Application Of Neural Networks To Adaptive Control

Neural network-based control architectures have been developed that can autonomously learn to perform kinematic and dynamic control of unknown systems and/or adapt to systems that change over time. Kinematic control of non-linear, time varying systems is demonstrated. The controller can control continuous-valued system variables to arbitrary accuracy using a small number of neurons. It learns to control the system more accurately than an analytically calculated control function. It is fault tolerant in the presence of a large number (e.g. 30%) of component failures. The architecture has been used to learn to control a simulated robot arm of initially unknown characteristics. Dynamic feed forward control of a second order system is also demonstrated. The simulations run in near real time, and custom VLSI hardware is under development.