Partial Consistency for Stabilizing Undiscounted Reinforcement Learning | IEEE Journals & Magazine | IEEE Xplore