Odyssey, an AI research company developing foundation world models, has announced Odyssey-3, which it describes as its most powerful world model to date. World models learn from large volumes of visual observations of the world evolving, predicting what happens next and how actions affect outcomes. Odyssey-3 is positioned as a pretrained foundation that gives physical agents a shared understanding of physics, dynamics, cause and effect, and human behavior before they are taught a specific task, so systems can adapt to new machines, environments and tasks with relatively little additional experience.
Key takeaways
- Robotics: controls multiple robot arms and demonstrated recovery behaviors it was not explicitly trained to perform.
- Humanoids: powers control policies developed with Flexion that perform tasks in real time and generalize to environmental changes better than the baselines Odyssey tested.
- Vehicles: drove autonomously on roads in India using a policy trained with approximately 20 hours of simulated driving data.
- Drones: powers aerial navigation policies trained with simulated data to navigate indoor environments and avoid obstacles.
- AI training and games: generates interactive environments for agents to act and learn from consequences, and plays Grand Theft Auto V with early transfer of learned behaviors into other games without additional policy training.
Odyssey-3 is designed as a foundation spanning domains rather than a model optimized for a single machine or application. For manufacturing automation, the emphasis on cross-machine transfer and simulated-data training points to shorter policy development cycles and broader reuse of learned skills across robot arms, mobile platforms and aerial systems.
Source: Robotics & Automation News · Published 2026-10-08T19:22:04 · “Odyssey launches world model to control robots, humanoids, vehicles and drones”