FlowAWR: Online Adaptive Flow Reinforcement via Advantage-Weighted Rectification
FlowAWR reframes continuous generative policy optimization as supervised regression toward an optimal velocity field, eliminating the need for stochastic SDE samplers and Classifier-Free Guidance (CFG). It achieves this by deriving a magnitude-aware, advantage-weighted rectification form of the optimal field. This method provides SDE-free optimization that is faster and more stable than existing implicit frameworks like DiffusionNFT.