PDRVLA Uses Physical Dynamics Tokens for Robot Control
October 26, 2026
PDRVLA introduces a Physical Dynamics Module that extracts physical tokens from vision-language features to condition action generation. This avoids the high latency of online video world models by predicting RGB, depth, and optical-flow features directly.
HOW THIS AFFECTS YOU
●
builderYou can implement lower-latency VLA control loops by using physical tokens instead of full video generation.
●
researcherThis demonstrates an effective way to incorporate embodied chain-of-thought and physical dynamics into VLA training.