Reflective Policy Optimization.

arXiv:2406.03678v2 Announce Type: replace-cross Abstract: On-policy reinforcement learning methods, like Trust Region Policy Optimization (TRPO) and Proximal Policy Optimization (PPO), often demand extensive data per update, leading to sample inefficiency. Thi