Draw the agent–environment boundary - On Policy Distillation | Zoonk

Draw the agent–environment boundary - On Policy Distillation | Zoonk

Draw the agent–environment boundary

x

A simple split diagram with a vertical boundary line: left side labeled “Agent: chooses action” showing a small decision box, right side labeled “Environment: reacts” showing a game world receiving the action arrow; keep only these labels and one arrow crossing the boundary.

Draw the line first

An agent–environment boundary is the line between the thing choosing actions and the world that reacts. In reinforcement learning, drawing this line clearly stops you from mixing up the decision-maker with its feedback system.