Amazon FAR's DreamFAR: a world model trained from scratch on robot data only, 1,000+ tasks
_krishna_murthy · x · 2026-10-30
Amazon FAR researchers introduced DreamFAR, a world model trained from scratch purely on robot-native data without internet video pretraining. They argue the era of finetuning large pretrained video models as robot world models is over. DreamFAR supports real-time teleoperation, minutes-long rollouts, and generalizes across 1,000+ tasks.
More from Embodied
- VisualForce: robots infer grip force from soft gripper deformation, no force sensors needed — du_yilun · 2026-10-30
- HALO (ICML 2026 Spotlight): certified-stable RL keeps humanoids coordinated with unpredictable humans — Scobleizer · 2026-10-30
- Humanoid cleaner robots now bookable on recurring subscription, 3 hours per visit — Scobleizer · 2026-10-30
- Neuralink lets ALS patient Kenneth call his mom independently for the first time — XFreeze · 2026-10-30
- Medusa uses nine synced iPhone 18 Pros and four Mac Studios to build reframable 3D scenes — SumitGup · 2026-10-30
- Proxie and Hippocratic AI hackathon: the patient fall detector worked too well, in unexpected ways — BradPorter_ · 2026-10-30