Odyssey-3 Tested: Frozen Backbone Trained on 20 Hours of Driving Data Transfers
testingcatalog · x · 2026-10-09
testingcatalog breaks down Odyssey-3, drawing a parallel between physics in world models and coding in language models: a model that accurately predicts falling, pouring and colliding objects shows an understanding of the world that robots and agents need.
On results: robot arms learned tasks from tens of hours of demonstrations; a driving policy trained on 20 hours of driving data kept the Odyssey-3 backbone frozen; Flexion's humanoid policies kept working under lighting changes that broke the VLA baselines Odyssey tested. Physical AI teams can contact Odyssey to build on the model.
More from Embodied
- ENPIRE turns coding agents into autonomous robotics researchers, hitting 99% success on dexterous tasks — chris_j_paxton · 2026-10-09
- IRVL Researcher to Present UHAS, HRT1 and VLA-Replica at Texas A&M Robotics Seminar — YuXiang_IRVL · 2026-10-09
- Google Gemma-Backed Hardware Hackathon at AGI House Gathers 120 Builders for On-Device AI — agihouse_org · 2026-10-09
- Odyssey launches Odyssey-3, claims SOTA world model on Physics-IQ benchmark — Scobleizer · 2026-10-09
- Figure valued at $39B: tracking Brett Adcock's promise of 100,000 robots in 4 years — annatonger · 2026-10-09
- Sentdex: a raw multimodal LLM drives a robot with zero training, VLA era may be over — Sentdex · 2026-10-09