Using Backpropagation to Learn the Dynamics of a Real Robot Arm.

摘要

Computing the inverse dynamics of a robot arm is an active area of research in the control literature. We apply a back propagation network to this problem and measure its performance on the CMU Direct-Drive Arm II for a family of pick and place trajectories. Trained on a random sample of these trajectories, the network is shown to generalize top new samples drawn from the same family. The weights developed during the learning phase are reminiscent of the velocity and acceleration filters used in standard control theory. Keywords: Robotics; Manipulators; Neural networks; Robot control; Learning manipulator dynamics; Artificial intelligence. (kt)