Calibration-Free Robots

Abstract

Robots not needing any quantitative models of themselves or of the world and, thus, no calibration of their sensors, their actuators or their kinematics promise great advantages over conventional robots in terms of robustness, adaptability and cost of ownership. Concepts for such robots are introduced and results obtained in real-world experiments are reported. Similarly to infants and young animals, such robots learn relationships between internal motion control commands on one hand and resulting effects on data produced by their sensors through interactions with the world on the other hand. On the basis of such knowledge they reach goals in the world by controlling their motions in such a way that equivalent goals in their internal sensor data space are reached. Experiments show that object manipulation and an automatic build-up of skill by such robots has been accomplished.