Robot learning assisted by perception-based information: a computing with words approach

Abstract

Sensor-based operation of autonomous robots in unstructured environments has been proved to be an extremely challenging problem. However, humans seem to cope very well with uncertain and unpredictable environments, often relying on their perceptions. Furthermore, humans can also utilize the perceptions to guide their learning on those parts of the perception-action space that are actually relevant for the task. To make use of perceptions to assist robot learning and control, by using computational theory of perceptions (CTP), a linguistic version of Lyapunov synthesis working with fuzzy arithmetic operations in the domain of computing with words (CW) is proposed to derive a set of stable fuzzy control rules from the perception-based information. Then the fuzzy rules are incorporated in a fuzzy reinforcement learning (FRL) agent to accelerate its learning. The experimental and simulation results show that it is possible for a robot to start with the perception-based information and then refine its behavior through further learning.