The Agent-Environment Boundary
Dhawal Gupta
Abstract
This paper examines the agent-environment boundary in reinforcement learning (RL) from a new perspective. We suggest that the traditional distinction between actions and observations can evolve as the agent's control and understanding of its environment grow. We illustrate these concepts with simple examples, showing how shifting boundaries allow us to define the notion of building knowledge and an interplay between RL and planning.
Chat is not available.
Successful Page Load