AI RESEARCH
WarmPrior: Straightening Flow-Matching Policies with Temporal Priors
arXiv CS.LG
•
ArXi:2605.13959v1 Announce Type: new Generative policies based on diffusion and flow matching have become a dominant paradigm for visuomotor robotic control. We show that replacing the standard Gaussian source distribution with WarmPrior, a simple temporally grounded prior constructed from readily available recent action history, consistently improves success rates on robotic manipulation tasks. We trace this gain to markedly straighter probability paths, echoing the effect of optimal-transport couplings in Rectified Flow.