Braitenbergian experiments with simple aquatic

Rustam Stolkin, Richard Sheryll, Liesl Hotaling Stevens Institute of Technology Hoboken, NJ 07030, USA

Abstract —This paper describes the development of a short introductory underwater course, aimed at college freshman and high school and middle school students. During these courses, students work in teams to build and program underwater robots using a combination of LEGO and other simple materials. As an introduction to ideas of and programming, students undertook a practical exploration of the concepts developed by cybernetician Valentino Braitenberg in his famous book “Vehicles: Experiments in Synthetic Psychology”. Over five laboratory sessions, students gradually evolved their own designs for waterborne “robotic amoebas” through a series of progressively more complex design challenges. These courses build on our previously reported work in which students have built underwater Remotely Operated Figure 1. A programmable AUV with light sensors, built using a Vehicles using similar materials and educational strategies. This combination of LEGO and other simple materials. work is now being adapted for dissemination to large numbers of middle and high schools across New Jersey through a grant from II. MATERIALS the National Science Foundation. Students were provided with a selection of LEGO including I. INTRODUCTION several motors, battery boxes and leads, gearing, structural and Valentino Braitenberg’s famous text “Vehicles-experiments in mechanical components. Also provided, were a selection of synthetic psychology”, [1], uses a series of elegant thought plastic propellers (obtainable from hobby stores) mounted on experiments, involving simple imaginary vehicles equipped LEGO axles. Additional materials included Styrofoam, with motors and sensors, to explain how seemingly complex modeling clay, a selection of weights (nuts and bolts work animal behaviours such as attraction, repulsion, fear and well), rubber bands, string and duct tape. A 30 inch deep aggression, can result from combinations of simple inflatable pool was used to test the designs. mechanisms. Braitenberg’s explanations are profound in their For programmable , students used the LEGO implications for roboticists and neuro-scientists, yet so simple NXT controller (figure 2), sealed inside a plastic box, LEGO and intuitive that they are immediately accessible to readers of robotics sensors, including touch sensors and light sensors all levels, without any prior knowledge or expertise. (which can be waterproofed using simple materials such as This paper describes the development of a short introductory clingflim), and the simple icon based NXT-G programming course, aimed at college freshmen, high school and middle system. school students, enabling a practical exploration of Braitenbergian ideas through constructing, programming and testing a series of progressively more complex waterborne robot vehicles, also known as Autonomous Underwater Vehicles (AUVs), e.g. figure 1.

Manuscript received August 10 th , 2007. We thank Costas Chassapis, Dir. Dept. Mech. Eng., Stevens Institute of Technology, for funding the equipment and materials to test and develop this project. R. Stolkin is a research Assistant Professor at the Center for Maritime Systems, Stevens Institute of Technology, Phone: 201-216-8217; e-mail: [email protected] . Richard Sheryll is an instrumentation designer and also a PhD candidate in Ocean Figure 2. The LEGO NXT programmable brick set in a watertight housing. Engineering at the Center for Maritime Systems, Stevens Institute of Rubber buttons, set in the housing, enable the controls on the NXT to be Technology, email: [email protected] . Liesl Hotaling is Assistant Director pressed. Alternatively a diver’s “pelican” box with snap shut lid can be of the Center for Innovation in Engineering and Science Education at Stevens used (figure 1). A LEGO plate is bonded to the underside of the housing so Institute of Technology, email [email protected] . that students can add their own LEGO structures and motors.

III. WHY BUILD UNDERWATER ROBOTS ? Although discovery learning is frequently employed in an When students design, build and program underwater robotic early childhood development setting, the instructional model vehicles, they are learning engineering fundamentals which offers several advantages to a high school or undergraduate span virtually every engineering discipline. Additionally, setting. It arouses students’ curiosity, motivating them to students are motivated by an exciting and stimulating design continue to work until they find answers, [6]. Students also scenario. learn independent problem solving and critical thinking skills The use of projects based on small robotic vehicles is now because they must independently analyze and manipulate widespread in engineering curricula, however these are information. predominantly wheeled, terrestrial vehicles. Such projects often Students often benefit more from being able to engage in reduce to little more than exercises in applied programming, active learning by “seeing” and “doing” things than from losing valuable opportunities to present substantial mechanical passive learning by listening to lectures. Tackling material challenges or to incorporate real interdisciplinary engineering from several perspectives and persevering with unresolved design. In contrast, the underwater environment presents problems improves students’ core intellectual skills - they learn unique design challenges and opportunities. The motion of an how to learn independently. Cognitive development is not the underwater vehicle, through a three dimensional space with six accumulation of isolated pieces of information; rather, it is the degrees of freedom, is more complex. Additional engineering construction by students of a framework for understanding issues include propulsion, drag, buoyancy and stability. their environment. Teachers should serve as role models and Practical construction problems include how to waterproof facilitators by solving problems with students, explaining the electrical components. The challenge of creating a robot which problem solving process and talking about the relationships can be sent to explore a hostile and inaccessible environment is between actions and outcomes. Observing students during their also motivating and stimulating to many students. activities, examining their solutions and listening carefully to The aquatic environment is also preferable for investigations their questions can reveal much about their interests, modes of of Braitenbergian ideas since it more closely resembles the thought and understanding or misunderstanding of concepts, “primordial soup” in which Braitenberg envisions the evolution [7]. of simple amoeba-like vehicle behaviours. Discovery based learning is a particularly effective means of teaching the iterative approach to engineering design. Our IV. WHY USE LEGO? students are encouraged to approach engineering problems through an iterative sequence of steps: Design/Test/Modify Our students work with a combination of LEGO and (figure 1). In contrast, surprisingly little of conventional additional simple materials. LEGO is particularly suited to engineering curricula are devoted to this design process, with discovery based learning due to its ease and speed of assembly, the learning experience of engineering students often bearing [2], [3]. This speed reduces the time between conception of an little resemblance to the activities of professional engineers in idea and its implementation, enabling students to discover industry. through trial and error, rapidly test a range of alternative designs and evolve their designs iteratively by observing the VI. OVERVIEW OF THE STEVENS “INTRODUCTION TO relationship between structure and function. In contrast, when UNDERWATER ROBOTICS ” PROGRAM students use conventional materials, which must be sawed, drilled, glued, screwed or welded, the construction process is Educators and engineers at Stevens Institute of Technology lengthy and frustrating. Time constraints prevent students from are currently engaged in developing a set of educational evolving their designs through multiple iterations of testing and modules, which teach fundamental engineering principles modification. Often there is no time allotted for the students to through the design, construction and testing of underwater fail, analyze the failure and then modify their design. In robotic vehicles. The strategies incorporated into our contrast “We know that students will learn most deeply and underwater robotics projects foster an active, discovery profoundly when they…have an opportunity to try, fail and learning environment that integrates many mathematical, receive feedback on their work”, [4]. scientific and engineering principles and will support conceptual and skill-based learning, application of principles to V. DISCOVERY BASED LEARNING novel situations, collaborative learning and cooperative group skills. As far as possible we try to build our LEGO underwater Initially we developed a Remotely Operated Vehicle (ROV) robotics classes upon “discovery learning” principles. project in which students build wire guided underwater Discovery learning, [5], is a cognitive instructional model in vehicles equipped with mechanical grabbers. Students then which students are encouraged to learn through active used their ROVs to retrieve objects from the bottom of a pool. involvement with concepts and principles, and teachers This paper describes the initial trial of a follow on course in encourage students to have experiences and conduct which students build programmable Autonomous Underwater experiments that permit them to discover principles for Vehicles (AUVs) which respond intelligently to sensor themselves. stimulus to complete a series of simple autonomous tasks.

These projects were initially pilot tested with high school 3) Add a third motor to the vehicle, enabling vertical motion junior students who participate in our Exploring Career in the water column. Options in Engineering and Science (ECOES) summer 4) Design a motorized mechanical which can program. Following positive feedback from ECOES students, grasp specified objects. the ROV course has now been introduced to our freshman 5) Combine the products of stages 3, 4 and 5 to produce a mechanical engineering curriculum. With a major grant from vehicle which can retrieve the greatest number of objects the National Science Foundation ITEST program, these from the bottom of the pool within a five minute period projects and materials are being adapted and disseminated to (figure 3). large numbers of middle and high school students across New Notice that these progressively more complex stages of the Jersey. robot design, naturally tend to correspond to adding each successive motor or each additional degree of freedom to the VII. PREVIOUS WORK – WIRE GUIDED ROV COURSE robot. Our previous work, [8], describes short courses, in which students design, build and test wire guided Remotely Operated VIII. BRAITENBERG VEHICLES Vehicles (ROVs) equipped with a mechanical grabbing device. “Vehicles – Experiments in Synthetic Psychology” is a book This same course has now been used successfully with middle by Valentino Braitenberg, [1], a famous cybernetician and school, high school and university level engineering students. neuro-anatomist. Braitenberg seeks to explain how the brain In accordance with the principles of discovery learning, may have evolved, how complex behaviors can result from students are not given detailed instructions or pre-packaged simple mechanisms, and particularly why one side of our brains “kits” with which to build their ROV. Instead they are set a controls the opposite side of our body. He does this through a series of design challenges for which they must independently series of elegant thought experiments with imaginary robot invent their own solution. These challenges begin very simply vehicles which consist of motors connected to sensors. and become progressively more complex until the student arrives at a completed ROV by the end of the course. As a final + challenge, each team has to use their ROV to retrieve and - manipulate objects on the bottom of a pool of water (figure 3).

Figure 4. Single motor Braitenberg vehicles with positive and negative sensory feedback (e.g. light-phobic and light-philic respectively).

The simplest Braitenberg vehicle is shown in figure 4. A single motor is connected to a single sensor (e.g. a light sensor). A positive connection indicates that the motor runs faster as the sensed quantity increases. If the sensed quantity were light, the vehicle would speed up and “run away” when it entered bright areas, and tend to slow down and settle in dark areas. Somewhat like a cockroach, we might say that this vehicle is “scared of light” and prefers darkness. Conversely a negative connection between sensor and motor will result in a vehicle that likes to bask in bright areas but “dislikes” darkness and runs away from dark areas. Figure 3. A LEGO ROV with mechanical grabber, built by high school students over five laboratory sessions. The ROV was used to Braitenberg next describes a series of vehicles which consist retrieve wiffle balls from the bottom of a pool. of two motors and two sensors. By either wiring same side or opposite side sensors to the motors using positive connections,

the vehicles will speed up as they approach light, either veering The intermediary design challenges include: away (“cowardice”) or homing in on and ramming 1) Design a surface vessel with a single motor and various (“aggression”) the light source, figure 5. Alternatively, using propeller options, optimizing gearing ratios to maximize negative connections results in vehicles which slow down as speed in a single direction. they approach light, either homing in and stopping (“love”) or 2) Design a surface vessel with steering, using two spending some time near the light before being attracted away independently controlled motors. The challenge involves again on a new journey (“the explorer”), figure 6. negotiating a figure eight course, around two buoys, in the

least amount of time.

Unfortunately, in our experience, relatively few simple educational robotics curricula emphasize this feedback process, which we believe encapsulates the fundamentals of real robotics. There are now numerous kits, projects or “camps on disk”, aimed at getting young students, from middle school age, interested in science and engineering through robotics projects. Frequently these involve students programming a simple, pre-determined sequence of events, without creating a robot that genuinely interacts with a changing and unknown environment.

IX. A PROTOTYPE SHORT COURSE IN AUTONOMOUS UNDERWATER VEHICLES In summer, 2007, 33 high school students participated in a Figure 5. “Aggression” and “Cowardice” behaviors, short course of five laboratory sessions (2 hours each), building using positive sensory feedback. and programming AUV robots, as part of the Stevens Exploring Career Options in Engineering and Science summer program. The aim of this course was to preserve the educational principles and progressive, step by step format of our successful ROV course, while exploring some of the ideas of Braitenberg vehicles. As with our ROV course, students were set a series of progressively more difficult design challenges, gradually adding more degrees of freedom of motion and finally arriving at a fully functional autonomous underwater robot. For challenge 1, students were given a single motor and a pair of touch sensors. They were told to build a simple vehicle which moves in a straight line across the surface of a pool. When the vehicle touches a wall of the pool, the robot’s Figure 6. “Love” and “Explorer” behaviors, using direction is reversed, figure 8. Because the vehicles tend to negative sensory feedback. deviate from straight line motion, this results in a primitive Braitenberg’s ideas are very powerful. They are simple and amoeba-like behavior with the robot repeatedly transecting the accessible to students without prior knowledge or training, yet pool in a random fashion. convey fundamental ideas of feedback control systems and hint at basic principles of neural networks and artificial intelligence. Our aim is to use these principles to convey basic ideas of feedback systems that enable a robot to interact with the world, figure 7.

motors

Figure 8. A simple “robot amoeba” uses touch sensors and “mechanical wiskers” to reverse direction when it encounters the boundary of the pool in which it lives. Figure 7. An intelligent robot learns about a changing world via its Using the highly accessible NXT-G programming system, sensors and responds by using motors to intelligently exert changes on the this behavior can be generated with a very simple program, world (or its own position in the world). This leads to an iterative feedback process. Unfortunately many educational robotics curricula do figure 9. not emphasize this feedback process, but instead have students program a simple pre-determined sequence of actions.

students thus can readily observe the Braitenbergian behaviors but are also able to remotely steer their robot around the pool, which they (the students and perhaps also the robots) find fun.

Figure 9. Icon based NXT-G programming language. “Within a continuous loop, move forwards until a touch sensor is bumped, then move backwards until a touch sensor is bumped. Repeat indefinitely.” For challenge 2, the students begin to implement Braitenbergian ideas. The behavior of challenge 1 is now modified so that the robot’s speed is proportional to light detected by a light sensor. This implements Braitenberg’s most simple robot, as in figure 4. Now the robots move randomly around the pool area, but dislike light and tend to settle in dark regions. This behavior can be coded as in figure 10.

Figure 12. 2D Braitenberg attraction and aversion behavior with the NXT-G language. Depending on which side of the vehicle the motors and sensors are placed, this code can result in the “Agression” or “Cowardice” behaviors – robots home in on the light Figure 10. “Move forwards while continually adjusting speed to be or move to avoid the light. proportional to sensed light level. Once a touch sensor is bumped, repeat but in opposite direction.” Figure 13 shows an example of a robot with two light sensors for Braitenberg homing behaviors, built by high school We can also explore negative Braitenbergian relationships students. This behavior can also be used to make a robot follow between sensed stimuli and motor speed, by setting motor a line of lights, figure 14. speed equal to “100 minus sensed light level” (where sensed light level is also measured on a scale from 0-100), figure 11. Light sensors

Figure 11. “Continuously monitor light levels. Set motor speed proportional to 100 minus light level.” Hence in bright light, vehicle moves slowly, whereas in darkness the vehicle will move fast.

For challenge 3, students are given a second motor and a Figure 13. Underwater robot with two light sensors (waterproofed with second light sensor. They now begin to explore the more clingfilm) for Braitenbergian light homing, built by high school students. advanced Braitenbergian attraction and aversion behaviors of figures 5 and 6. These behaviors can be easily coded in the NXT-G language by using two parallel threads, figure 12. The code in figure 12 causes a robot to continuously update the speeds of motor A and motor C with light levels measured by sensor 4 and sensor 1. Depending on whether sensors 4 and 1 are placed on the same sides or opposite sides of the vehicle as motors A and C, this robot will perform the “Agression” behavior or the “Cowardice” behavior shown in figure 5. We note that LEGO light sensors have a rather narrow field of view, so that it can be frustrating to try to replicate the scenario envisaged by Braitenberg, where robots are naturally attracted to or averted from ambient regions of brightness or Figure 14. Braitenberg’s “aggression” behavior can also be used to darkness. Instead our students were issued with flashlights. The follow a line of lights. robots are attracted to or repulsed by the flashlight beams. The

For challenge 4, students begin sending their robots X. STUDENT FEEDBACK underwater, modifying them to dive and surface. Students are Out of the first 17 high school students to try this underwater given additional motors and learn about buoyancy and robotics course, 14 completed anonymous questionnaires. Archimedes’ principle. They modify the weights and floats on their robots to achieve neutral buoyancy, and can then control Q1) On a scale of 1 to 5, how interesting did you find the depth with motors connected to vertical propellers. The course? students write a simple program that demonstrates this 1 2 3 4 5 Average capability by repeatedly diving to the bottom of the pool and totally very response then re-surfacing, figure 15. boring interesting Num. of 2 5 7 4.4 responses

Q2) On a scale of 1 to 5, how fun did you find the course? 1 2 3 4 5 Average totally very fun response boring Num. of 1 5 8 4.5 responses

Q3) On a scale of 1 to 5, how much do you feel you learned about the following areas of engineering? Rating 1 2 3 4 5 Average Figure 15 . Underwater robot submerged at bottom of pool. response Robotics 2 1 7 4 3.9 The first four challenges were completed in three laboratory Underwater 1 1 5 7 4.3 sessions. The fourth and fifth laboratory sessions were devoted technology to a final challenge – to create a robot that can be deployed Interdisciplinary 3 7 4 4.1 anywhere in the pool and which will seek out and home in on a engineering light source placed on the bottom of the pool, figure 16. Computer 2 6 2 4 3.6 programming Teamwork 1 5 3 5 3.9 skills

Q4) On a scale of 1 to 5, would you have liked to do this activity in your high school or middle school classroom? 1 2 3 4 5 Average certainly very much response not Num. of 3 1 10 4.5 responses

Figure 16. An underwater robot seeks out an underwater light source. Q5) On a scale of 1 to 5, has this course helped stimulate your interest in pursuing an engineering degree? The final challenge was attempted in various ways. Some 1 2 3 4 5 Average students tried to extend the Braitenburg behaviors and combine put me off increased response them with search strategies. Some students tried random engineering interest searches followed by a dive command when a downwards Num. of 1 1 7 5 4.1 responses looking light sensor exceeded a threshold. Other students used

Braitenburg behaviors to guide their robots across the surface of the pool using a flashlight, followed by a dive command XI. LESSONS LEARNED AND FUTURE WORK when a downwards looking light sensor exceeded a threshold. One of the key reasons for attempting an educational course around the theme of programmable underwater robots, was that it would provide a project in which mechanical issues and programming issues were truly integrated and interdependent. Our earlier ROV course successfully explored a range of mechanical design problems over five laboratory sessions. However, trying to squeeze both mechanical tasks and programming / algorithmic tasks into the same short amount of time proved problematic. We suggest that to explore both these

issues properly needs more time. One possibility is to run both Note, although several of our programmable NXT the ROV course and then the AUV course consecutively. controllers did indeed become a little damp from time to time Students might first explore the mechanical issues of during this project, they all subsequently made a full recovery developing a wire guided submersible. They might then begin and appear to have suffered no long term ill effects from their using the NXT computer to control the completed submersible, underwater experience. progressing from a mechanical focus to a programming and algorithmic focus. REFERENCES Submerging computers in a classroom is risky and [1] V. Braitenberg. Vehicles: Experiments in Synthetic Psychology. MIT problematic. It is difficult to waterproof a programmable Press, 1984. controller in a manner which is robust against heavy classroom [2] M.D. Portsmore, C. Rogers. Bringing Engineering to Elementary School. Journal of STEM Education. Vol 5. 2004. wear and tear, remains accessible and usable and is also cost [3] Wang. E., LaCombe, J., and Rogers, C., Using LEGO Bricks to Conduct effective. Diver’s “pelican” boxes provide a very reliable seal Engineering Experiments. Proceedings of the ASEE Annual Conference and a snap-open lid which enables the microprocessor controls and Exposition. 2004. [4] Bain. K., Creating a Natural Critical Learning Environment in Large to be accessed. However, the easily openable lid is source of Lecture Classes. Invited lecture, Stevens Institute of Technology. 2005. worry in a classroom which will always have some disengaged [5] Bruner, J., Toward a Theory of Instruction. Harvard University Press. and inattentive individuals. We have also tried industrial, 1966. [6] Berlyne, D.E., Curiosity and education. In J.D. Krumboltz (ed.), waterproof boxes which bolt closed, with rubber buttons set Learning and the educational process. Rand McNally. 1965. into the lid to enable operation of the microprocessor controls. [7] Slavin, R., Educational Psychology: Theory and Practice. Allyn and With these, we experienced several leaks due to rubber buttons Bacon. 1994. being torn by fingernails or other abuse. The manufacturers [8] R. Stolkin, L. Hotaling, R. Sheryll. A simple ROV project for the engineering classroom. Proc. IEEE / Marine Technology Society seals also proved of poor quality and failed on several OCEANS conference, 2006. occasions. In future work, care must be taken to experiment with a wider variety of boxes and button covers, to determine robust and reliable brands. Another issue with controllers sealed in boxes, is how to download new programs to the controllers. Our students wrote their programs on laptop computers. These programs were then downloaded to the LEGO NXT controllers via Bluetooth, which is able to penetrate the plastic boxes without the need to unseal and reseal them. The NXT controllers are fully Bluetooth enabled and are capable of communicating wirelessly with PCs as well as with each other at ranges of up to 100 meters. This is a powerful capability, however classroom use was problematic. PCs frequently lose contact with their associated NXT and the reconnection process can be highly temperamental, time consuming and frustrating. It is hard to teach students to do this for themselves, especially in a small number of lab sessions, and so it is necessary to have at least one instructor dedicating a large proportion of class time to helping students reconnect their controllers. For this reason, this approach necessitates two instructors for each class. Note also that Bluetooth will only transmit through air and cannot communicate with a vehicle while it is underwater. An alternative solution, which might solve all three of the above concerns, may be to work with “semi-autonomous” robots, i.e. keep the microprocessor outside the water and use it to control the underwater vehicle by wire. This is frustrating in that some of the autonomous nature of the robots would be diminished, however a richer range of classroom activities may be enabled with this approach. Partly, the success of this approach would hinge on finding suitably thin and flexible connecting cables for controlling motors and receiving data from sensors.