Oscillatory Neuromorphic Computing for Robotics

Physical Sciences Inc. proposes to develop an oscillator-based neurocomputer for associative memory and sensor fusion. The oscillatory-neurocomputer exploits the properties of nonlinear oscillators to form Central Pattern Generators (CPGs). CPGs are common elements throughout the entire nervous system. Investigators have already started to use silicon central pattern generators for small-scale sensor fusion and direct control of robots. No one has yet examined the use of CPGs as computational elements to build scalable neuromorphic systems. In contrast to competing approaches which typically require n-squared programmable connections and therefore do not scale well to large real-world problems, these systems require only order n connections and the sensing elements feed directly into the network to change the connection. Dynamic changes in the sensors or actuators (e.g. motor drive current) automatically change the dynamics of the network. Our Phase I research will focus on mapping the detailed dynamics of small-scale hardware networks of CPGs with the objective of using them as components for building larger neuromorphic systems for controlling real-world processes. In Phase II we propose to integrate our neuromorphic computational engine with a mobile robot.