Independent RL Lab

Experiments

A chronological record of the experiments, including their questions, methods, status, and reproducibility materials.

Shortcut or shelter?

The policy boundary in a noisy maze

An exposed route is shorter. A protected route is slower. This experiment solves for the reliability at which the preference changes, then tests whether three TD-control methods recover the boundary implied by the objective they learn.

Q-learning · SARSA · Expected SARSA