Acquisition by robots of danger-avoidance behaviors using probability-based reinforcement learning

Abstract

Robots are being used more and more in dangerous environments such as space and disaster areas. However, when robots are at risk in dangerous environments, the time during which robot operators can issue risk avoidance instructions is limited. Therefore, robots should be able to acquire behaviors that enable them to autonomously avoid danger. In this paper, we present a probability-based reinforcement learning (PrRL) method and apply it to robot behavior acquisition.