WorldDiT unifies robot action generation and future frame prediction in one diffusion model
bageldotcom · hf · 2026-07-28
WorldDiT proposes a unified diffusion transformer for robot world and action modeling
Bagel Labs introduces WorldDiT, a single diffusion transformer that jointly learns continuous action chunks and future visual patches, instead of relying on a large pretrained VLM as the action backbone.
Key results:
- Evaluated on four LIBERO simulation suites
- Reported to sit on the Pareto frontier for total parameters vs. mean success among methods that report all four suites
- Positioned as a sub-billion-parameter baseline for future scaling studies
Related event: Bagel Labs Unveils WorldDiT for Unified Robot Control(3 posts)→
More from Embodied
- Sentdex trains a fall-recovery policy for the Asimov humanoid: getting up is hard — Sentdex · 2026-10-16
- LeRobot adds VLM-powered feature that turns plain-language goals into robot skills — Gradio · 2026-10-16
- Optimus Robot Shows Off Its Driving in a Tesla Roadster — debreuil · 2026-10-16
- Qdrant Edge gives the Reachy Mini robot local persistent semantic memory — tomaarsen · 2026-10-16
- Tesla Roadster Thruster Teardown: SpaceX COPV Tank Plus Optimus Actuator — XFreeze · 2026-10-16
- Critic Calls Figure's Terminator-Style Robot Demo Publicity Over Substance — FuSheng_0306 · 2026-10-16