Pursuit, Avoidance, and Cohesion in Flight: Multi-Purpose Control Laws and Neuromorphic VLSI

Abstract : Pursuing evasive prey, avoiding collisions with obstacles, and coordinating one's position in a group (formation) are some of the most important sensorimotor tasks that animals must perform. When studied in isolation, best-fit control laws can be dramatically different, but in an animal brain these behaviors are likely to be implemented by a small number of simultaneously-active and coordinating brain areas that work together to form a flexible amorphous behavior that can dynamically emphasize one or more subgoals. In this proposal, we seek to unify a number of different task-dependent control laws at multiple levels of abstraction from theoretical equations of control down to known neural substrates. Through carefully-chosen physical implementations (custom Neuromorphic VLSI and robotics) we will apply important practical constraints that can lead to deeper insight into how and why efficient, real-time systems exhibit certain behavior. Using the echolocating bat as a model system, we seek to unify prey pursuit control laws with formation control and collision avoidance, and to understand the theoretical consequences of stochastic sensory data and delays in the system. To support this work, our budget will cover the training of graduate students in the Controls area and the Microelectronics area, the submission of VLSI designs for fabrication through the MOSIS fabrication service, and the construction of robot and sonar testbeds for closed-loop behavioral experiments.