities. US Peña, J. M., J. Torres-Sánchez, A. Serrano-Pérez, and Field Evaluation of a Robotic Apple Patent 10,762,571,fi led 2 Sept 2014 and A. I. De Castro, and F. López-Granados. Harvesting System with a 3D-Printed issued 1 Sept 2020. 2015. Quantifying efficacy and limits of Soft-Robotic End-Effector.T ASABE62 Martinez-Guanter, J., P. Aguera, and M. Perz- unmanned aerial vehicle (UAV) technol- (2): 405–414, https://doi.org/10.13031/ Ruiz. 2020. Spray and economics assess- ogy for weed seedling detection as affected trans.12986. ment of a UAV-based ultra-low-volume by sensor resolution. Sensors 15 (3): Jacobs, J. A. and J. M. Siegford. 2012. Lactat- application in olive and citrus orchards. 5609–5626. ing dairy cows adapt quickly to being Precision Agriculture 21:226–243. Pezzuolo, A., M. Guarino, L. Sartori, L.A. milked by an automatic milking system. J Maselyne, J., A. Van Nuffel, P. Briene, J. Vang- González, and F. Marinello On-barn pig Dairy Sci95 (3): 1575–1584, https://doi. eyte, B. De Ketelaere, S. Millet, J. Van den weight estimation based on body measure- org/10.3168/jds.2011-4710. Hof, D. Maes, and W. Saeys. 2017. Online ments by a Kinect v1 depth camera. Com- Jasiński, M. J. Mączak, P. Szulim, S. Radkows- warning systems for individual fattening put Electron Agr148:29–36, 10.1016/j. ki. 2016. Autonomous Agricultural Robot pigs based on their feeding pattern.Biosyst compag.2018.03.003. - Collision Avoidance Methods Overview Eng173:143–156, doi:10.1016/j.biosys- Poblete, T., S. Ortega-Farías, M. A. Moreno, Proc Inst Veh2:106 temseng.2017.08.006 and M. Bardeen. 2017. Artificial neural John Deere. 2017. Deere to Advance Machine Microsoft. 2019. “FarmBeats: AI, Edge & IoT network to predict vine water status spatial Learning Capabilities in Acquisition for Agriculture.” https://www.microsoft. variability using multispectral information of Blue River Technology. https:// com/en-us/research/project/farmbeats-iot- obtained from an unmanned aerial vehicle www.deere.com/en/our-company/news- agriculture/ (UAV).Sensors 17 (11): 2488, https://doi. and-announcements/news-releases/ Murman, J. N. 2019. Flex-Ro: A Robotic High org/10.3390/s17112488. 2017/corporate/2017sep06-blue-river- Throughput Field Phenotyping System, Posselius, J. H., C. H. Foster, S. K. Pitla, S. technology/ (3 June 2019). Master’s thesis, University of Nebraska- A. Shearer, J. D. Luck, M. P. Sama, R. S. Kapun, A., F. Adrion, and E. Gallmann. 2017. Lincoln, Lincoln, Nebraska. Zandonadi, Multi-Robot System Control Behaviour and activity monitoring of National Academies of Sciences, Engineer- Architecture. US Patent 9,527,211,fi led 9 growing-finishing pigs with UHF-RFID. ing, and Medicine. 2018.Science Break- May 2014 and issued: Dec 27th, 2016. In8th European Conference on Precision throughs to Advance Food and Agricultural Poultry World. 2019. ”Robotics revolution hap- Livestock Farming, Nantes, France, 12–14 Research by 2030.The National Academies pening now”, https://www.poultryworld. September. Press, Washington, D.C., https://doi. net/Home/General/2019/5/Robotics-revolu- Kashiha, M., C. Bahr, S. Ott, C. Moons, T. org/10.17226/25059 (18 March 2019). tion-happening-now-425378E/. Niewold, F. Ödberg, and D. Berckmans. National Academy of Engineering Grand Chal- Raven Autonomy. 2020. “DOT®”. https://raven- 2014. Automatic weight estimation of lenges for Engineering. 2008. 14 game- precision.com/raven-autonomy/dot individual pigs using image analysis.Com- changing goals for improving life on the Reed, J. N., S. J. Miles, J. Butler, M. Baldwin, put Electron Agr107:38–44. 10.1016/j. plant. Available at: http://www.engineer- and R. Noble. 2001. Automatic mushroom compag.2014.06.003. ingchallenges.org/challenges.aspx harvester development.Journal Agr Eng King, M. T. M., E.A. Pajor, S. J. LeBlanc, and Oberti, R., M. Marchi, P. Tirelli, Calcante, Iriti, Res78:15–23. T. J. DeVries. 2016. Associations of herd- Hočevar, J. Baur, J. Pfaff, C. Schützm, Rushen, J., A. M. de Passillé, and L. Munks- level housing, management, and lameness and H. Ulbrich. 2016. Selective spraying gaard. 1999. Fear of people by cows and prevalence with productivity and cow of grapevines for disease control using a effects on milk yield, behavior, and heart behavior in herds with automated milking modular agricultural robot.Biosyst Eng rate at milking.J Dairy Sci82:720–727. systems.J Dairy Sci99:9069–9079. 146:203–215, https://doi.org/10.1016/j. Russell, S. J. and P. Norvig. 2003. Artificial King, M. T. M., S. J. LeBlanc, E.A. Pajor, T. C. biosystemseng.2015. 12.004. Intelligence: A Modern Approach (2nd ed.). Wright, and T. J. DeVries. 2018. Behavior Octopus Robotics. 2019. Octopus Poultry Safe. Prentice Hall, Upper Saddle River, New and productivity of cows milked in auto- https://octopusrobots.com/en/octopus- Jersey mated systems before diagnosis of health poultry-safe-en/. Salfer J., M. Endres, W. Lazarus, K. Minegishi, disorders in early lactation.J Dairy Sci Olsson, A.-Ch., M. Andersson, D. Rantzer, B. Berning. 2019. Dairy Robotic Milk- 101:4343–4356. J. Svendsen, and T. Hellström, T., 1986 ing Systems – What are the Economics? Kongsro, J. 2014. Estimation of pig weight us- .Group housing of sows in gestation: DAIReXNET. https://dairy-cattle.exten- ing a Microsoft Kinect prototype imaging Comparison of a computer-controlled indi- sion.org/2019/08/dairy-robotic-milking- system. Comput Electron Agr109:32-35, vidual feeding system with a group feeding systems-what-are-the-economics/ 10.1016/j.compag.2014.08.008 system based on biologicalfi xation (Report Shearer, S. A. and S. K. Pitla. 2013. Field Crop Leu, A., M. Razavi, L. Langstadtler, D. Ristic- 51).Swedish Univ Agric Sci Dept Farm Production Automation. Pp. 97-122. In Q. Durrant, H. Raffel, C. Schenck, A. Graser, Buildings, Lund, Sweden. Zhang and F.J. Pierce (eds.)Agricultural and B. Kuhfuss 2017. Robotic green Parajuli, P., Y. Huang, Y. Zhao, T. Tabler, and Automation: Fundamentals and Practices. asparagus selective harvesting.IEEE/ J.L. Purswell. 2018. Comparative evalu- CRC Press, Taylor & Francis Group: New ASME Transactions on Mechatronics22 ation of poultry avoidance distances to York. (6): 2401–2410, https://doi.org/10.1109/ human vs. robotic vehicle. In10th Inter- Shi, C., J. Zhang, G. Teng. 2016. Mobile TMECH.2017.2735861. national Livestock Environment Sympo- measuring system based on LabVIEW Lopez Benavides, M. G. and C. O. Paulrud. sium (ILES X),Omaha, Nebraska, 25–27 for pig body components estimation in

18 COUNCIL FOR AGRICULTURAL SCIENCE AND TECHNOLOGY a large-scale farm. Comput Electon Agr (USDA). 2018. USDA Partners with Com- ment. (2016). 2016 Agricultural Workforce 156:299–405, https://doi.org/10.1016/j. munities to Bring High-Speed Broadband Report. https://esd.wa.gov/labormarketinfo/ compag.2018.11.042. e-Connectivity Infrastructure to Rural ag-employment-and-wages (3 June 2019). Shockley, J. M., C. R. Dillon, and S. A. Shearer. Areas. https://www.usda.gov/media/ Westwood J. H., R. Charudattan, S. O. Duke, S. 2019. An Economic Feasibility Assessment press-releases/2018/11/13/usda-partners- A. Fennimore, P. Marrone, D. C. Slaughter, of Autonomous Field Machinery in Grain communities-bring-high-speed-broadband- C. Swanton, and R. Zollinger. 2018. Weed Crop Production.Precis Agric 20 (5): e-connectivity Management in 2050: Perspectives on the 1068–1085. United States Department of Agriculture- Future of Weed Science.Weed Sci10.1017/ Silwal, A.,A. Gongal, andM. Karkee, M. 2014. National Agricultural Statistics Service wsc.2017.78 Identification of red apples infi eld environ- (USDA-NASS). 2019. Cotton Ginnings Wiegman, C.R., S. K. Pitla, and S. A. Shearer. ments with over-the- row machine vision 2018 Summary. https://www.nass.usda.gov/ 2019 Advances in Communication systems system.Agric Eng Int: CIGR Journal16 Publications/Todays_Reports/ in Agricultural Robotics. Pp. 49–76. In J. (4): 66–75. reports/ctgnan19.pdf Billingsley (ed.)Robotics and Automation Silwal, A., J. R. Davidson, M. Karkee, C. Mo, Q. Usher, C. T., W. D. Daley, B. P. Joffe, and A. for Improving Agriculture. Burleigh Dodds Zhang, and K. Lewis, K. 2017. Design, in- Muni. 2017. Robotics for Poultry House Science Publishing, Cambridge, United tegration, andfi eld evaluation of a robotic Management. In2017 ASABE Annual In- Kingdom apple harvester.J Field Robot34 (6): 1140- ternational Meeting,Spokane, Washington, Williams, H. A. M., M. H. Jones, M. Nejati, 1159, https://doi.org/10.1002/rob.21715. 16–19 July. M. J. Seabright, J. Bell, N. D. Penhall, J. Slaughter, D. C., D. K. Giles, and D. Downey. Van Henten, E. J., J. Hemming, B. A. J. van Tui- Barnett, M. D. Duke, A. J. Scarfece, H. 2008. Autonomous robotic weed control jl, J. G. Kornet, J. Meuleman, J. Bontsema, S, Ahn, JY Lima, and B. A. MacDonald. systems: A review.Comput Electron Agr E. A. van Os. 2002. An autonomous robot 2019. Robotic kiwifruit harvesting using 61 (1): 63–78, https://doi.org/10.1016/j. for harvesting cucumbers in . machine vision, convolutional neural compag.2007.05.008. Autonomous Robots13:241–58. networks, and robotic arms.Biosyst Eng Stajnko, D., M. Brus, and M. Hočevar.2008. Vanegas, F., D. Bratanov, K. Powell, J. Weiss, 181:140–156, https://doi.org/10.1016/j. Estimation of bull live weight through ther- and F. Gonzalez. 2018. A novel methodol- biosystemseng.2019.03.007. mographically measured body dimensions. ogy for improving plant pest surveillance Wongsriworaphon, A., B. Arnonkijpanich, and S. Comput Electon Agr61 2:233–240. in vineyards and crops using UAV-based Pathumnakul, S., 2015. An approach based Stavrakakis S., W. Li, J. H. Guy, G. Morgan, G. hyperspectral and spatial data.Sensors 18 on digital image analysis to estimate the Ushaw, G. R. Johnson, and S. A. Edwards. (1): 260. live weights of pigs in farm environments. 2015. Validity of the Microsoft Kinect VanEs, H. and J. Woodard. 2017. Innovation in Comput Electron Agric115:26–33. sensor for assessment of normal walking agriculture and food systems in the digital Wood, R., R. Nagpal, and G.-Y. Wei. 2013. patterns in pigs.Comput Electron Agr age. InThe Global Innovation Index 2017: Flight of the Robobees.Sci Am, March 117:1–7. Innovation Feeding the World, 10th ed. 2013, p. 61–65, https://doi.org/10.1038/ Stehr, N. J. 2015. Drones: The newest technol- World Intellectual Property Organization, scientific american0313-60 ogy for precision agriculture.Natural https://www.wipo.int/edocs/pubdocs/en/ Xiong, Y., C. Peng, L. Grimstad, P. J. From, Sciences Education44 (1): 89–91. wipo_pub_gii_2017.pdf (6 March 2019) and V. Isler. 2019. Development andfi eld Tsolakis, N., D. Bechtsis, and D. Bochtis. Varela, S., P. Reddy Dhodda, W. H. Hsu, P. V. evaluation of a strawberry harvesting 2019. AgROS: A Robot Operating System Vara Prasad, Yared Assefa, N. R. Peralta,T. robot with a cable-driven gripper.Comput Based Emulation Tool for Agricultural Griffin, A. Sharda, A. Ferguson, and I. A. Electron Agr157:392–402, https://doi. Robotics. Agronomy 9:403, https://www. Ciampitti. 2018. Early-Season Stand Count org/10.1016/j.compag.2019.01.009. mdpi.com/2073-4395/9/7/403. Determination in Corn via Integration of Ye, Y., Z. Wang, D. Jones, L. He, M. Taylor, G. The Poultry Site. 2018. Robots: The new frontier Imagery from Unmanned Aerial Systems Hollinger, and Q. Zhang. 2017. Bin-Dog: in poultry production. https://thepoultrysite. (UAS) and Supervised Learning Tech- A Robotic Platform for Bin Management com/articles/robots-the-new-frontier-in- niques.Remote Sens10 (2): 343, https:// in Orchards.Robotics 6 (2): 12, https://doi. poultry-production doi.org/10.3390/rs10020343. org/10.3390/robotics6020012 TIBOT Technologies. 2019. Proven results in Wall, F. 2018. Decoding trust in data manage- Young, S. N., E. Kayacan, and J. M. Peschel. Layers farming. https://tibot.fr/spoutnic- ment in Canada. In FCC Vision—Shape 2019. Design andfi eld evaluation of a nav-poule-pondeuse-380.php. the Future of Canadian Agriculture, ground robot for high-throughput phe- Tilman, D., J. Fargione, B. Wolff, C. D’Antonio, https://static.producer.com/wp-content/ notyping of energy sorghum.Precision A. Dobson, R. Howarth, D. Schindler, uploads/2018/12/24102739/2018- Agric20:697–722, https://doi.org/10.1007/ W.H. Schlesinger, D. Simberloff, and D. 11-Farm-Data-Usage_Precision- s11119-018-9601-6. Swackhamer. 2001. Forecasting agricul- Ag_FW_CHARTS-ONLY.pdf#_ga= Yuan, H., N. Wang, R. Bennett, D. Burditt, A. tural driven global environmental change. 2.168146147.313887200.1551376149- Cannon, and K. Chamberlin. 2018. Devel- Science292:281–284 363111074.1551376149 (18 March 2019). opment of a Ground-Based Peanut Canopy Troyer, T. A, S. K. Pitla, E. Nutter. 2016. Inter- Wang, T., J. A. Thomasson, C. Yang, T. Isakeit, Phenotyping System.IFAC-PapersOnLine row using 1D Ranging and R. L. Nichols 2020. Automatic Classi- 51 (17): 162–65. https://doi.org/https://doi. Sensors Inter-row Robot Navigation using fication of Cotton Root Rot Disease Based org/10.1016/j.ifacol.2018.08.081. 1D Ranging Sensors. IFAC-PapersOnLine on UAV Remote Sensing.Remote Sensing Zhang, J., L. Gong, C. Liu, Y. Huang, D. Zhang, 49(16):463–468. 12 (8): 1310. and Z. Yuan. 2016. Field Phenotyp- Tzounis, A., N. Katsoulas, T. Bartzanas, and C. Wang, Y., W. Yang, L T. Walker, amd T. M. ing Robot Design and Validation for the Kittas. 2017. Internet of things in agricul- Rababah. 2008. Enhancing the accuracy Crop Breeding.IFAC-PapersOnLine 49 ture, recent advances and future challenges. of area extraction in machine vision-based (16): 281–86. https://doi.org/https://doi. Biosyst Eng164 (1): 31–48; https://doi. pig weighing through edge detection.Int J org/10.1016/j.ifacol.2016.10.052. org/10.1016/j.biosystemseng.2017.09.007. Agric Biol Eng1 (1): 37–42. United States Department of Agriculture Washington State Employment Security Depart-

COUNCIL FOR AGRICULTURAL SCIENCE AND TECHNOLOGY 19 CAST Member Societies, Companies, Nonprofit Organizations, and Universities

AGRICULTURAL AND APPLIED ECONOMICS ASSOCIATION ■ AMERICAN ASSOCIATION OF AVIAN PATHOLOGISTS ■ AMERICAN ASSOCIATION OF BOVINE PRACTITIONERS ■ AMERICAN BAR ASSOCIATION, SECTION OF ENVIRONMENT, ENERGY, & RESOURCES ■ AMERICAN DAIRYSCIENCE ASSOCIATION ■ AMERICAN FARM BUREAU FEDERATION AMERICAN MEAT SCIENCE ASSOCIATION ■ AMERICAN METEOROLOGICAL SOCIETY, COMMITTEE ON AGRICULTURAL AND FOREST METEOROLOGY ■ AMERICAN SEED TRADE ASSOCIATION■ AMERICAN SOCIETY OF AGRICULTURAL AND BIOLOGICAL ENGINEERS ■ AMERICAN SOCIETY OF AGRONOMY■ AMERICAN SOCIETY OF ANIMAL SCIENCE ■ AMERICAN SOCIETY OF PLANT BIOLOGISTS■ AMERICAN VETERINARY MEDICAL ASSOCIATION■ AQUATIC PLANT MANAGEMENT SOCIETY ■ BASF CORPORATION ■ BAYER CROP SCIENCE ■ CAL POLY STATE UNIVERSITY ■ CORNELL UNIVERSITY ■ CORTEVA AGRISCIENCE ■ CROP SCIENCE SOCIETY OF AMERICA ■ CROPLIFE AMERICA ■ ENTOMOLOGICAL SOCIETY OF AMERICA■ INNOVATION CENTER FOR U.S. DAIRY ■ IOWA STATE UNIVERSITY ■ KANSAS STATE UNIVERSITY■ MISSISSIPPI STATE UNIVERSITY ■ NATIONAL CATTLEMEN’S BEEF ASSOCIATION■ NATIONAL CORN GROWERS ASSOCIATION/IOWA CORN PROMOTION BOARD ■ NATIONAL MILK PRODUCERS FEDERATION■ NATIONAL PORK BOARD ■ NORTH CAROLINA A&T STATE UNIVERSITY■ NORTH CAROLINA BIOTECHNOLOGY CENTER ■ NORTH CAROLINA STATE UNIVERSITY ■ NORTH CENTRAL WEED SCIENCE SOCIETY ■ NORTHEASTERN WEED SCIENCE SOCIETY ■ PENN STATE UNIVERSITY ■ POULTRY SCIENCE ASSOCIATION ■ RURAL SOCIOLOGICAL SOCIETY■ SOCIETY FOR IN VITRO BIOLOGY■ SOIL SCIENCE SOCIETY OF AMERICA■ SYNGENTA CROP PROTECTION■ TEXAS A&M UNIVERSITY■ THE FERTILIZER INSTITUTE■ THE OHIO STATE UNIVERSITY ■ TUSKEGEE UNIVERSITY ■ UNITED SOYBEAN BOARD ■ UNIVERSITY OF ARKANSAS ■ UNIVERSITY OF CALIFORNIA–DAVIS■ UNIVERSITY OF FLORIDA■ UNIVERSITY OF KENTUCKY ■ UNIVERSITY OF MISSOURI■ UNIVERSITY OF NEBRASKA ■ UNIVERSITY OF NEVADA–RENO ■ U.S. POULTRY AND EGG ASSOCIATION■ WEED SCIENCE SOCIETY OF AMERICA■ WESTERN SOCIETY OF WEED SCIENCE The mission of the Council for Agricultural Science and Technology (CAST):CAST, through its network of experts, assembles, interprets, and communicates credible, balanced, science-based information to policymakers, the media, the private sector, and the public. The vision of CAST is a world where decision making related to agriculture and natural resources is based on credible information developed through reason, science, and consensus build- ing. CAST is a nonprofit organization composed of scientific societies and many individual, student, company, nonprofit, and associate society members. CAST's Board is composed of representatives of the scientific societies, commercial companies, nonprofit or trade organizations, and a Board of Directors. CAST was established in 1972 as a result of a meeting sponsored in 1970 by the National Academy of Sciences, National Research Council. ISSN 1070-0021

Additional copies of this Issue Paper are available from CAST, http://www.cast-science.org.

Citation:Council for Agricultural Science and Technology (CAST). 2020.Ground and Aerial Robots for Agricultural Production: Opportunities and Challenges. Issue Paper 70. CAST, Ames, Iowa.

4420W est LincolnW ay Ames, Iowa 50014-3447, USA (515) 292-2125 E-mail: [email protected] Web: www.cast-science.org

20 COUNCIL FOR AGRICULTURAL SCIENCE AND TECHNOLOGY