Advancing the Scientific Frontier with Increasingly Autonomous Systems

Rashied Amini1,a, Abigail Azari2, Shyam Bhaskaran1, Patricia Beauchamp1, Julie Castillo-Rogez1, Rebecca Castano1, Seung Chung1, John Day1, Richard Doyle1, Martin Feather1, Lorraine Fesq1, Jeremy Frank3, P. Michael Furlong3, Michel Ingham1, Brian Kennedy1, Ksenia Kolcio4, Issa Nesnas1, Robert Rasmussen1, Glenn Reeves1, Cristina Sorice1, Bethany Theiling5, Jay Wyatt1

Endorsements Tibor Balint1, Luther Beegle1, Steve Chien1, Chad Edwards1, Carolyn Ernst6, Terrance Fong3, Abigail Fraeman1, Anthony Freeman1, Carly Howett7, Robert Lillis2, Gentry Lee1, Michael Mischna1, Marc Rayman1, Mark Robinson8

September 15, 2020

1NASA Jet Propulsion Laboratory 2Space Sciences Laboratory, University of California, Berkeley 3NASA 4Okean Solutions 5NASA Goddard Space Flight Center arXiv:2009.07363v1 [astro-ph.IM] 15 Sep 2020 6Johns Hopkins University Applied Physics Laboratory 7Southwest Research Institute 8Arizona State University

a rashied.amini@jpl..gov, (626) 720-9942 c 2020 California Institute of Technology. This work was carried out at the Jet Propulsion Laboratory, California Institute of Technology, under a contract with the National Aeronautics and Space Administration (80NM0018D0004). 1 Executive Summary ence and Decadal Survey (PSADS) and the 2018 Workshop on Autonomy for Future A close partnership between people and partially NASA Science Missions will not be achievable us- autonomous machines has enabled decades of ing the current paradigm of spacecraft control [3, space exploration. But to further expand our 4]. horizons, our systems must become more capa- To answer ever more challenging science ques- ble. Increasing the nature and degree of auton- tions, we will need spacecraft that can explore omy - allowing our systems to make and act on unknown and dynamic environments with less in- their own decisions as directed by mission teams - put from human operators. This will demand an enables new science capabilities and enhances sci- integrated approach to autonomy, because auton- ence return. The 2011 Decadal omy is a system-level technology that requires Survey (PSDS) and on-going pre-Decadal mis- an interdisciplinary approach to technology de- sion studies have identified increased autonomy velopment. The need for a revolutionary ad- as a core technology required for future missions. vance in spacecraft autonomy to meet the needs However, even as scientific discovery has neces- of NASA’s missions was identified as early as the sitated the development of autonomous systems 1980 Carl Sagan, et al. report to NASA on Ma- and past flight demonstrations have been success- chine Intelligence and Robotics [5]. Yet despite ful, institutional barriers have limited its matura- the need for and past uses of autonomous capa- tion and infusion on existing planetary missions. bilities, these capabilities are rarely used or even Consequently, the authors and endorsers of this considered in most planetary missions due to in- paper recommend that new programmatic path- stitutional barriers. ways be developed to infuse autonomy, infrastruc- In the Sagan Report, one of the primary barri- ture for support autonomous systems be invested ers to progress in autonomous systems was iden- in, new practices be adopted, and the cost-saving tified as culture: “Technology decisions are, to value of autonomy for operations be studied. much too great a degree, dictated by specific mis- 2 Motivations sion goals, powerfully impeding NASA utilization of modern computer science and technology. Un- From the beginning of interplanetary exploration, like its pioneering work in other areas of science reliance on on-board decision-making has been and technology, NASA’s use of computer science critical for mission success. The use of time- and machine intelligence has been conservative driven command sequences and critical sequence and unimaginative.” Similar cultural issues were retries on the Viking and Mariner 6 & 7 orbiters identified at ESA, based on the experience of 2019 resulted in reduced risk and increased the abil- OP-SAT mission: “...resistance to experimenta- ity of spacecraft to return science data [1,2]. tion and innovation, especially when the projected Over the following decades, the evolutionary in- benefits are not yet flight proven. Time after clusion of autonomous functions, primarily in the time projects settle for reuse rather than inno- domains of spacecraft fault protection, and guid- vation.” [6] To wit, the highest level recommen- ance and control, further reduced mission risk and dation of this paper is one made by the Sagan increased a spacecraft’s ability to perform and Report: “The advances and developments in ma- downlink science measurements. For instance, chine intelligence and robotics needed to make fu- the Jovian radiation environment caused multiple ture space missions economical and feasible will safe mode events during the mission; but not happen without a major long-term commit- “smart safing” enabled it to maintain thermally- ment and centralized, coordinated support.” safe attitude to protect its instruments despite immediate loss of operator control. However, the 3 The Need for Autonomy mission and system complexity needed to answer The 2015 NASA Technology Roadmap defines au- new questions in planetary science has outpaced tonomy as “the ability of a system to achieve efforts to mature autonomous capabilities. As de- goals while operating independently of external tailed in this paper, many of the ambitious mis- control.” [7] External control is exemplified by sion concepts described in the Planetary Mission the traditional commanding control loop, sum- Concept Studies (PMCS) for the Planetary Sci- marized in Figure1. Autonomy reduces the char-

1 Figure 1: The traditional commanding control loop has a characteristic timescale that may be greater than Figure 2: Autonomy is necessary when a spacecraft the timescales needed to exercise control. In these has to react to changes in the environment or within cases, autonomous functions are required. itself at a shorter timescale than afforded by communi- acteristic control timescale by moving analysis cation constraints. Past missions relied on their abil- and planning on board the spacecraft [8]. Mov- ity to predict but future missions that operate in in- creasingly unknown environments would fall outside ing functions traditionally performed by mission the controllability threshold. teams to on board the spacecraft requires changes to both flight systems and ground systems. a priori planning cannot be assured and au- When the timescale associated with traditional tonomous functions are required for safe control. commanding becomes larger than timescales re- Figure2 illustrates the control regimes and as- quired for responding to spacecraft critical events sociated risks of using a priori planning. It’s or interacting in an unpredictable environment, for this reason that autonomous functions were using a priori planning for control increases mis- first used for mission aspects with short dynam- sion risk. Deploying autonomy, in situ percep- ical timescales, e.g. guidance, control, and fault tion, reasoning, and acting under both nominal protection [2]. With incremental confidence from and off-nominal situations, mitigates this risk and these early missions, more ambitious autonomous allows the spacecraft to make decisions based on functions were flown to enable more ambitious current circumstances. science measurements. For example, in 2004 Commanding timescales are partly driven by ’s Impactor spacecraft required au- communications for commands and status. For tonomous navigation functions (AutoNav) to in- instance, in the case of planetary rotorcraft like tercept Tempel 1 [10]. This was driven by Ingenuity or Dragonfly, communication is con- two primary goals: first, colliding with the comet strained by limited data rates (e.g. due to using the Impactor spacecraft, and second, ob- weak signal), restricted link availability (e.g. a serving the impact on the Flyby vehicle. Because body’s orbit and day/night cycle), and large the exact size, shape, and orbit of the comet could data products required for ground-based plan- not be determined in advance from the ground, ning (e.g. contextual maps for path planning) closing the navigational loop on-board was nec- [9]. The timescale of systems responding to essary to successfully impact the 7 km size comet critical events or performing critical operations at a speed of 10 km/s. The use of AutoNav was is driven by the dynamical nature of system- unavoidable to meet the science objective and environment and system-system interactions and AutoNav’s decision-making could be sufficiently how predictable and observable these interactions modeled and supervised that it could be trusted. are. Dynamic rotorcraft mobility in planetary at- While using autonomy may be enabling for mospheres is incompatible with communication some missions, it can enhance mission produc- constraints and has to occur in situ. tivity and science return for nearly all missions, It is this relationship between communications- increasing a mission’s science-to-dollar ratio. On- and dynamical-related timescales that has tra- board science data analysis has been used on var- ditionally driven requirements for autonomy. ious Mars rovers [11, 12] and on Observing- Where the communications timescale exceeds the 1 [13] to improve science return and dynamically dynamical timescale, safe control using strict target and image novel signatures. For instance,

2 incorporated Autonomous Exploration known surface for sampling, and “smart” target- for Gathering Increased Science (AEGIS) to tar- ing to increase science return and reduce risk un- get the ChemCam laser-induced breakdown spec- der off-nominal conditions [19]. On-board anal- troscopy (LIBS) instrument, targeted preferred ysis of “remote”, low-cost data, like Raman or outcrop terrain over 93% of the time as opposed LIBS, can significantly reduce risk of performing to blind targeting, which targeted outcrops 24% high-cost sampling, like drilling [20, 21, 22]. of the time [14]. Two additional white papers Just as critical to these time-limited missions submitted to the PSADS, Azari et al and Theil- is a capability for restorative fault management ing et al, further describe the advantages of using that can restore functionality after a safing event, science autonomy [15, 16]. or take action to avoid the need for a saf- However, a lesson learned from Mars rovers, ing event. As more decision-making capabil- EO-1, Deep Impact, and is that flight- ity is moved on-board, the scope of autonomous proven and high value-to-risk-ratio autonomous fault detection, isolation and recovery (FDIR) functions are not regularly used on Flagship or functions will increase, relying on on-board re- competed missions. Moreover, these deployments planning/scheduling and execution of real-time have not led to the development of system-level contingency actions. This sort of functionality is autonomy, which is able to integrate numerous broadly applicable to science missions operating autonomous and traditional functions. Given in all contexts but critical to missions with signif- NASA’s track record of developing and de- icant and challenging operations outside available ploying autonomous systems and the inter- communication windows. (E.g. Mercury action of autonomous functions necessary nighttime operations with six weeks of no ground for strategic planetary science missions, it contact. Also, Intrepid operations that have to is clear a shift in institutional culture is cover an 1,800 km distance with hundreds of in- necessary. strument placements in four years, where manual interventions cannot occur more than once every 3.1 Future Planetary Missions and Their 6–16 km of traverse, or missions seeking to access Autonomy Requirements extreme surface or subsurface terrains. [23, 24, All of the missions under evaluation by the 25, 26]) PSADS are either enabled by, or would be en- Generally, by adopting an approach where hanced by, the use of increasingly autonomous situational awareness (perception, mapping, es- systems. By evaluating aspects of mission ar- timation, see [27]), hazard assessment, plan- chitecture that constrain communications or are ning/execution, payload data analysis, science driven by the dynamics of system-system and planning, and FDIR functions are moved on- system-environment interactions we can identify board, landed missions will be more produc- how these missions are enabled and enhanced. tive. As an example, results from the Self-Reliant These evaluations are summarized in Table1, and Rover study, wherein a terrestrial rover was oper- additional description for two mission types are ated by campaign intent rather sequenced activi- included below. ties, showed an 80% reduction in sols required to Landed Missions (landers, hoppers, rovers, complete a campaign and 267% increase in loca- aerial systems) Landed systems are impacted by tions surveyed per week [28]. similar issues that drive the use of autonomy. Deep Space Missions Although deep-space pro- Across these classes of missions, the dynamic as- vides a more predictable environment than that pects of operations during entry, descent, and near, on, or into planetary bodies (rendezvous, landing (EDL), roving, flying, and hopping re- proximity operations, surface mobility, below- quire situational awareness and in situ reason- surface access), autonomy enables operations ing and acting. Limited line of sight to Earth with reduced communication burden and allows and communication constraints restrict contact scaling to multi-craft missions for deep-space (or- opportunities and data rates. Limited lifetime biters and flybys). Using autonomous naviga- and data volume constraints, such as a tion functions provides tighter turnaround loops or a Venus lander, will need situational aware- in high dynamic environment situations where ness, assessment of the interaction with the un- long light-times precludes ground processing to

3 Table 1: Summary of how different aspects related to mission architecture can be enabled or enhanced through the use of integrated autonomous functions. Enabling/Enhancing Aspect of Mission Architecture Drivers Autonomous Functions Entry, Descent, and Landing, e.g. - Terrain-relative navigation (TRN) - Mercury Lander, Enceladus Orbilander - Short dynamic timescales and dynamic control Surface & Aerial Mobility, e.g. - Limited a priori atmosphere/surface - Hazard assessment and avoidance - Intrepid Rover characterization - Payload data analysis - Lunar/Vesta Geochronology Hopper - One-way light times (OWLTs) - Planning/execution - Dragonfly, Mars Helicopter - Restorative fault management Short-Lived Landers, e.g. - Science competitiveness - Payload data analysis - Venus Flagship lander - Limited lifetime - Planning/execution - - Limited bandwidth and contact opportunity - Restorative fault management - VIPER Rover Missions with Opportunistic Science - Autonomous navigation (enabling - Limited communications and time to Potentially all missions, but particularly: for fast flybys) perform critical targeting - Fast flyby missions, e.g. [17] - Planning/execution - Science competitiveness and impact - Monitoring missions, e.g. MOSAIC - Science planning - Cost/risk reduction - In-situ missions, e.g. Ceres Habitability - Payload data analysis - Autonomous navigation Interplanetary Cruise, e.g. - Mitigating impact of safing on trajectory - Planning/execution - All missions beyond Earth- system (EP missions) - Science planning - Missions using electric propulsion, e.g. Persephone - Cost/risk reduction - Payload data analysis Missions with Coordinated Observations, e.g. - Multi-SC mapping missions, like MOSAIC - Limited time to coordinate with ground-in- - Multi-agent coordination - Bistatic radar experiments, e.g. CONCERT the-loop - Planning/execution - Planetary defense & impactor/observer missions, e.g. - Science competitiveness - Science planning DART, Small Bodies DRM [18] - Cost/risk reduction - Autonomous navigation - Landed system coordination, e.g. M2020/Mars Helicopter, lava tube exploration Mapping Missions, e.g. - Mitigating the impact from anomaly - Planning/execution - - Science competitiveness - Science planning - MORIE - Cost/risk reduction Operations in High-Radiation Environments, e.g. - Communications delay to restore science - Europa Clipper operations may not be acceptable - Restorative fault management - Volatiles Explorer - Reduce risk/cost - Autonomous navigation Approach/Rendezvous with Unexplored Bodies, e.g. - Uncertainty in relative spacecraft/body position - Autonomous mapping - NEO/NEA missions [18] - Unknown irregular body shape and gravity - TRN, hazard assessment, landing - Comet Sample Return - Unknown geotechnical properties for landing - Autonomous surface navigation - Transneptunian bodies/KBO etc. - Limited a priori surface characterization - Restorative fault management achieve required accuracy. Autonomy can en- In architectures with multiple spacecraft, e.g. a able operations in less predictable scenarios, like two spacecraft flyby of a NEO for bistatic radar atmospheric aerocapture at icy giants, where investigation, multi-agent coordination, naviga- turnaround time on ground-based navigation may tion, and planning/execution can be used to per- induce additional risk, and for planetary constel- form coordinated measurements to achieve chal- lation where coordinated, multi-spacecraft oper- lenging measurement objectives and maximize ations are required [29, 30]. payload utilization based on available resources. As the sensitivity of ground-based surveys im- The Mercury, Venus, Ceres, MORIE, MO- prove over the next decade, the detection fre- SAIC, and Persephone PMCS reports have base- quency of interstellar visitors, e.g., ‘Oumuamua lined electric propulsion [24, 32, 33, 34, 35]. Elec- and 2I/Borisov, and long-period , like tric propulsion trajectories are non-Keplerian and C/2017 K2, is expected to increase [31]. At the use continuous thrusting to benefit from their same time, numerous proposals have called for high ISP . However, safe mode events can result flybys of distant objects such as Trojans, Jovian in missed thrust, risking mission success in terms and Saturnian , trans-Neptunian objects, of schedule and excessive propellant use to cor- and Kuiper Belt objects. Both sets of missions rect trajectory. For , a four-day period of face similar challenges: flybys of these bodies in- missed thrust resulted in a 26-day delay to the volve high relative velocities, limiting the effec- first planned Ceres orbit; a projected seven-day tiveness of ground-based navigation and science missed thrust could have resulted in a ∼50-day planning to the point where the mission may not delay [36]. Restorative fault management capa- be feasible. Integrating autonomous navigation, bility would mitigate the impact of missed thrust. payload data processing, and planning/execution Missions in the Jovian environment, like Eu- functions may be enabling for these missions [17]. ropa Clipper, at Mercury, like Mercury Lan-

4 der, and missions during solar maxima face in- grating different autonomous functions can afford creased risk of radiation-induced events that can PIs more flexibility in meeting science require- induce component resets and induce safe mode ments, relaxing system requirements on mission events. Suspending operations as operators in- systems, and potentially reducing cost and sci- tervene may be acceptable during cruise, but dur- ence risk. For instance, using autonomy like au- ing critical mission phases like a Europa flyby, it tonomous targeting can enable performing novel could cause a mission to fail its mission require- and opportunistic measurements, raising a mis- ments. Restorative fault management could en- sion’s threshold science return. Use of auton- able a spacecraft to recover to an operable state omy like autonomous retry with on-board plan- sufficient to perform critical measurements neces- ning/execution, can reduce mission risk by allow- sary for meeting mission requirements. ing a spacecraft to dynamically adopt to on-board Like EO-1, orbital mapping missions would anomalies and reattempt measurements. benefit from autonomous planning/execution, be- A notable example of a mission that could be- ing able to opportunistically detect novel signa- come more competitive should systems autonomy tures, e.g. methane plumes at Mars or volcanic be used is a New Frontier Venus lander - a con- plumes at Io, and to perform follow-up obser- cept that has seen multiple failed proposals over vations. Missed coverage imaging, e.g. due to two decades, including the step-two Venus In situ an anomaly, could also be replanned on board. Composition Investigations (VICI) and the Venus Multi-spacecraft monitoring, like MOSAIC or the Surface and Atmosphere Geochemical Explorer NanoSWARM concept, could use multi-agent co- (SAGE) proposals [39, 40]. By integrating the ordination for event tracking and map complete- autonomous functions in Table 1, a Venus lan- ness [37]. In these cases, autonomy could improve der could more effectively utilize its time before a mission’s science-to-dollar ratio by increasing end-of-mission. Contextual images and remote productivity and the value of returned data. Raman/LIBS measurements can direct sampling, 3.2 Institutional Drivers of Autonomy resulting in lower risk on science return. Based on current development of the NASA ARC Volatiles There are also institutional motivations for us- Investigation Polar Exploration Rover (VIPER), ing autonomous functions on-board spacecraft. this could enable higher performance and better A shift in culture that is more accepting of au- handling mission-ending faults. tonomous systems would see its benefits in relax- Cost Autonomy could reduce costs to NASA ing constraints on deep space communication and by moving some of operations team functions on scheduling, mission competitiveness, and achiev- board. In the last 10 years, SMD has spent ing more with the planetary exploration budget. $2.4B FY20 on mission operations [41]. If these Deep & Communications In- missions had been operated 20% more efficiently, frastructure All missions utilizing the Deep Space $480M could have been available to NASA. As Network (DSN) could be expanded and see im- predicted in the 1980 Sagan Report, moving to provements in science return using adaptive op- “Autonomous Mission Control” by the year 2000 erations. The introduction of deep space Small- would result in mission operations costs 1% of Sat missions is affecting the roadmap for NASA’s those in 1975. While “Autonomous Mission Con- Deep Space Network (DSN) and is leading to trol” has yet to be realized, the experience from a fundamental change in the way future deep past missions and current research supports that space missions will interact with ground systems prediction. On EO-1, significant mission time was [38]. With the projected increase in the num- saved by having the spacecraft discard imagery ber of DSN users, e.g. SmallSats funded through obscured by cloud cover. SIMPLEx, the DSN will need to more efficiently schedule tracking passes. Increasing the degree of 4 Why Autonomy Isn’t Standardized spacecraft autonomy will allow improvements in Despite decades of incremental progress in achiev- the efficiency of DSN use. ing remarkable successes with the autonomous Mission Competitiveness The interrelated fac- functions used on EO-1, different Mars rovers, tors of cost, perceived risk, and science return and missions like Deep Impact and Stardust, the impact selection of competed missions. Inte- broad use of autonomous functions to achieve

5 mission objectives is still not standard practice. advances. The nature of the competitive mission While the reasons are interconnected, they can process drives engineers at NASA and in industry be generalized as three barriers to infusion. toward heritage solutions in mission proposals to Barrier 1: Unlike most other NASA avoid perceived risk. Integrated over time, this technology investments, autonomy is steers missions away from innovative advances system-level technology. Most technologies, that could be enabling for more demanding mis- e.g. detectors and propulsion systems, fulfill sions. There is little interdisciplinary coopera- specific needs. Their operational conditions tion to offer guidance on performing software- and can be readily modeled for laboratory testing autonomy-related trade studies, which could have and interfaces and behavior well-defined for significant implications for mission architecture. mission development and operations. For these This lack of exposure and strained mapping of technologies, the NASA definitions of Technology TRL to multi-mission autonomous functions fur- Readiness Level (TRL) are a relevant and useful ther reinforces a culture wary of adopting more tool for appraising flight-readiness. Moreover, advanced autonomy. incremental investments that raise TRL are ap- Barrier 3: Lack of Inter-Directorate Co- propriate. Autonomy, especially where multiple ordination. The Human Exploration Direc- functions are integrated, has implications for the torate (HEOMD), the Space Technology Mis- system’s architecture, design, development and sion Directorate (STMD), and divisions of Sci- operations processes, and personnel throughout. ence Mission Directorate (SMD) have struggled It requires coordinated efforts throughout devel- to coordinate and lack direct incentive to develop opment phases and across domains and cannot NASA-wide plans to implement autonomous sys- be comprehensively adopted through incremental tems. This results in fragmented and incremen- investments. This means standard paths to tal progress that will not lead the agency to sus- mature technologies do not apply to autonomous tainable advances in multi-mission autonomy. Al- systems. Experience has shown that technology ternatively, coordinated investments could allow demonstration missions, like the Remote Agent directorates to leverage developments by others, Experiment on Deep Space-1, do not transition promoting agency-wide sharing of standards and to routine science mission use [42]. practices, increasing the likelihood autonomous Barrier 2: Institutional environment lim- functions are reflown, and minimizing duplicated its autonomy to incremental maturation investments. Meanwhile, private industry bene- and thus restricts NASA’s ability to deploy fits from investments in autonomy and can push autonomous systems. Even though autonomy forward with increasingly autonomous solutions, has the potential to reduce mission risk, using e.g. the Falcon fly-back boosters used by SpaceX. it is still a perceived risk. NASA has success- 5 Recommendations fully executed missions without significant auton- omy for decades and is capable of developing and Given how autonomy would enable and enhance evaluating missions that are enhanced or enabled strategic missions of interest to the planetary sci- by autonomy. As exemplified by AutoNav and ence community, we offer the following high-level AEGIS and described in the Sagan Report and recommendations for enabling the routine deploy- ESA OPS-SAT paper, a cultural barrier limits ment, and continued evolution, of autonomy for opportunities to fly enhancing technologies. Risk future planetary science missions. These high- aversion results in incremental maturation of spe- level recommendations include several potential cific autonomous functions, which struggle again implementations described as examples. to find their place on science missions despite past 1. Create programmatic pathways that success and potential advantages. Even when prepare integrated autonomy systems risks are taken on competed missions and they for future missions and build institu- would stand to benefit in science-to-dollar ratio, tional trust, e.g.: e.g. Dragonfly, opportunistic science is not con- sidered an intrinsic component of baseline mis- • Commit to advancing autonomy by setting sions. As also noted in the Sagan Report and still a series of capability deadlines to include true today, NASA struggles to pioneer software increasing amounts of autonomy on all

6 planetary science missions. This would en- 3. Invest in practices that promote the sure that NASA is ready with the needed multi-mission use of autonomous sys- processes and capabilities when the time tems. Practices includes design, devel- comes to fly the more ambitious missions opment, test, verification, and opera- that are enabled by autonomy tions processes and standards, e.g.: • Incentivize adoption of autonomous sys- • Develop common architectural patterns, tems for PSD Announcements of Opportu- principles and standards to enable con- nity (AOs), e.g. through a cost incentive. fident integration of autonomy technolo- • Coordinate STMD investments with com- gies; invest in updates to development and peted missions, e.g. SIMPLEx, so missions operations processes to enable the trust- push boundaries of science exploration and worthy deployment of increasingly au- technology demonstration tonomous missions • Expand programs like the New Frontiers Homesteader Program and ROSES to in- • Update mission selection and review pro- clude autonomous functions, system inte- cesses to consider assessment of agency gration, operator tools, and verification risk posture, e.g., the risk of not includ- methods for autonomous systems. (See ing new technology or methods in NASA Rec #2) planetary science missions. • Instrument AOs for Flagships should offer • With inter-directorate coordination, in- an opportunity for collaborative propos- vest in laboratory and virtual autonomy als with complementary instruments using testbeds so NASA centers and industry autonomy for payload data processing and can test and validate autonomy software science planning and train on new processes. (See Rec #2) • Use all extended missions for demonstrat- • Spur the adoption of integrated au- ing autonomous functions tonomous functions in industry to support NASA’s competed missions. (See Rec #1) 2. Invest in infrastructure for developing and supporting autonomous systems in • Adopt new maturity evaluations for soft- space, e.g.: ware and model trust to augment TRL in assessing autonomous functions, their in- • With inter-directorate coordination, invest tegration, and applications. in an in-space autonomy testbed, poten- tially utilizing SmallSats, so NASA cen- 4. Set goals to reduce operations costs ters and industry can test and flight vali- and determine the degree of au- date flight and ground software and train tonomous operations required to on new processes (See Rec #3) achieve these goals. NASA should commission an independent study, poten- • Make DSN and Advanced Multi-Mission tially performed by the National Academy Operations System (AMMOS) invest- of Sciences, assessing existing operations ments that support the anticipated growth to evaluate how operations costs can be of customer missions and use of au- reduced by adopting autonomy and what tonomous systems funding and savings profiles would result. • Expand Homesteader and ROSES op- portunities to cover interdisciplinary au- 6 Acknowledgements tonomous research and development (See Thanks to Ellen Van Wyk (NASA JPL) for illus- Rec #1) trations. • Set specific objectives and a time frame for transitioning AMMOS away from a priori References planning (e.g., change from time-based se- [1] Stewart A. Collins. The Mariner 6 and 7 quencing to goal-based commanding). Set pictures of Mars. Scientific, Technical In- a date where all new missions would be formation Office, National Aeronautics, and expected to use this new paradigm. Space Administration, 1971.

7 [2] Mark Brown. Programmable Deep Space [14] R Francis et al. “AEGIS autonomous tar- Autonomy: The First 25 Years. Apr. 1997. geting for ChemCam on Mars Science Lab- [3] Planetary Mission Concept Studies Work- oratory: Deployment and results of initial shop 2020. https://www.hou.usra.edu/ science team use”. In: Science Robotics 2.7 meetings / pmcs2020 / format/. (Accessed (2017). on 09/08/2020). [15] Abigail R Azari et al. “Integrating Ma- [4] 2018 Workshop on Autonomy for Fu- chine Learning for Planetary Science: Per- ture NASA Science Missions: Output spectives for the Next Decade”. In: arXiv and Results — Science Mission Direc- preprint arXiv:2007.15129 (2020). torate. https : / / science . nasa . [16] Bethany Theiling et al. “Non-Robotic Sci- gov / technology / 2018 - autonomy - ence Autonomy Development”. In: Plan- workshop/output-results. (Accessed on etary Science Decadal 2020 White Paper 08/13/2020). (2020). [5] Carl Sagan and Rajashekar Reddy. Ma- [17] Benjamin Donitz and et al. New Frontiers chine Intelligence and Robotics Report of Mission Concept Study to Explore Oort the NASA Study Group. 1980. Cloud Comets. 2020. [6] David J Evans. “OPS-SAT: FDIR Design [18] Shyam Bhaskaran et al. 2018 Workshop on on a Mission that Expects Bugs-and Lots of Autonomy for Future NASA Science Mis- Them”. In: 14th International Conference sions: Small Bodies Design Reference Mis- on Space Operations. 2016, p. 2481. sion Reports. 2018. [7] NASA Autonomous Systems – Systems [19] Kiri L Wagstaff et al. “Enabling Onboard Capability Leadership Team. Autonomous Detection of Events of Scientific Interest for Systems Taxonomy. 2019. the Europa Clipper Spacecraft”. In: Pro- [8] Steve Chien and Kiri L Wagstaff. “Robotic ceedings of the 25th ACM SIGKDD Inter- space exploration agents”. In: Science national Conference on Knowledge Discov- Robotics 2.7 (2017). ery & Data Mining. 2019, pp. 2191–2201. [9] Ralph D Lorenz et al. “Dragonfly: A ro- [20] Glenn Reeves and et al. “Autonomy Chal- torcraft lander concept for scientific explo- lenges and Development for Europa Lan- ration at Titan”. In: Johns Hopkins APL der”. In: Planetary Decadal 2020 White Pa- Technical Digest 34.3 (2018), pp. 374–387. per (2020). [10] Daniel G Kubitschek et al. Deep Impact Au- [21] D. Wang et al. “Using Flexible Execution, tonomous Navigation: the trials of targeting Replanning, and Model Parameter Updates the unknown. 2006. to Address Environmental Uncertainty for a Planetary Lander”. In: Intl Symposium [11] Andres Castano et al. “Automatic detec- on AI, Robotics, and Automation for Space tion of dust devils and clouds on Mars”. (2020). In: Machine Vision and Applications 19.5- 6 (2008), pp. 467–482. [22] Martha Gilmore et al. “Venus Flagship Mis- sion Planetary Decadal Study”. In: Interna- [12] Tara A Estlin et al. “Aegis automated tional Planetary Probe Workshop Webinar science targeting for the mer opportunity Series. 2020. rover”. In: ACM Transactions on Intelli- gent Systems and Technology (TIST) 3.3 [23] John Elliott, Mark Robinson, and et (2012), pp. 1–19. al. “Intrepid Planetary Mission Concept Overview”. In: International Planetary [13] Steve Chien et al. Using autonomy flight Probe Workshop. 2020, Webinar Session 6. software to improve science return on Earth Observing One. 2005. [24] Carolyn Ernst and et al. Mercury Lander. url: https://www.youtube.com/watch? v=SKH6oymc93M.

8 [25] Sanae Kubota, Carolyn Ernst, and et al. [37] Ian Garrick-Bethell. “NanoSWARM: A “Mercury Lander Planetary mission con- Nano-satellite Mission to Measure Particles cept study for the 2023-2032 Decadal”. and Fields Around the Moon”. In: AGUFM In: International Planetary Probe Work- 2015 (2015), EP53C–1043. shop Webinar Series. 2020. [38] Les Deutsch. Today’s Deep Space Com- [26] Chad Edwards and et al. “Emerging Capa- munications System: The Earth-based Ele- bilities for Mars Exploration”. In: Planetary ments. 2019. Science Decadal 2020 White Paper (2020). [39] Lori Sherea Glaze and James Brian Garvin. [27] LH Matthies et al. “Robotics Technology “VICI (Venus in situ composition inves- for In Situ Mobility and Sampling”. In: tigations): the next step in understand- Planetary Science Decadal 2020 White Pa- ing Venus climate evolution”. In: AGUFM per (2020). 2017 (2017), P52A–06. [28] Daniel Gaines et al. Self-Reliant Rover De- [40] Ross M Jones. Surface and atmosphere geo- sign for Increasing Mission Productivity. chemical explorer (SAGE) baseline design 2018. from March 2003 Team X studies. 2003. [29] Soumyo Dutta and et al. “Aerocapture as [41] The Planetary Society. The Planetary Ex- an Option for Ice Giants Missions”. In: In- ploration Budget Dataset. https : / / ternational Planetary Probe Workshop We- www . planetary . org / space - policy / binar Series. 2020. planetary - exploration - budget - [30] Alex Austin and et al. “Enabling and En- dataset. (Accessed on 09/07/2020). hancing Science Exploration Across the So- [42] Douglas Bernard et al. Spacecraft autonomy lar System: Aerocapture Technology for flight experience: The DS1 Remote Agent SmallSat to Flagship Missions”. In: Inter- experiment. 1999. national Planetary Probe Workshop Webi- nar Series. 2020. [31] T. Marshall Eubanks et al. “Exobodies in Our Back Yard: Science from Missions to Nearby Interstellar Objects”. In: Planetary Decadal 2020 White Paper (2020). [32] Julie Castillo and John Brophy. Ceres Plan- etary Mission Concept Study: Sampling Brines from an Evolved . 2020. [33] Wendy Calvin. PCMS MORIE. A Plan- etary Mission Concept Study: Mars Or- biters for Resources, Ices and Environ- ments. 2020. [34] Rob Lillis and the MOSAIC Team. MO- SAIC: A Planetary Mission Concept Study. Mars Orbiters for Surface-Atmospheric- Ionospheric Connections. 2020. [35] Carly Howett. Persephone: A Pluto-System Orbiter & Kupier Belt Explorer. 2020. [36] Daniel Grebow et al. “Dawn safing ap- proach to Ceres re-design”. In: AIAA/AAS Astrodynamics Specialist Conference. 2016, p. 5426.

9