Deva-3 Access

For warehouse robots, breaking a glass bottle is expensive. DEVA-3 allows robots to "simulate" a grasp in their head before moving a muscle. If the simulation shows the object slipping, the robot adjusts its grip pressure. This reduces real-world trial-and-error by 90%.

The model hallucinated cars sliding, pedestrians walking cautiously, and brake lights flashing. It had never seen snow, but it had learned friction and low-traction behavior from dry roads. It generalized the concept of slipperiness. deva-3

It is called .

Current AVs rely on "predictive models" that assume other drivers are rational. DEVA-3 simulates irrational behavior. It can predict the "jerk" who cuts across three lanes without a blinker because it has seen that episode 10,000 times in training data. Wayve and Ghost Autonomy are rumored to be testing DEVA-3 variants on public roads in London right now. For warehouse robots, breaking a glass bottle is expensive

We have tried rule-based systems (they break in the real world), end-to-end deep learning (they hallucinate), and large language models (they lack physics). But a new architecture is emerging from the labs that might finally crack the code. This reduces real-world trial-and-error by 90%

They trained DEVA-3 on nothing but dashcam footage from Phoenix, Arizona. Then, they gave it a single frame from a snowy street in Oslo—something it had never seen.

For the last decade, the holy grail of robotics and autonomous driving has been a simple question: How do we teach machines to predict the future?