Scheduling That Speaks: An Interpretable Programmatic Reinforcement Learning Framework

Researchers introduced ProRL, an interpretable program-based reinforcement learning framework for job shop scheduling.

ProRL searches a policy space defined by a domain-specific language and uses Bayesian optimization to complete partial scheduling programs.

On benchmark scheduling tasks, it matches or outperforms heuristic and deep RL baselines even with limited training episodes.

The approach makes scheduling decisions more editable and inspectable, helping address trust and compute constraints in neural policies.