Scientists at ETH Zurich have demonstrated how an anthropomorphic robotic hand can use its fingers not just for manipulation but in addition for locomotion, enabling the hand to crawl, get better from falls and transfer round its atmosphere with no separate mobility system.
The analysis used an off-the-shelf robotic hand from WUJI, a developer of dexterous robotic arms and teleoperation expertise.
Slightly than mounting the hand on a traditional robotic arm or cellular platform, the researchers skilled it to assist its personal weight and transfer utilizing the identical fingers usually supposed for greedy and manipulation.
The ensuing system was able to untethered crawling, steering and recovering after falls. It might additionally carry out manipulation duties whereas supporting itself, together with urgent successive keys on a keyboard and pushing an object towards specified targets.
The researchers describe the system as a “compact cellular manipulator” that eliminates the necessity for a separate locomotion mechanism.
The work was introduced in a paper titled “Fingers as Legs: Studying Self-Supported Locomotion and Manipulation with an Anthropomorphic Hand”, by Amirhossein Kazemipour, Hehui Zheng and Robert Katzschmann.
The researchers used reinforcement studying to develop the locomotion capabilities, bearing in mind the unequal dimensions and traits of the person fingers.
Coaching occurred in simulation, with the simulated atmosphere calibrated utilizing measurements from the bodily {hardware}.
In response to the researchers, their reward formulation allowed the hand to maneuver sooner in simulation than when utilizing tuned rewards initially developed for quadruped robots.
The bodily system integrated onboard energy and computation, permitting it to function as a self-contained platform.
For manipulation experiments, the hand carried out successive keyboard instructions with out counting on imaginative and prescient. In one other demonstration, it used overhead visible suggestions to push an object towards goal positions.
The analysis illustrates an uncommon strategy to cellular manipulation by permitting the identical mechanisms to carry out each locomotion and bodily interplay.
It additionally highlights the growing position of commercially accessible dexterous arms as analysis platforms for robotic studying.
WUJI has been creating this space additional with its WUJI Hand 2 and WUJI Glove, that are designed to assist real-time teleoperation and the gathering of movement and tactile information for robotic studying.
The corporate demonstrated the 2 programs at IROS 2026, held in Pittsburgh from September 27 to October 1.
Throughout demonstrations, the WUJI Glove was used to manage the corporate’s dexterous hand in actual time, offering a method of capturing human movement and translating it into robotic manipulation.
Such teleoperation programs are more and more getting used not merely for distant robotic management, however as data-generation instruments for coaching embodied AI and robotic manipulation fashions.
By recording human demonstrations, builders can acquire examples of how objects are grasped, repositioned and manipulated, offering coaching information for programs supposed finally to carry out these duties autonomously.
WUJI can be among the many {hardware} producers offering native machine assist for Nvidia Isaac Teleop, Nvidia’s teleoperation framework for robotics.
The mix of more and more succesful dexterous {hardware}, teleoperation and reinforcement studying is broadening the position of robotic arms past their conventional place as finish effectors hooked up to robotic arms.
The ETH Zurich work takes that concept to an uncommon excessive: as a substitute of giving a robotic a hand, the researchers successfully turned the hand itself into the robotic.
