Odyssey

Odyssey is an AI lab that develops general world models. It provides a foundation world model designed to power physical and virtual systems, including robots, cars, drones, and video games. The model aims to enable intelligent agents to interact with and understand the world through learning physics, dynamics, and human behaviors. The work targets applications across robotics, autonomous driving, science, healthcare, education, gaming, defense, and more, presenting a common foundation for multiple domains.

Odyssey-3 is presented as a foundation world model that can operate across many machines and tasks. It describes a single learned intelligence that can interface with physical and virtual systems, adapting its knowledge to new tasks with relatively little task-specific experience. The goal is to provide a broad, reusable representation of world knowledge that supports both real-world control and simulated environments.

The toolkit demonstrates capabilities in controlling robots and humanoids, driving vehicles, piloting drones, training AIs, and even playing video games. It emphasizes learning from large-scale visual observations and using a fixed backbone representations to inform task-specific policies. The work also discusses collaborations with industry partners to evaluate transfer, robustness, and generalization across bodies, viewpoints, and controls.

Odyssey-3 is framed as a base model for physical intelligence, powering both real and simulated systems, with examples in robotics, autonomous driving, humanoid control, drone navigation, and game-playing. It highlights a shared world model underpinning multiple tasks and environments, and mentions ongoing collaborations to assess performance across different robots and settings.

Use cases

• Control robots and humanoids in real-world tasks
• Autonomous driving in simulated or real settings
• Aerial navigation for drones