Decart Launches Oasis 3: A Photorealistic World Model to Bridge Sim-to-Real Gap

Decart, a frontier AI analysis lab, has launched Oasis 3, a world model designed to bridge the simulation-to-real gap in robotics training. The robotics industry faces a significant challenge: the lack of high-quality training data. Traditional approaches to VLA model training include teleoperation data collection (expensive), web video data (messy), and synthetic data (limited by physics engines). Oasis 3 addresses these limitations by combining photorealistic, interactive motion graphics with a powerful physics engine. Key features: runs at 22fps on CoreWeave infrastructure, achieves less than 200ms latency, and supports native three-camera view for spatial consistency. The model was co-designed with Nvidia's physical AI ecosystem.