CRPO: A New Approach for Safe Reinforcement Learning with Convergence Guarantee.

Tengyu Xu, Yingbin Liang, Guanghui Lan

Browse the full ICML paper archive.