New Techniques for Teach-In, Acceleration and Learning in Sensor-Controlled Robots

Abstract The paper tries to outline a unified approach to sensory feedback in robotics . By a few generalizing definitions relating positions/orientations and (pseudo-) forces/torques the problem of teaching a robot a certain task including continuous sensory Information, is Simplified. Already in the teach phase the paths are generated via and together with sensory pattern motion commands and sensor data are stored together. The latter ones are then availab1e as reference values for the repetition mode in a possibly changing environment. Self-adaptions and teacher-induced corrections may be superposed in an associative memory. A New data compression concept that allow to store long paths with only a few parameters and that Compensate robot dynamics in high speed operations are outlined.