GlanceWAM Enables Real-Time Asynchronous Visual Imagination for Robotics
October 5, 2026
GlanceWAM decouples visual imagination from control using a single video Diffusion Transformer backbone. By generating lookahead frames asynchronously in the background and consuming them in latent space, the model achieves a 48 ms control rate without sacrificing the task success provided by visual priors.
HOW THIS AFFECTS YOU
●
builderYou can implement real-time world-action models by offloading visual imagination to an asynchronous proposer.
●
designerThis enables more fluid, responsive robot behaviors through latent-space visual foresight.