Odyssey-3: one frozen world model drives humanoids, cars, drones, and games
thione · x · 2026-09-21
Odyssey, the Palo Alto lab founded by Oliver Cameron and Jeff Hawke, unveiled Odyssey-3: a single frozen foundation world model that controls robot arms, humanoids, autonomous vehicles, drones, and game characters via lightweight decoders trained on top.
Key numbers:
- Driving policies trained entirely in simulation retained 77% of real-data-trained performance on unstructured Indian roads, measured by distance between safety-driver interventions;
- The backbone stayed frozen; the driving decoder trained on just 20 hours of simulated data;
- Academic benchmarks typically show 24–30% drops for sim-to-real transfer, so 77% retention would be an architectural advance if independently verified.
Caveat: independent evaluation remains pending.
Related event: Odyssey-3: One Frozen World Model for Robots, Cars, Drones and Games(2 posts)→
More from Embodied
- Apple wins the AI infra lottery: M5 Ultra packs 512GB unified memory for local AI — Hesamation · 2026-09-21
- DeliveryGym: adaptive UE5 RL environment boosts Qwen3-VL-4B delivery earnings 54.3% — Lianhuiq · 2026-09-21
- Train a balance bot with PPO: Reinforcement Learning for Robotics Part 2 — ShawnHymel · 2026-09-21
- Google's $899 Googlebook bets you'll buy a new laptop just for Gemini — TechCrunch AI · 2026-09-21
- Archtyp AI CEO: Robots must be predictable in ways humans are not — xmercury_one · 2026-09-21
- Unitree's Dex5 hand packs 22 DOF at $6.5k, undercutting rivals by up to 3x — RevolutionaryJob2409 · 2026-09-21