Dexterity has announced a major advancement in its Physical AI stack with Foresight, a physics-consistent world model designed for real-time industrial manipulation. Foresight powers the company's dual-armed robot, Mech, enabling autonomous truck loading through a 4D box packing agent that reasons across three spatial dimensions plus time.
Key takeaways
- Foresight makes placement decisions in under 400 milliseconds, optimizing density, stability, reachability, and dual-arm parallelism.
- The system handles up to 400 potential placements per box, managing multiple walls simultaneously—a combinatorial problem more complex than Go.
- Foresight is trained on over 100 million autonomous actions and is proven across six applications, four robot types, and five hand types.
- The agentic framework is interpretable and safety-first, providing operators visibility into decision-making.
Dexterity is also launching the Foresight API Challenge, offering up to $50,000 in prizes for student teams to build packing agents without a simulator, encouraging direct engagement with physical reasoning. This development signals a shift toward production-grade world models that perceive, decide, and act in real-world logistics environments.
Source: Robotics & Automation News (roboticsandautomationnews.com) · Published 2026-03-06 · “Dexterity says its new world model delivers ‘major leap forward’ for physical AI-powered truck loading”