Episode Summary: Dr. Ayers provides a comprehensive overview of his development of autonomous underwater robots, intended to help discover and destroy dangerous underwater land mines. He provides his perspective on two major obstacles facing robotics, including the concept of autonomy, providing valuable insight in light of the current events surrounding the development of autonomous AI.
Guest: Dr. Joseph Ayers
Expertise: Behavior Biomimetics, Robotics, Neurophysiology
Recognition in Brief: Dr. Ayers is the author of several published books, including Neurotechnology for Biomimetic Robots and Dr. Ayers Cooks with Cognac. He has had 50+ academic articles published in the areas of Robotics, Lobsters, and Lower Vertebrates, and has also developed over 35 analytical programs for Macintosh over the past two decades. Not least, he spearheaded the development of the RoboLobster and related research projects, in part with funding by DARPA, for which he has received coverage by The New York Times, The Wall Street Journal, Wired, and many other reputable publications.
Current Affiliations: Professor of Marine and Environmental Sciences at Northeastern University
Biomimetics Unlocks Doors for Robotics
Joseph Ayers got his start in biomimetics studying lobsters. As a graduate student in 1970, he began focusing on the control of walking legs in lobsters and became interested in the problem of how they use the same neural circuit to walk in different directions. This led to establishing a hypothetical network model and doing some behavioral experiments.
Around that time, he also did some work on the sea lamprey, a parasitic organism found on the coast of the Northern Atlantic Ocean. The lamprey can recover from almost total spinal cord damage, and Ayers set to work on figuring out how its nerve system rewires and allows it to swim soon after damage.
Based on his research, he discovered that the lobster and lamprey are similar in that they both have a command neuron, coordinating neuron and central pattern generator (CPG) model of the organization of innate motor systems. Ayers recognized this as a general model that could potentially be used to build robots with a similar architecture. He was funded by DARPA to build a lobster and a lamprey robot.
The robots that Joseph eventually built were unique compared to the more standard models being delivered at the time by institutions such as MIT, which were typically built based on a known operation of neural circuits. After initially getting the robots operational, Ayers quickly realized that using this algorithmic state machine architecture would require a human to program an escape strategy for every possible contingency, a next to impossible task; otherwise, the robot gets stuck. This is an ongoing obstacle in the development of autonomous robots. Ayers realized that the robot’s ability to recover its telecommunication link would be imperative to its success.
The solution was using nonlinear dynamic models of neurons. Lobster neurons have only 4 degrees of dynamical freedom, and Ayers teamed up with physicists who were able to come up with equations that described these relations. What the team discovered was a model for an analog computer.
“The electronic neurons become parts, like legos”, remarked Ayers. “When we first started working with them (the electronic neurons)…we first did perturbation experiments and realized they were indistinguishable from living neurons.” Ayers realized that he could model a nervous system with this technology. The next step was to take the model of the lobster walking legs that he had built in grad school, and use this system in tandem with the neuronal model. The result was a robot that behaved like a real lobster – a RoboLobster.
How the World of Biology is a Model for the World of Robots
Ayers’ team’s success in developing a complete robot in lobster form that successfully mimics how the animal looks, moves, behaves – and perhaps thinks – is a tribute to the continuing potential of biomimetics in AI. There is an important point to be gleaned, says Ayers. “If you try and control robots with computer programs, unless you anticipate all possible contingencies and program them in, the robot is going to get stuck; animals never get stuck.”
Animal movements are chaotic, and their analog-like neuronal networks are able to account for the continuous spectrum of possibilities that exist in the natural environment. Electronic neurons, like the ones created for the lobster and lamprey robots, account for variable chaos. The long-term goal in robotics is to build controllers with a level of perceived chaos that gets them to wiggle and squirm like real animals, learning through sensory input.
The current trend is to use comparative physiology i.e. find an animal model that exhibits the type of behavior that you want to mimic, and develop potential models for creating similarly-performing AI. Both lamprey and lobsters are predators, and this is the type of investigative behavior you want in underwater machines that hunt. “Mimicking behavior needs us to approximate physics as close to (the actual animal) as possible”, remarks Ayers. Lampreys and lobsters are a logical jumping off point, as both happen to have a uniquely replicable neuron system.
Why DARPA is Interested in Robotic Lobsters
Ayers was funded by DARPA because of its continuing interest in robots that can do what humans can not or will not do – which includes destroying underwater mines that are difficult to locate and destroy in shallow water or hidden near beaches. An undulatory robot (like a lamprey) with scanning sonar can pragmatically look for hanging mines, while a walking lobster robot can deftly traverse the clutter of the sea floor in its search for mines. The ultimate goal is for the two robots to be able to communicate with each other through stigmergy, an organizational principal used by social insects.
Ayers is currently being funded by the Office of Naval Research to give these robots a sense of smell using synthetic biology, with the hope that they will eventually be able to “sniff out” explosive materials and other agents of harm.