The Superhuman-Scale Dexterity Project
Robust Robotic Hand in Optimized Tendon-Driven Design for Field-Deployment on Heavy Machinery
At the heart of the demonstration was the “GigaHand,” a tendon-driven robotic hand three times the size of a human palm. Engineered by ETH Zurich’s Soft Robotics Lab, the hand was designed to bridge the dexterity gap in heavy machinery, enabling excavators not only to lift heavy debris but also to grasp delicate and more complex objects.
During the demonstration, the GigaHand performed a series of grasping tasks, from lifting sandbags to manipulating more complex and fragile items, such as a bucket or a hose, all controlled via a glove-based teleoperation system. The hand’s musculoskeletal design, inspired by biological tendons and joints, allowed for precise and fluid movements and remarkable strength.
One impressive aspect of the robotic hand is that the system is completely self-contained, featuring the battery for the motors, and a computer with all of the required wireless and power functionalities, allowing it to be mounted on existing excavators in under two minutes. This plug-and-play design is critical for rapid deployment in disaster zones.
The current setup requires two operators, one for the excavator and one for the robotic hand. Looking ahead, they hope to develop an interface that enables more intuitive operation to streamline control into a single-user interface.
Since summer 2024, the Soft Robotic Lab has been conducting research under the Moonshot R&D program Goal 3 which aims to develop AI robots that learn and coexist with human. More specifically, the GigaHand is part of the CAFE project led by Prof. Keiji Nagatani of the University of Tsukuba, seeking to revolutionize infrastructure resilience through adaptive robotics. The idea to integrate a robotic hand to heavy machinery was first pitched by Prof. Robert Katzschmann, founder of the Soft Robotics Lab, which ultimately led to this Swiss-Japanese collaboration.
In a separate afternoon demonstration, researchers from NAIST (Nara Institute of Science & Technology) showcased an AI-driven excavator trained in virtual environments to detect and remove buried obstacles, such as pipes or fallen trees, using reinforcement learning.
Japan’s frequent natural disasters demand innovative solutions. While heavy machinery is powerful, it lacks dexterity, requiring human assistance in high-risk zones. These robotic systems aim to shift dangerous tasks away from frontline workers, allowing remote operation and reducing exposure to hazardous conditions.
