Averaged Soft Actor-Critic for Deep Reinforcement Learning

With the advent of era artificial intelligence, deep reinforcement learning (DRL) has achieved unprecedented success in high-dimensional and large-scale intelligence tasks. However, insecurity instability DRL algorithm have an important impact on its performance. The Soft Actor-Critic (SAC) uses advanced functions to update policy value network alleviate some these problems. SAC still In order reduce error caused by overestimation SAC, we propose a new called Averaged-SAC. By averaging previously learned action-state estimates, it reduces problem soft Q-learning, thereby contributing more stable training process improving We evaluate performance Averaged-SAC through games MuJoCo environment. experimental results show that effectively improves stability process.