SALE-Based Offline Reinforcement Learning with Ensemble Q-Networks
This is an incremental improvement for offline RL practitioners, addressing stability and performance in model-free actor-critic algorithms.
The paper tackles the challenge of out-of-distribution actions in offline reinforcement learning by integrating ensemble Q-networks and a gradient diversity penalty, achieving higher convergence speed, stability, and performance on D4RL MuJoCo benchmarks compared to existing methods.
In this work, we build upon the offline reinforcement learning algorithm TD7, which incorporates State-Action Learned Embeddings (SALE) and a prioritized experience replay buffer (LAP). We propose a model-free actor-critic algorithm that integrates ensemble Q-networks and a gradient diversity penalty from EDAC. The ensemble Q-networks introduce penalties to guide the actor network toward in-distribution actions, effectively addressing the challenge of out-of-distribution actions. Meanwhile, the gradient diversity penalty encourages diverse Q-value gradients, further suppressing overestimation for out-of-distribution actions. Additionally, our method retains an adjustable behavior cloning (BC) term that directs the actor network toward dataset actions during early training stages, while gradually reducing its influence as the precision of the Q-ensemble improves. These enhancements work synergistically to improve the stability and precision of the training. Experimental results on the D4RL MuJoCo benchmarks demonstrate that our algorithm achieves higher convergence speed, stability, and performance compared to existing methods.