Expert Behavior Prior Reinforcement Learning via Q-CVAE

The Expert Behavior Prior (EBP) algorithm improves online reinforcement learning sample efficiency by using a Q-guided conditional variational autoencoder (Q-CVAE). This method generates high-value expert policy priors directly from online replay buffers to overcome the limitations of static, low-diversity offline datasets.