Proximal policy optimization via enhanced exploration efficiency

Proximal policy optimization (PPO) algorithm is a deep reinforcement learning with outstanding performance, especially in continuous control tasks. But the performance of this method still affected by its exploration ability. Based on tasks, paper analyzes original Gaussian action mechanism PPO algorithm, and clarifies influence ability performance. Afterward, aiming at problem exploration, an enhancement based uncertainty estimation designed paper. Then, we apply theory to propose proximal intrinsic module (IEM-PPO). In experimental parts, evaluate our multiple tasks MuJoCo phsysical simulator, compare IEM-PPO curiosity (ICM-PPO). The results demonstrate that performs better terms sample efficiency cumulative reward, has stability robustness.