Physical AI Gets a New Sandbox: Decart’s Oasis 3 Simulates the Real World
Decart has unveiled Oasis 3, a real-time world model capable of simulating photorealistic driving environments. This breakthrough allows developers to test autonomous systems in high-fidelity virtual worlds that react dynamically to AI inputs.
The boundary between simulation and reality is thinning with the release of Oasis 3 by Decart. This new "world model" is designed to generate hours of photorealistic driving footage in real-time, providing a sandbox for Physical AI to interact with complex environments without the risks associated with public roads.
Unlike traditional simulators that rely on pre-built 3D assets and rigid physics engines, Oasis 3 uses generative AI to predict and render the next frame of a driving sequence based on user or AI control inputs. This allows for an infinitely variable range of scenarios, from treacherous weather conditions to unpredictable pedestrian behavior. By training on vast datasets of real-world driving, the model captures the nuances of lighting, reflection, and motion that often baffle less sophisticated vision systems.
For developers of Physical AI, the implications are significant. The ability to run "edge case" scenarios—such as a child darting into a highway or a sudden mechanical failure—thousands of times in a photorealistic virtual environment accelerates the safety validation process. However, Decart notes that while the visual fidelity is high, the model still faces challenges with long-term temporal consistency across many hours of simulation. Despite these caveats, Oasis 3 represents a major step toward autonomous systems that can "dream" and learn from imaginary experiences before ever touching asphalt.
Source: TechCrunch
