A Strategy-Oriented Bayesian Soft Actor-Critic Model

Adopting reasonable strategies is challenging but crucial for an intelligent agent with limited resources working in hazardous, unstructured, and dynamic environments to improve the system's utility, decrease overall cost, increase mission success probability. This paper proposes a novel hierarchical strategy decomposition approach based on Bayesian chain rule separate intricate policy into several simple sub-policies organize their relationships as networks (BSN). We integrate this state-of-the-art DRL method – soft actor-critic (SAC) build corresponding (BSAC) model by organizing joint policy. compare proposed BSAC SAC other approaches such TD3, DDPG, PPO standard continuous control benchmarks Hopper-v2, Walker2d-v2, Humanoid-v2 MuJoCo OpenAI Gym environment. The results demonstrate that promising potential of significantly improves training efficiency.