Neuromorphic architectures for fast adaptive robot control

Abstract

We suggest an architecture for an adaptive neuromorphic system designed to control a robot. The proposed architecture utilizes two important features of neural networks: the abundance of local minima in the network's state space and the uniformity of convergence to these minima in the face of growing dimensionality. The proposed approach is expected to yield controllers which are both faster and simpler than controllers which are designed via the methods of Model Reference Adaptive Control and Self Tuning Regulator. We expect the controller's complexity not to grow exponentially with the number of unknown parameters, and to allow adaptation in both continuous and discrete parameter domains. The possible benefits of our architecture are demonstrated on a single degree-of-freedom manipulator, whose controller is assisted by a neural estimator.