Special-Purpose Computing for Neurobiologically Inspired Networks

Current practice in robotics is to control robots with finite state machines operating on sequential processors. However, deterministic algorithmic programming limits the potential for adaptation in unpredictable environments. We propose an alternative approach, which is to control robot behavior with electronic nervous systems that implement the adaptive control schemes of simple animal models. Animals readily adapt to new circumstances by squirming out of impediments. Discrete time map-based neurons and synapses can be configured to implement the major features of the command neuron, coordinating neuron, and the central pattern generator organization of the lamprey nervous system. Exteroceptive reflexes can be readily instantiated for heading control, beacon tracking, as well as flow and orientation relative to gravity. Coupling feedback indicating hindrance or precarious conditions to increases in intrinsic variability at specific loci in the control system can maximize adaptability when entangled or trapped. In this proposal, we plan to evaluate the performance of an electronic nervous system in a beacon-tracking mission of significance to mine neutralization in natural sea environments. We will compare its performance with an identical robot controlled by a more conventional state machine. The mission will involve homing on a sonar beacon.