‘Soft’ robot hands mimic human touch

The research, led by the Computational Robot Design & Fabrication (CREATE) Lab in EPFL’s School of Engineering, explores how soft, flexible materials can enable robots to perform complex tasks in uncertain environments without sophisticated programming.

Kai Junge, a PhD student in the CREATE Lab, explained the inspiration behind the project: “As humans, we don’t really need too much external information to grasp an object, and we believe that’s because of the compliant – or soft – interactions that happen at the interface between an object and a human hand,” he said. “This compliance is what we are interested in exploring for robots.”

The result of this investigation is the ADAPT hand (Adaptive Dexterous Anthropomorphic Programmable sTiffness), a robot hand built with strips of silicone, spring-loaded joints, and a bendable arm. These materials allow for ‘self-organised’ grasps – movements that arise from the mechanics of the hand itself, rather than from programmed instructions.

In practical trials, the ADAPT hand successfully picked up 24 different objects with a 93% success rate. The grasps showed a 68% similarity to natural human grasps. These findings, recently published in Nature Communications Engineering, suggest that a mechanically intelligent design can allow for effective task execution even in the absence of environmental precision.

Unlike traditional robotic hands that rely on individual motors for each joint, the ADAPT hand contains only 12 motors in its wrist to control 20 joints. Its operation depends largely on physical properties – spring stiffness and silicone coverage – to provide mechanical feedback. The robot moves through four basic positions, with additional adaptation happening autonomously, a process known in robotics as ‘open-loop’ control.

For instance, when programmed with a single motion, the hand adjusted itself to grasp items as varied as a bolt and a banana. More than 300 grasps were analysed and compared to those performed by a rigid version of the robot, showing the benefits of distributed compliance.

Junge noted the broader implications of this bottom-up approach: “Developing robots that can perform interactions or tasks that humans do automatically is a lot harder than most people expect,” he said. “That’s why we are interested in exploiting this distributed mechanical intelligence of different body parts like skin, muscles, and joints, as opposed to the top-down intelligence of the brain.”

While the aim of the ADAPT hand was not to perfectly replicate human grasping, the project has provided a clear demonstration of how far compliant materials alone can go in robotic manipulation. The EPFL team is now moving towards integrating closed-loop control elements, including pressure sensors in the silicone skin and artificial intelligence, to combine mechanical robustness with refined feedback and control.

Junge concluded: “A better understanding of the advantages of compliant robots could greatly improve the integration of robotic systems into highly unpredictable environments, or into environments designed for humans.”