A Generalizing Spatial Representation for Robot Navigation with Reinforcement Learning

Abstract

In robot navigation tasks, the representation of the surrounding world plays an important role, especially in reinforcement learning approaches. This work presents a qualitative representation of space consisting of the circular order of detected landmarks and the relative position of walls towards the agent’s moving direction. The use of this representation does not only empower the agent to learn a certain goaldirected navigation strategy, but also facilitates reusing structural knowledge of the world at different locations within the same environment. Furthermore, gained structural knowledge can be separated, leading to a generally sensible navigation behavior that can be transferred to environments lacking landmark information and/or totally unknown environments.