A technique that restricts an RL agent to only take feasible actions by preventing invalid choices during decision-making.