Author: Denis Avetisyan
Researchers have developed a novel robotic hand that combines hard and soft materials to achieve robust and adaptable manipulation capabilities.

This paper details the design and control of CRAFT, a tendon-driven hand with hybrid compliance and rolling-contact joints for dexterous manipulation and improved durability.
Achieving robust and repeatable dexterity in robotic hands remains a challenge due to the trade-off between fragility and durability. This paper introduces [latex]CRAFT[/latex]: A Tendon-Driven Hand with Hybrid Hard-Soft Compliance, a novel design that leverages soft joints and rigid links to enhance both manipulation performance and mechanical resilience. By employing rolling-contact surfaces and a tendon-driven architecture, [latex]CRAFT[/latex] demonstrates improved strength, endurance, and handling of delicate objects, successfully completing 33/33 grasps in the Feix taxonomy. Could this hybrid approach unlock more adaptable and reliable robotic manipulation capabilities for a wider range of real-world applications?
The Limits of Rigidity: Towards Compliant Grasping
Conventional robotic hands frequently employ direct-drive actuation, where motors directly control joint movements – a design that, while powerful, introduces significant challenges when interacting with fragile objects or unpredictable environments. This approach often results in high impact forces and limited adaptability, making delicate manipulation – such as grasping a ripe tomato without bruising it – remarkably difficult. The rigidity inherent in these systems struggles to absorb unexpected collisions, increasing the risk of both damage to the robot and the objects it handles. Consequently, these hands often lack the finesse required for complex real-world tasks, highlighting the need for alternative designs that prioritize controlled interaction and robustness over sheer force.
Conventional robotic hands, engineered with stiff links and direct-drive actuators, frequently falter when confronted with the unpredictable nature of real-world interactions. This inflexibility renders them susceptible to damage upon impact – a dropped object or unexpected collision can easily strain or break delicate components. More significantly, these designs struggle with tasks requiring subtle adjustments or the ability to conform to irregularly shaped objects; grasping a ripe tomato without crushing it, or assembling components with varying tolerances, proves exceptionally difficult. The lack of adaptability isnât simply a matter of precision; it fundamentally limits the range of tasks a robot can reliably perform outside of highly controlled environments, hindering progress in areas like in-home assistance, disaster response, and complex manufacturing.
Current robotics research increasingly focuses on designs that move beyond the limitations of traditionally rigid robotic hands. These new approaches prioritize inherent compliance – the ability to yield and adapt to external forces – as a pathway to both increased robustness and enhanced dexterity. Rather than resisting contact, compliant hands absorb impact, reducing the risk of damage to both the robot and the objects it manipulates. This is achieved through innovative materials, bio-inspired designs mimicking the human handâs musculotendinous system, and novel actuation methods like series elastic actuators. The result is a robotic hand capable of grasping delicate or irregularly shaped objects with greater security and performing complex tasks in unstructured environments, opening doors to applications ranging from collaborative robotics to advanced manufacturing and even surgical assistance.
![CRAFT successfully demonstrates a versatile range of 33 grasp types from the Feix taxonomy [8], including power, precision, and intermediate configurations, validating its compliant design for general-purpose manipulation.](https://arxiv.org/html/2603.12120v1/x10.png)
CRAFT: A Synthesis of Strength and Suppleness
CRAFTâs mechanical design incorporates a hybrid compliance strategy achieved through the integration of rigid Polylactic Acid (PLA) segments and compliant Thermoplastic Polyurethane (TPU) joints. This approach allows for targeted stiffness and flexibility; PLA segments provide structural integrity and efficient force transmission during interaction, while the TPU joints introduce compliance that absorbs impact and distributes loads. The combination optimizes both the force applied to objects and the robotâs resilience to external forces, minimizing damage risk and maximizing operational robustness. This hybrid method contrasts with fully rigid or fully compliant hands, offering a balance between precision and shock absorption capabilities.
Tendon-driven actuation in the CRAFT hand design reduces overall weight and mechanical complexity by positioning the actuation motors within the forearm housing. This architecture eliminates the need for individual motors at each finger joint, thereby decreasing the mass concentrated in the hand itself and simplifying the mechanical linkages. Relocating the motors also enables faster response times and improved energy efficiency, as shorter and lighter tendons transmit the actuation force to the fingers. The system utilizes a series of tendons routed through the fingers, allowing for a compact design and a greater range of motion compared to direct-drive systems.
The CRAFT hand design incorporates rolling-contact joints and a bidirectional mechanical linkage to improve both durability and precision. This configuration achieves a joint angle tracking error of less than 0.01 radians, a performance level comparable to that of traditional rigid robotic hands. The rolling-contact design minimizes wear and maximizes lifespan, while the bidirectional linkage ensures consistent load distribution across the fingers and facilitates coordinated movement. This mechanical approach allows for repeatable performance and reduces the potential for positional drift during operation.
![Despite employing flexible tendons, CRAFT consistently maintains a joint angle tracking error of less than [latex]0.01[/latex] radians over an hour of continuous grasp-release cycles, performing comparably to the rigid LEAP baseline and demonstrating robust repeatability.](https://arxiv.org/html/2603.12120v1/x6.png)
Validating Dexterity: Teleoperation and Simulated Performance
Teleoperation studies were conducted to assess the robotic handâs performance in complex manipulation scenarios, employing vision-based whole-body tracking to capture nuanced control and dexterity. This methodology allowed researchers to directly observe human operators controlling the hand in real-time, providing data on the handâs kinematic capabilities and identifying limitations in its range of motion and grip stability. The captured data included operator joint angles, force exertion, and manipulation strategies, which were then analyzed to evaluate the handâs usability and effectiveness in tasks requiring precise movements and delicate handling. These studies served as a crucial benchmark for evaluating the handâs design and control algorithms prior to automated operation.
Simulations were conducted using the MuJoCo physics engine, employing Universal Robot Description Format (URDF) models to represent the robotic hand. This approach enabled comprehensive testing of the handâs kinematic and dynamic properties without physical hardware limitations. Specifically, researchers were able to systematically vary loading conditions – including force, weight, and object characteristics – to assess the handâs range of motion, stability, and control performance across numerous iterations. The URDF models facilitated accurate representation of the handâs geometry, mass distribution, and joint limits, providing a reliable platform for evaluating its mechanical behavior and identifying potential design improvements prior to physical prototyping.
During teleoperation studies, the CRAFT hand achieved complete coverage of the 33-grasp Feix taxonomy, indicating its ability to perform a wide variety of grasp types. Quantitative performance data revealed a 100% success rate when manipulating fragile objects, a significant improvement over the 60% success rate achieved by the LEAP hand under identical testing conditions. This indicates a superior level of dexterity and control for CRAFT during delicate manipulation tasks, as demonstrated through operator-controlled teleoperation.

Towards Accessible Robotics: Impact and Future Directions
The design of the CRAFT hand prioritizes resilience and flexibility, positioning it as a strong candidate for robotic systems operating in real-world, unpredictable settings. This capability extends to applications like assistive robotics, where gentle yet firm grasping is crucial for handling objects for individuals with limited mobility, and in-home care, where robots might assist with tasks requiring delicate manipulation amidst cluttered or changing environments. Unlike systems demanding highly structured surroundings, CRAFTâs inherent adaptability allows it to maintain a secure grip and perform tasks reliably even when confronted with unexpected obstacles or variations in object shape and texture, offering a pathway towards more practical and user-friendly robotic assistance.
The collaborative potential of the CRAFT hand is significantly amplified by its open-source design. This accessibility allows researchers worldwide to not only scrutinize and validate the robotic handâs mechanics and control algorithms, but also to actively contribute to its improvement and adaptation. Hobbyists and independent makers gain a readily available, fully-documented platform for experimentation, fostering a broader understanding of robotics and accelerating innovation outside of traditional research institutions. By removing barriers to entry, the open-source nature of CRAFT encourages a decentralized approach to development, potentially leading to unforeseen applications and specialized customizations tailored to diverse needs and environments – ultimately driving the field of accessible robotics forward at an unprecedented pace.
The newly developed CRAFT hand demonstrates a significant leap in robotic grasping capabilities, achieving a pull-out strength of 15.29 N – nearly double that of the LEAP hand, a previously established benchmark. Importantly, this enhanced performance isnât achieved at the cost of efficiency; CRAFT consumes approximately 50% less current during holding tasks. This reduction in power consumption is crucial for extending operational lifespan, particularly in mobile or untethered robotic applications. Rigorous testing confirms that this improvement in strength and efficiency doesnât compromise precision or versatility, as CRAFT maintains comparable repeatability and a complete range of grasp types – demonstrating its potential for reliable and adaptable manipulation in diverse environments.

The development of CRAFT exemplifies a principle central to robust design: minimizing unnecessary complexity. The handâs hybrid hard-soft compliance, particularly the innovative rolling-contact joints, demonstrates a focus on fundamental mechanics over elaborate solutions. As Grace Hopper once stated, âItâs easier to ask forgiveness than it is to get permission.â This resonates with the approach taken in CRAFTâs creation; rather than adhering to conventional rigid hand designs, the researchers prioritized a system allowing for both durability and adaptability through a novel, yet elegantly simple, mechanical architecture. This willingness to explore unconventional methods yields a system capable of dexterous manipulation while streamlining the learning process-a testament to the power of foundational design.
Future Directions
The presented work, like any attempt to replicate the intricacy of the human hand, reveals as much about what remains unsolved as it accomplishes. CRAFTâs hybrid approach offers a promising step toward durable, adaptable manipulation, but it is crucial to recognize this as infrastructural development, not wholesale reconstruction. The hand itself is not the ultimate goal; it is a node within a larger system of perception, control, and learning. Future iterations must move beyond isolated performance metrics and address the complexities of integrating such a hand into a complete robotic architecture.
A key limitation lies in the current reliance on pre-programmed control and reinforcement learning within constrained environments. True dexterity demands a system capable of anticipating unforeseen contact, adapting to novel objects, and learning from sparse, real-world experience. This necessitates a shift toward more biologically-inspired control strategies, focusing on intrinsic motivation and embodied intelligence. The challenge is not simply to teach the hand to grasp, but to allow it to discover how to manipulate.
Ultimately, the long-term success of this field hinges on recognizing that robustness and adaptability are not features to be added, but emergent properties of a well-structured system. Just as a cityâs resilience comes not from fortifying individual buildings, but from the flow of its streets and the diversity of its inhabitants, so too must robotic hands evolve through incremental improvements, built upon a foundation of elegant design and systemic understanding.
Original article: https://arxiv.org/pdf/2603.12120.pdf
Contact the author: https://www.linkedin.com/in/avetisyan/
See also:
- CookieRun: Kingdom 5th Anniversary Finale update brings Episode 15, Sugar Swan Cookie, mini-game, Legendary costumes, and more
- Robots That React: Teaching Machines to Hear and Act
- Call the Midwife season 16 is confirmed â but what happens next, after that end-of-an-era finale?
- PUBG Mobile collaborates with Apollo Automobil to bring its Hypercars this March 2026
- Taimanin Squad coupon codes and how to use them (March 2026)
- Heeseung is leaving Enhypen to go solo. K-pop group will continue with six members
- How to get the new MLBB hero Marcel for free in Mobile Legends
- Alan Ritchsonâs âWar Machineâ Netflix Thriller Breaks Military Action Norms
- Gold Rate Forecast
- Robots Learn by Example: Building Skills from Human Feedback
2026-03-13 13:31