Robotic Hand Achieves Human-Like Grasp Through Adaptive Compliance

Researchers at EPFL's CREATE Lab have developed a robotic hand that can replicate human-like grasps without precise spatial data, using a principle known as compliance. As Kai Junge, a PhD student in the School of Engineering's Computational Robot Design and Fabrication (CREATE) Lab, explains, human hands grasp objects with minimal external guidance, thanks to the soft, compliant interactions at the interface between the hand and the object. This principle inspired the design of the lab's ADAPT hand (Adaptive Dexterous Anthropomorphic Programmable sTiffness), which leverages compliant materials like silicone and spring-loaded joints to perform self-organized grasps.

The ADAPT hand, with just 12 motors controlling its 20 joints, relies on springs and a silicone 'skin' to adjust its grip, minimizing the need for precise positioning. This design allows the hand to pick up a wide range of objects with a 93% success rate in trials, achieving a 68% similarity to natural human grasping. Unlike conventional robotic hands that depend on detailed programming for each joint, the ADAPT system uses an 'open loop' approach, moving through only four general waypoints without constant feedback. This distributed compliance enables it to grasp items as varied as bolts and bananas without additional programming.

The research team is now exploring ways to enhance the ADAPT hand's performance by integrating closed-loop control elements, including pressure sensors and AI, to improve its adaptability to uncertain environments. This hybrid approach aims to combine the flexibility of compliant mechanisms with the precision of traditional robotic systems, potentially transforming robotic interactions in complex, human-centric settings.

Research Report:Spatially distributed biomimetic compliance enables robust anthropomorphic robotic manipulation