Odyssey is an AI lab developing general-purpose world models: causal, multimodal systems designed to learn how the world works and predict what may happen next. Its research focuses on representing physical properties, motion, cause and effect, and the behavior of agents. The company aims to provide a shared foundation for AI systems to learn through experience and for people to explore ideas in simulated environments, with potential applications across robotics, autonomous driving, science, healthcare, education, gaming, energy, and other fields.

Odyssey’s products include Odyssey-3, a foundation world model focused on physical accuracy; Agora-2, a real-time simulation where multiple human or AI participants can interact; and Starchild-1, which explores learning through richer multimodal interaction. Its PROWL-2 framework supports reinforcement learning by allowing teams of agents and their world model to learn through separate curricula in one training loop. Together, these technologies are intended to support robots, vehicles, drones, and game-playing agents, as well as training and evaluating AI in interactive environments. The company also shares research presentations and technical reports.