# ETH Zurich Unveils 'Fingers as Legs': An Anthropomorphic Robotic Hand Capable of Autonomous Walking and Manipulation

Source: TechNewsList (https://technewslist.com)
Canonical URL: https://technewslist.com/en/article/eth-zurich-develops-fingers-as-legs-autonomous-walking-robotic-hand-2026-09-20-m
Section: Drones & Robots (https://technewslist.com/en/drones-robotics)
Author: TechNewsList
Language: en
Published: 2026-09-20T09:57:11.591+00:00
Updated: 2026-09-20T09:57:11.747011+00:00

> Researchers at ETH Zurich's Soft Robotics Lab demonstrate an untethered, 818-gram robotic hand that uses reinforcement learning to repurpose its 20 powered finger joints as walking legs across uneven terrain.

## TL;DR
- ETH Zurich developed an untethered 818-gram robotic hand capable of walking on its fingers and manipulating objects.
- The hand features 20 independently powered joints controlled by an onboard Raspberry Pi Zero 2 W processor.
- Researchers used simulation-to-real reinforcement learning to train stable gaits and dynamic fall recovery.
- The innovation enables dual-use robotic appendages that combine locomotion and fine dexterity in compact form factors.

## Key points
- The system reconfigures anthropomorphic fingers as multilegged walking appendages without adding dedicated wheels or limbs.
- An onboard lithium-polymer battery and microcontroller enable completely autonomous, cable-free locomotion.
- The neural locomotion policy allows the hand to walk forward, steer omnidirectionally, and climb over small obstacles.
- When overturned, the hand executes autonomous righting maneuvers by coordinating high-torque finger extensions.
- Potential applications include confined industrial pipeline inspection, disaster search-and-rescue, and aerospace exploration.

## What happened

Robotics researchers at ETH Zurich's renowned Soft Robotics Lab unveiled a remarkable engineering breakthrough on September 19, 2026, demonstrating an untethered, fully anthropomorphic robotic hand capable of walking across complex surfaces using its fingers as legs. Detailed in a peer-reviewed research preprint and accompanied by physical hardware demonstrations, the project—titled "Fingers as Legs: Autonomous Locomotion and Manipulation with an Anthropomorphic Hand"—reimagines the mechanical role of robotic end-effectors, transforming a delicate grasping tool into an independent, self-contained mobile exploration agent.

For decades, robotics engineering has maintained a strict architectural divide between locomotion systems and manipulation tools. Mobile robots utilize dedicated wheeled platforms, continuous tracks, or specialized quadrupedal limbs to travel through environments, mounting separate articulated arms and grippers to manipulate objects once they arrive at a target location. This division adds substantial weight, mechanical complexity, and power consumption, limiting robot mobility within tightly confined spaces.

The ETH Zurich research team discarded this traditional design separation by designing a 20-joint anthropomorphic robotic hand weighing just 818 grams that integrates all required power, actuation, and compute electronics directly inside its palm structure. Operating without any external tether cables, the hand can crawl across floors, navigate uneven terrain, recover from physical falls, and seamlessly transition into grasping tools and pushing buttons.

![Compact servo-electric five-finger gripping assembly demonstrating fine dexterity and independent finger joint motor control](https://rkhynbcsbnkkcwgexzwg.supabase.co/storage/v1/object/public/media/api/1789898222349-ynineb-eth-zurich-develops-fingers-as-legs-autonomous-walking-robotic-hand-2026-09-20-m-inside-1-e9eaa7e35a.webp)
*Anthropomorphic joint actuation: Multi-degree-of-freedom finger mechanisms serve as both locomotion legs and precision grasp tools.*

## Why it matters

The development of multi-modal, dual-purpose appendages addresses one of the most persistent bottlenecks in field robotics: morphological versatility under tight weight constraints. In applications ranging from search-and-rescue within collapsed earthquake rubble to internal inspection of jet engines and offshore petrochemical conduits, conventional mobile robots are frequently too bulky to penetrate narrow access points.

A robotic hand that can detach from an industrial robotic arm, crawl autonomously through a pipe or crawlspace, inspect a valve, manipulate a mechanical switch, and return to dock represents a massive leap in operational capability. By eliminating the necessity for heavy, dedicated walking chassis, robotic systems can achieve unprecedented payload efficiency and access previously unreachable physical domains.

Beyond immediate industrial utility, the research advances the frontier of bio-inspired morphological robotics. Biological hands evolved primarily for manipulation, but biological organisms frequently repurpose limbs for alternative locomotion tasks when under environmental pressure. By demonstrating that neural reinforcement learning policies can unlock unexpected mobility modes from existing kinematic structures, the study provides a roadmap for designing more adaptable, general-purpose robotic hardware.

## Technical details

The physical construction of the walking hand features a highly integrated mechanical layout optimized for strength-to-weight performance. The hand incorporates 20 active degrees of freedom, driven by compact coreless DC servo motors routed through miniature low-friction tendon cables. The palm houses an onboard Raspberry Pi Zero 2 W single-board computer running Linux, an inertial measurement unit (IMU), motor driver arrays, and a rechargeable lithium-polymer battery providing up to 25 minutes of continuous untethered crawling.

To overcome the immense control challenge of coordinating 20 high-dimensional joints during dynamic ground contact, the researchers employed simulation-to-real (sim-to-real) reinforcement learning within the Isaac Gym physics simulation environment. The neural network controller was trained across millions of parallelized physics simulations, exposing the agent to varied ground frictions, payload masses, and external lateral disturbances.

![Industrial articulated robotic manipulator arm exhibiting wrist mount integration for specialized end-effector modules](https://rkhynbcsbnkkcwgexzwg.supabase.co/storage/v1/object/public/media/api/1789898224255-z4orrw-eth-zurich-develops-fingers-as-legs-autonomous-walking-robotic-hand-2026-09-20-m-inside-2-6972cea273.webp)
*Detachable end-effectors: Hand mechanisms capable of self-locomotion can detach from manipulator arms to inspect confined pipes.*

The resulting locomotion policy generates a stable, multi-beat walking gait where four fingers act as supportive walking legs while the thumb provides lateral balance and directional steering. The policy dynamically compensates for irregular ground obstacles up to 20 millimeters in height. Furthermore, the controller features an autonomous righting reflex: if the hand is flipped onto its back by an external collision, it coordinates a simultaneous extension of all five fingers against the floor, executing a dynamic somersault that flips the palm upright onto its walking tips.

## Market / industry impact

The breakthrough from ETH Zurich is drawing intense interest from industrial automation suppliers, defense research agencies, and space exploration programs. Commercial robotics manufacturers have spent recent years striving to commercialize dexterous hands for humanoid warehouse robots, but high costs and delicate mechanical linkages have slowed deployment. Proving that an anthropomorphic hand can endure the high-impact stresses of physical ground locomotion underscores the mechanical durability of modern tendon-driven architectures.

In aerospace and planetary exploration, space agencies like NASA and the European Space Agency are actively investigating miniature crawling robots for planetary surface exploration and spacecraft maintenance. An astronaut assistance robot equipped with detachable crawling hands could deploy secondary exploration scouts into lunar lava tubes or narrow satellite service panels where full-size rovers cannot fit.

The research also impacts the educational and open-source robotics communities. By utilizing affordable microcontrollers like the Raspberry Pi Zero alongside 3D-printable structural components, the ETH Zurich platform demonstrates that cutting-edge reinforcement learning locomotion does not require multi-million-dollar laboratory rigs, democratizing advanced quadrupedal and multi-limb locomotion research for universities worldwide.

## What to watch next

Following the initial publication, the ETH Zurich research team plans to integrate miniature tactile optical sensors into the finger pads. High-resolution tactile feedback will allow the neural controller to sense surface textures and edge boundaries in real time, enhancing the hand's ability to traverse slick metallic surfaces and climb vertical mesh barriers.

Observers will also watch for experimental demonstrations involving coordinated multi-hand cooperative locomotion. Researchers have proposed linking two walking hands together via magnetic wrist latches, allowing them to collaborate in carrying heavier payloads or navigating steep inclines before unlatching to manipulate dual controls.

Finally, the robotics sector will track commercial licensing discussions. Several European industrial robotics manufacturers are in talks with ETH Zurich's technology transfer office to evaluate whether the crawling end-effector concept can be adapted into commercial pipeline inspection tools for the oil, gas, and municipal water utilities sectors.

## Sources

- [ETH Zurich Research Preprint on arXiv](https://arxiv.org/abs/2609.09841) — Full scientific publication detailing 20-joint kinematic modeling, simulation-to-real reinforcement learning, and untethered gait metrics.

- [ETH Zurich Soft Robotics Lab Project Overview](https://srl.ethz.ch/research/fingers-as-legs-locomotion.html) — Laboratory project overview featuring locomotion videos, fall recovery trajectories, and onboard computing hardware specifications.

- [Hackaday Robotics Engineering Feature](https://hackaday.com/2026/09/19/eth-zurich-robotic-hand-walks-on-its-fingers) — Independent robotics analysis assessing mechanical tendon routings, 818g weight balance, and dual manipulation-locomotion versatility.

Mentions: ETH Zurich, Soft Robotics Lab, Reinforcement Learning, Robotics, Raspberry Pi

## Sources
- [ETH Zurich Research Preprint on arXiv](https://arxiv.org/abs/2609.09841)
- [ETH Zurich Soft Robotics Lab Project Overview](https://srl.ethz.ch/research/fingers-as-legs-locomotion.html)
- [Hackaday Robotics Engineering Feature](https://hackaday.com/2026/09/19/eth-zurich-robotic-hand-walks-on-its-fingers)