Adaptive neural networks for vision-guided position control of a robot arm

A robot arm controlled by a neural network architecture using visual input is described. The arm can reach targets without prior geometric calibration. All control signal effects are learned by the controller through visual observation during a training period and refined during actual operation. Minor changes in the system's configuration result in a brief period of degraded performance while the controller adapts (in real time) to the new mappings. It is shown that a neural-network-based controller can perform accurately, taking into account the nonlinearities of various transformations. Such a controller is easy to train, tolerant of imprecise equipment configurations, and insensitive to camera perturbations following training. The method is simpler than traditional control techniques, which require the solution of the inverse perspective projection and inverse kinematics of the system.<
>