#robotic adaptation

DSRL introduces a novel method to adapt diffusion-based robotic policies via latent-space reinforcement learning, significantly boosting real-world task performance without modifying base models.