This work adds one feature-alignment term to ordinary VLA training: a frozen world model is run over the training frames once and cached, and the student learns to agree with that cache, indicating a broad representational prior rather than a fragile alignment between two particular networks.
Abstract
Vision-Language-Action (VLA) models map observations to actions with no objective that accounts for how the world responds, so their robustness is bounded primarily by data coverage. World models carry precisely that missing objective and are better grounded for it, yet rolling the future forward costs seconds per decision and rules them out of the control loop. We show the two can be separated. What a world model knows about physical scenes lives in its internal features; generating the future is merely the objective that produced them, so the grounding can be inherited while the generative machinery is left behind. We add one feature-alignment term to ordinary VLA training: a frozen world model is run over the training frames once and cached, and the student learns to agree with that cache. No teacher is loaded during training, the projector is discarded after it, and the deployed policy is identical to the undistilled baseline, running in 32ms and 1.86GB on a consumer RTX5090, so every gain is attributable to the representation rather than to added capacity or test-time compute. A 0.8B student reaches 97.9% on LIBERO, improves from 48.2% to 50.5% on RoboCasa-GR1 humanoid manipulation, and the same objective carries over to real hardware, on both a single-arm and a bimanual platform. The gain survives changes of student scale, backbone, alignment layer, and teacher, indicating a broad representational prior rather than a fragile alignment between two particular networks. Project page: https://thaw-vla.trung-dt.com/.
Specifying a goal in language rather than as a goal frame is a natural interface for planning with a latent world model, but testing it needs scenes in which language must discriminate between several objects. We build SLIM, a pushing benchmark with several small objects and paired visual and language goals on identica...
Florian Strohm, Patrick Wagner, Jannik Schwab et al.· 0 citations
A pretrained video world model admits many plausible futures for a scene, but a robot must realize the exact task-conditioned one. To turn world models into executable robot policies, existing methods fine-tune the heavy world model backbone using large-scale robot data and computational resources. Challenging this sta...
Bang Du, Yi-Chen Xie, Shu-Qi Zhao et al.· 0 citations
AWM-VLA is presented, a unified framework that embeds aligned world modeling directly inside a diffusion-transformer policy and is compatible with any diffusion or flow-matching policy, making aligned world modeling an inexpensive, broadly applicable component of generalist manipulation.
WorldToken, a time-first policy instantiation that fuses multiview images, proprioception, and task conditioning within each policy timestep into one world token is introduced and its data-scaling and temporal-context behavior under the tested recipes are characterized.
Chunkai Yang, An-Dong Yang, Di Huang et al.· 0 citations
FIND is introduced, an agentic real-world RL framework that closes the loop between scene understanding, weakness-aware practice, self-evaluation, and policy improvement in a persistent workspace and reframes autonomous practice as a scene-conditioned, performance-aware task-selection problem.
Token-World is introduced, an action-conditioned world model that compresses VLM features into a compact token state, learns future dynamics in this reduced space, and maps predicted states back to the original policy-facing representation for downstream use.
Chu-Yao Fu, Xiao-Wei Chi, Yu-Han Rui et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.