Dexterity introduces Foresight, a world model trained on 100M+ production robot actions that plans Mech's box placements

Dexterity/Mech

Dexterity introduced Foresight, a world model trained on over 100 million autonomous robot actions in real warehouse operations. It enables Mech to predict and weigh box placements for dense, stable loads before execution, making placement decisions in under 400 milliseconds while balancing density, stability, and reachability.

• World model trained on 100M+ real warehouse actions (not simulation); predicts physical outcomes of potential placements.
• Decision-making: predictive branching, pragmatic adequacy, capability-aware orchestration, predictive pipelining.
• 400 ms decision cycle optimizes multiple factors: load density, stability, robot reachability, execution feasibility.
• Enables warehouse robots to reason about load planning with nuance previously requiring human expertise.