We're pioneering world models: causal, multimodal systems that learn to understand and simulate the world
A new, foundational technology
World models learn the physics, dynamics, cause-and-effect and human behaviors that shape our reality, building an understanding of how the world works. That understanding provides a shared foundation for robots to perform useful work, AIs to learn through experience, and humans to explore, create and test ideas within simulated worlds
How does a world model learn?
A world model learns to predict the most likely future
Physics
The properties and constraints that govern how objects behave
Dynamics
How objects and systems move, interact, and change over time
Cause & Effect
How actions and events influence what happens next in the world
Agent Behaviors
How agents act, pursue goals, and interact with others
Our team brings experience from
DeepMind
Waymo
Tesla
Meta
Apple
XAI
We’re backed by
Odyssey-3 · Foundation World Model
Odyssey-3 powers a variety of physical and virtual systems
Odyssey-3 draws on a learned understanding of physics, motion, and cause-and-effect to act in physical and virtual worlds. One foundation world model powers policies for robots, humanoids, cars, drones, video games, and more
Learn from the World
Broad world experience builds knowledge of motion, physics, and how interactions with the world shape what happens next
Learn from Experience
Generate interactive environments where humans and AI can take actions and experience their consequences
One Shared Foundation
Power robots, humanoids, cars, drones, and game-playing agents with the same underlying world model
Multi-Agent World Models
Multi-agent worlds, where agents learn through interaction
We’re building multi-agent world models that enable humans and AI to interact inside the same simulated environment, exploring how agents learn through shared experience, and how those interactions can improve the worlds themselves
World Model
Agora-1
A multi-agent world model, enabling multiple participants—human or AI—to share and interact within the same world simulation in real-time
Reinforcement Learning
PROWL-1
A novel RL-driven adversarial framework where an RL agent explores game environments with the objective to improve world model performance
