Sensor-based control of robotic manipulators using a general learning algorithm

A practical learning control system is described which is applicable to the control of complex robotic systems involving multiple feedback sensors and multiple command variables during both repetitive and nonrepetitive operations. In the controller, a general learning algorithm is used to learn to reproduce the relationship between the sensor outputs and the system command variables over particular regions of the system state space. The learned information is then used to predict the command signals required to produce desired changes in the sensor outputs. The learning controller requires no a priori knowledge of the relationships between the sensor outputs and the command variables, facilitating control system modification for specific applications. The results of two learning experiments using a General Electric P-5 manipulator are presented. The first involved learning to use the video image feedback to position the robot hand accurately relative to stationary objects on a table, assuming no knowledge of the robot kinematics or camera characteristics. The second involved learning to use video image feedback to intercept and track objects moving on a conveyor. In both experiments, control system performance was found to be limited by the resolution of the sensor feedback data, rather than by control structure limitations.