TMTPOST — Autonomous driving and robotics startup CooWA AI launched its CooWAIM 2.0 interactive world model to optimize multi-form robotic navigation on Wednesday. The model utilizes a proprietary DAWN architecture that jointly trains a world predictor and an action denoiser within a latent space. This design forms a bidirectional closed loop where world hypotheses guide action generation, which concurrently reshapes world deductions. The integration leverages efficient feature compression to reduce the overall required computational workload by approximately 75 percent. CooWA AI refined the model using real-world data from a fleet of 10,000 robots deployed across more than 50 cities. The technology has accumulated over 45 million kilometers of operational mileage and fully adapts to wheeled chassis, quadruped robotic dogs, and humanoid robotics configurations.