Odyssey unveils Odyssey-3, a versatile world model for AI systems

Odyssey: Odyssey is an AI research lab pioneering foundation world models that enable simulation and understanding of physical environments across robotics, autonomous systems, and virtual worlds. Founded by self-driving technology veterans, the lab focuses on models that capture physics, dynamics, and cause-and-effect for broad applicability. In this news, Odyssey announced Odyssey-3 as its latest advancement, a single model designed to power multiple physical AI systems without requiring separate specialized AIs for each domain. Odyssey-3: Odyssey-3 is a general-purpose foundation world model built as an autoregressive diffusion transformer trained on diverse visual observations to develop an internal understanding of the physical world. It supports control of robot arms, humanoids, vehicles, drones, AI agents, and video game systems from the same pretrained base, adapting to new tasks with limited additional data. The model is directly featured in the news as the newly unveiled system demonstrating capabilities like learning autonomous driving from simulated data and enabling physical agents. Release Timeline: Odyssey plans to make Odyssey-3 publicly available in the coming weeks following its announcement. Model Architecture: Odyssey-3 uses an autoregressive diffusion transformer trained to simulate diverse scenarios and translate internal representations into system-specific actions via a learned decoder. Development Approach: The model emphasizes a shared understanding of physics and cause-and-effect across environments, moving away from building separate AI systems for individual machines or tasks.