What Makes Value Learning Efficient in Residual Reinforcement Learning?

Residual reinforcement learning (RL) enables stable online refinement of expressive pretrained policies by freezing the base and learning only bounded corrections. However, value learning in residual RL poses unique challenges that remain poorly understood. In this work, we identify two key bottlenecks: cold start pathology, where the critic lacks knowledge of the value landscape around the base policy, and structural scale mismatch, where the residual contribution is dwarfed by the base action. Through systematic investigation, we uncover the mechanisms underlying these bottlenecks, revealing that simple yet principled solutions suffice: base-policy transitions serve as an essential value anchor for implicit warmup, and critic normalization effectively restores representation sensitivity for discerning value differences. Based on these insights, we propose DAWN (Data-Anchored Warmup and Normalization), a minimal approach targeting efficient value learning in residual RL. By addressing these bottlenecks, DAWN demonstrates substantial efficiency gains across diverse benchmarks, policy architectures, and observation modalities.

Modern robots are often controlled by AI trained from demonstrations, but they still stumble in situations slightly different from what they were taught. A safer fix than retraining everything is to keep the original controller fixed and train a small "corrector" by trial and error. Yet this corrector learns frustratingly slowly, often needing millions of attempts. We traced the slowness to the robot's internal "judge" that scores how good an action is. It struggles because at first it knows nothing about the original controller, and because the corrections are deliberately tiny, their effect is drowned out and the judge barely notices them. We show that two simple remedies fix both problems and make learning about five times faster, bringing reliable, self-improving robots meaningfully closer to real-world use.