Langevin Soft Actor-Critic: Efficient Exploration through Uncertainty-Driven Critic Learning
This work addresses sample efficiency issues for researchers and practitioners in reinforcement learning, presenting an incremental improvement with novel application of Langevin Monte Carlo in continuous control.
The paper tackles the problem of poor sample efficiency in continuous control reinforcement learning by proposing LSAC, a model-free algorithm that uses uncertainty-driven critic learning for exploration, and demonstrates that it outperforms or matches existing methods in experiments.
Existing actor-critic algorithms, which are popular for continuous control reinforcement learning (RL) tasks, suffer from poor sample efficiency due to lack of principled exploration mechanism within them. Motivated by the success of Thompson sampling for efficient exploration in RL, we propose a novel model-free RL algorithm, Langevin Soft Actor Critic (LSAC), which prioritizes enhancing critic learning through uncertainty estimation over policy optimization. LSAC employs three key innovations: approximate Thompson sampling through distributional Langevin Monte Carlo (LMC) based $Q$ updates, parallel tempering for exploring multiple modes of the posterior of the $Q$ function, and diffusion synthesized state-action samples regularized with $Q$ action gradients. Our extensive experiments demonstrate that LSAC outperforms or matches the performance of mainstream model-free RL algorithms for continuous control tasks. Notably, LSAC marks the first successful application of an LMC based Thompson sampling in continuous control tasks with continuous action spaces.