UC Irvine Researchers Develop Open-source, 3D-printed Robotic Hand with Human-like Dexterity
By Brian Bell, UC Irvine
The CRAFT hand features an innovative hybrid of hard and soft materials, designed to make robotics more precise, durable and accessible.
● Precise and reliable: it can perform delicate manipulation tasks, such as picking up a raspberry without crushing it.
● Durable: a new knuckle design spreads out the force of impacts, increasing durability.
● Accessible and affordable: fully open-source and 3D-printable, the hand costs less than $600 to build.
A research team from UC Irvine and the University of Illinois Urbana-Champaign (UIUC) has developed a new robotic hand that combines rigid and soft materials to produce a design that is both precise and resilient. The CRAFT Hand is designed to make robot learning, teleoperation and dexterous manipulation more reliable, durable and accessible.
At less than $600 to build and weighing just 800 grams, the open-source hand offers enhanced durability and precision when compared with similar open-source models that cost at least twice as much.
Human-like properties created from a hybrid design
“When a robot hand hits an obstacle, one of two things happens: either the obstacle gets damaged, or the hand does,” said project advisor Unnat Jain, UC Irvine assistant professor of computer science. “Human hands avoid that problem because our muscles and tendons have some give in them — the same compliance that lets you pick up a raspberry without crushing it, or lift a quarter from a flat table without fumbling. CRAFT tries to give robots that same property: enough softness to absorb contact, enough structure to still be precise.”
Most robotic hands fall into one of two camps: rigid designs that move predictably but crack or break on impact, or soft designs that handle contact better but lose their shape under load, making precise motion harder to control.
Rather than choosing one or the other, the team developed a hybrid design, using polylactic acid (PLA) — the same plastic used in desktop 3D printers — to print the finger segments and thermoplastic polyurethane (TPU), a rubber-like material with more give, for the joints. The result is a hand that allows for precision during normal operation but has the flexibility to handle a range of different load scenarios without breaking. This approach preserves rigidity where it is needed and localizes compliance.
Rocking chair vs. door hinge
A key feature of CRAFT is how the finger knuckles are designed. The two joints closest to the fingertip use a rolling-contact mechanism rather than a conventional hinge. Think of a door hinge versus a rocking chair: a hinge pivots at a single fixed point and concentrates stress there, while a rocking chair rolls and spreads the load across a broader surface. CRAFT’s outer knuckles work like the rocking chair, which is why they hold up better under repeated contact than standard robotic finger designs.
These two outer knuckles are also mechanically linked, so they bend together in one smooth arc, much like the middle and tip joints of a human finger move in tandem. This means one motor drives a natural, coordinated curl rather than requiring two separate motors per finger.
“Instead of making the entire hand soft, we localized compliance to the joints and preserved rigidity where load-bearing and geometric predictability matter most,” said Jain. “This approach delivers both robustness and precision.”
Lightweight design, easy to repair
The hand is driven by 15 motors positioned behind the wrist, connected to the fingers through thin cables — much like tendons in a human hand — with three motors per finger. Relocating the motors away from the fingers keeps the system lightweight, compact and proportional to a human hand. Because the structural parts are 3D-printable, damaged parts are inexpensive and easy to replace.
Consistently higher success rates under testing
To support robot learning and data collection, the researchers paired CRAFT with a vision-based teleoperation system that allows users to control both a robotic arm and hand simultaneously using only a single digital camera. The system combines whole-body tracking and hand-pose estimation software to translate human motion directly into robotic movement without requiring gloves, wrist-mounted sensors or other wearable devices.
In structural testing, CRAFT matched the motion repeatability of the LEAP Lite Hand while tolerating nearly double the applied force, demonstrating superior resilience to impact. CRAFT also achieves simultaneous adduction-abduction and active fingertip joint motion — a combination that comparable open-source hands such as RUKA and ORCA do not currently offer together. In teleoperation tests conducted using a variety of small and delicate everyday objects — including a wine glass, an egg, a raspberry and a chip — users consistently achieved higher success rates with CRAFT than with a rigid robotic hand.
The benefits were particularly evident when handling delicate objects. CRAFT achieved a 100-percent success rate when manipulating fragile items, while the rigid comparison hand frequently crushed objects due to its inability to adapt its grip. Users also completed some tasks significantly faster because the hand’s give meant they didn’t need to slow down to manage every touch.
One task — picking a coin up off a flat table — illustrates why fingertip dexterity matters: it requires sliding a fingertip flush against a hard surface to get underneath a thin, low-profile object, something that is easy for a human hand but a common failure point for rigid robotic fingers.
The hand successfully covered all 33 grip types catalogued in the Feix grasp taxonomy, a standard reference describing the distinct ways human hands hold objects, from power grips on large tools to fine pinches of small items.
A video shows CRAFT picking up a raspberry without crushing it — a task that trips up rigid hands, which tend to apply uniform force regardless of how delicate the object is.
The work was presented at Robotics: Science and Systems (RSS) 2026, a premier international conference focusing on the algorithmic, mathematical and foundational aspects of robotics, in Sydney from July 14–17, where it won best paper at a dexterous manipulation workshop.
From the classroom to cutting-edge research
Since the release, researchers in Jain’s Visual and Embodied Intelligence Lab (UnVEIL) at UC Irvine have been putting CRAFT to work on harder problems. Graduate student Aditya Mittal first built his own CRAFT hand as part of Jain’s newly designed course, Deep Learning for Robotics — a first-of-its-kind offering at UC Irvine that drew students from both the computer science and engineering schools. He documented the full process in a blog post, covering everything from printing the parts and routing tendons to mounting the hand on a robot arm and running VR-based teleoperation, serving as a practical guide for others hoping to replicate the build.
“This was the first piece of hardware I’d ever built. What surprised me was how much the give of the joints changes the experience of actually using the hand. The joint bends on impact instead of snapping, and you stop worrying about breaking the object and start focusing on the task,” said Mittal.
Yuchen Song, a Ph.D. student in Jain’s lab, and Mittal have gone on to use CRAFT for research that pushes beyond what a human operator can do: a robotic arm fitted with a CRAFT hand held a wine glass steady while a second arm tipped a bottle to pour, with a learned policy executing the entire sequence autonomously. Sicheng He, a Ph.D. student who joined Jain’s lab this month, is researching ways to utilize physics-conforming simulation to further this research beyond what human operators can demonstrate.
Paying it forward with open source
The researchers believe CRAFT could help accelerate progress in robotic learning by providing an affordable, durable platform capable of collecting the large amounts of interaction data needed to train advanced manipulation policies. The hand is already in use at other universities and research institutions, where groups are building on the open-source design for their own work.
“The robotics community has been incredibly generous — LEAP, RUKA, ORCA and others built the foundation that made CRAFT possible. We wanted to pay that forward. By releasing the complete hardware and software stack as open source, we hope to make dexterous robot learning easier to reproduce, scale and deploy,” said Jain. “Our goal is to enable more researchers to build robust manipulation systems without the high costs and fragility that have traditionally limited the field.”
Additional information
The project is a collaboration with the University of Illinois Urbana-Champaign, where Leo Lin, Shivansh Patel and Jay Moon co-led the hand’s design and build, with Svetlana Lazebnik as co-advisor. Full hardware documentation, software, assembly videos and build guides are available at the project’s GitHub page.
The team has posted a step-by-step assembly video walking through the full build — from 3D-printed parts to tendon routing and final calibration — as documentation for anyone looking to build their own.
Robotis has created a bundle providing hardware components to build the hand for $471.
About the University of California, Irvine: Founded in 1965, UC Irvine is a member of the prestigious Association of American Universities and is ranked among the nation’s top 10 public universities by U.S. News & World Report. The campus has produced five Nobel laureates and is known for its academic achievement, premier research, innovation and anteater mascot. Led by Chancellor Howard Gillman, UC Irvine has more than 36,000 students and offers 224 degree programs. It’s located in one of the world’s safest and most economically vibrant communities and is Orange County’s second-largest employer, contributing $7 billion annually to the local economy and $8 billion statewide. For more on UC Irvine, visit www.uci.edu.
Media access: Radio programs/stations may, for a fee, use an on-campus studio with a Comrex IP audio codec to interview UC Irvine faculty and experts, subject to availability and university approval. For more UC Irvine news, visit news.uci.edu. Additional resources for journalists may be found at news.uci.edu/media-resources.
Related links
CRAFT project page https://craft-hand.github.io/
Paper (appeared at RSS 2026, Sydney) https://arxiv.org/abs/2603.12120
Assembly video https://www.youtube.com/watch?v=qfrquzk72vA
Build log: replicating CRAFT in the lab https://adityamittal03.github.io/blog/recreating-craft-hand/
CRAFT explainer thread on X/Twitter https://x.com/unnatjain2010/status/2043799865868857562
Department of Computer Science https://cs.ics.uci.edu/
Donald Bren School of Information and Computer Sciences https://ics.uci.edu