πR^2 Enables Real-Time Reactive Flow Policies for Robotics
July 27, 2026
πR^2 overcomes the high latency of action-chunking flow policies by using a per-position noise schedule based on diffusion forcing. This allows large-backbone policies to react to mid-execution sensory input for closed-loop control without sacrificing multi-modal expressivity.
HOW THIS AFFECTS YOU
●
builderYou can implement reactive, low-latency robot control using large pretrained backbones.
●
researcherThis approach bridges the gap between open-loop action chunking and closed-loop reactivity.