Self-Motivated, Task-Independent Reinforcement Learning for Robots

摘要

This paper describes a method for designing robots to learn self-motivated behaviors rather than externally specified behaviors. Self-motivation is viewed as an emergent property arising from two competing pressures: the need to accurately predict the environment while simultaneously wanting to seek out novelty in the environment. The robot’s internal prediction error is used to generate a reinforcement signal that pushes the robot to focus on areas of high error or novelty. A set of experiments are performed on a simulated robot to demonstrate the feasibility of this approach. The simulated robot is based directly on an existing platform and uses pixelated blob vision as its primary sensor.