RL Post-Training Finally Works for Diffusion Navigation Policies — and the Trick Is Reweighting, Not Gradients

RL Post-Training Finally Works for Diffusion Navigation Policies — and the Trick Is Reweighting, Not Gradients | BedrockNews