Retimed Bellman Flows: Escaping the Impossible Triangle of Velocity Bootstrapping
ReBF constructs a provably conditionally unbiased velocity target that preserves the Bellman fixed point and contracts under Wasserstein distances and outperforms existing flow critics across 38 challenging OGBench and D4RL offline reinforcement learning tasks.
Bo-Yang Xu, Sheng-Zhe Chen, Hao Yan
· 0 citations