The Gentle Grip: A New Robotic Hand for Delicate Tasks

by

Author: Denis Avetisyan

Researchers have developed a remotely controlled, tendon-driven robotic hand powered by soft actuators, enabling compliant and safe manipulation of fragile objects.

The developed musculoskeletal hand system exhibits inherent compliance, demonstrated by the actuator packs’ ability to deform under external force and accommodate twisting, allowing for easy object removal even during operation-a characteristic that, unlike conventional motor-driven hands exhibiting rigid responses, enables safe absorption of impacts, such as that of a tennis ball, through large deformation.
The developed musculoskeletal hand system exhibits inherent compliance, demonstrated by the actuator packs’ ability to deform under external force and accommodate twisting, allowing for easy object removal even during operation-a characteristic that, unlike conventional motor-driven hands exhibiting rigid responses, enables safe absorption of impacts, such as that of a tennis ball, through large deformation.

This work presents a sensorless, anthropomorphic musculoskeletal hand driven by HASEL actuators, achieving inherent compliance and remote actuation capabilities.

Achieving both high dexterity and safe, compliant interaction remains a central challenge in robotic manipulation. This is addressed in ‘A Sensorless, Inherently Compliant Anthropomorphic Musculoskeletal Hand Driven by Electrohydraulic Actuators’, which presents a novel remotely actuated, tendon-driven hand powered by [latex]HASEL[/latex] actuators. The resulting design prioritizes compliant grasping and inherent safety through actuator force-limiting characteristics, while simultaneously achieving sensorless operation via self-sensing capabilities. Could this approach pave the way for more intuitive and robust robotic systems capable of seamlessly interacting with fragile objects and unstructured environments?

The Fragility of Precision: Towards Compliant Grasping

Conventional robotic grippers, despite advances in motor control, frequently struggle with tasks requiring delicate manipulation or uncertain object properties. Their rigid designs and reliance on precise positional control often result in either crushing fragile objects or failing to secure a firm grasp on slippery or irregularly shaped items. This inherent inflexibility stems from a lack of intrinsic adaptability – the human hand effortlessly conforms to an object’s surface, distributing force and adjusting grip strength based on tactile feedback. Consequently, current robotic systems demonstrate a high propensity for causing damage during handling, experiencing unintentional slippage, and requiring significant recalibration for even minor variations in object shape or material – limitations that hinder their broader implementation in unstructured environments and alongside human workers.

Replicating the dexterity of the human hand necessitates a departure from conventional robotic actuation methods, which typically rely on rigid joints and discrete movements. Instead, bioinspired designs are exploring soft robotics principles, utilizing compliant materials and distributed actuation to achieve a more fluid and adaptable grasp. This shift extends beyond mechanics; truly nuanced control demands sophisticated sensing capabilities integrated directly within the robotic hand itself. Rather than external vision systems alone, researchers are embedding sensors that detect force, texture, and slippage across the entire grasping surface, providing real-time feedback for precise adjustments. This intrinsic sensing allows the robotic hand to ‘feel’ the object it is manipulating, enabling it to conform to irregular shapes, apply appropriate pressure, and maintain a secure hold without causing damage – mirroring the subtle, subconscious control humans exhibit with effortless ease.

Achieving truly robust and delicate manipulation with robotic systems necessitates a departure from rigid designs and a move toward compliant mechanisms – structures that yield and adapt to contact forces, much like a human hand. This isn’t simply about softness; it’s about distributing forces, improving contact stability, and preventing damage to both the robot and the objects it handles. Crucially, this approach must be paired with intrinsic sensing – the ability of the robotic hand itself to ‘feel’ the forces and textures it encounters, without relying on external vision systems. Embedded sensors within the compliant structures provide real-time feedback on grip force, object shape, and slippage, enabling the robot to dynamically adjust its grasp and perform complex tasks with a level of finesse previously unattainable. This combination allows for adaptable grasping, making robots suitable for handling fragile or irregularly shaped objects in unpredictable environments.

Current robotic grasping systems frequently struggle with tasks requiring finesse or adaptability, often resulting in dropped objects or damage to fragile items. This shortfall stems from designs prioritizing force over feel, and rigid structures over natural compliance. Consequently, researchers are actively pursuing novel robotic hand architectures that move beyond conventional, gear-driven mechanisms. These emerging designs emphasize soft materials, underactuation, and intrinsic sensing – features mirroring the human hand’s ability to conform to object shapes and regulate grip force through tactile feedback. The goal is to create hands capable of not just grasping, but of feeling their way around objects, enabling truly robust and delicate manipulation in unstructured environments and ultimately bridging the gap between robotic precision and human dexterity.

A tendon-driven robotic hand utilizes soft HASEL actuators and a pulley system to achieve compliant, independent finger control for grasping objects, as demonstrated with a cube.
A tendon-driven robotic hand utilizes soft HASEL actuators and a pulley system to achieve compliant, independent finger control for grasping objects, as demonstrated with a cube.

The Resilience of Distributed Force: Tendon-Driven Design

Tendon-driven systems achieve inherent compliance through the elastic properties of the tendons themselves and the distributed nature of force transmission. Unlike rigid linkages, tendon pathways can deform and absorb impact forces, increasing system robustness and reducing the risk of damage upon collision. This distributed force profile minimizes stress concentrations and allows the system to passively adapt to external disturbances and varying loads. Consequently, tendon-driven architectures improve safety in human-robot interaction scenarios and enhance adaptability to unstructured environments by allowing for more forgiving and resilient movements.

The system employs a pulley routing mechanism to mechanically amplify displacement generated by the actuators. This configuration enables a larger range of motion at the robotic hand compared to the limited stroke of the HASEL actuators themselves. Specifically, the pulley arrangement provides a mechanical advantage, trading force for displacement; for each unit of actuator travel, the tendon extends a greater distance, effectively scaling the output motion. Multiple pulleys and strategically routed tendons are used to achieve complex movements and distribute forces evenly, maximizing the efficiency of the limited actuator stroke and enabling precise control of the robotic hand.

The core of this design relies on the use of Dielectric Elastomer Actuators (DEAs), specifically those employing a Hydraulic Amplification System for Enhanced Lift (HASEL). These actuators demonstrate response times in the millisecond range and exhibit high strain potential, exceeding 30% in some configurations. Versatility is achieved through tunable mechanical properties enabled by varying elastomer composition and pre-strain levels. Furthermore, HASEL actuators possess inherent self-sensing capabilities; changes in capacitance, directly proportional to applied strain, can be measured and used for position and force feedback without the need for external sensors. This intrinsic sensing reduces system complexity and improves overall control accuracy.

Remote actuator placement, achieved through tendon transmission, significantly reduces mass at the terminal end of the robotic system – specifically, the hand – by relocating heavier components such as motors and control electronics to a central body location. This mass reduction directly improves dynamic performance by lowering inertia and enabling faster, more precise movements. Furthermore, minimizing distal mass simplifies control algorithms and reduces the energy required for actuation and stabilization, leading to increased efficiency and responsiveness. The resulting system exhibits improved agility and a greater capacity for interaction with delicate or sensitive objects.

This tendon-driven hand design utilizes a modular structure with stacked actuator packs and a coupled PIP/DIP mechanism to achieve precise finger control through flexor tendons with a pulley system and passively restored extensor tendons, all managed via adjustable wire tensioning gears.
This tendon-driven hand design utilizes a modular structure with stacked actuator packs and a coupled PIP/DIP mechanism to achieve precise finger control through flexor tendons with a pulley system and passively restored extensor tendons, all managed via adjustable wire tensioning gears.

The Eloquence of Internal Feedback: Intrinsic Sensing

Conventional robotic grasp detection typically employs external sensors – such as force/torque sensors, tactile arrays, or vision systems – to determine contact and grasp stability. These external systems introduce significant complexity to the robotic system through additional hardware, wiring, and calibration procedures. Furthermore, data acquisition and processing from these sensors introduce latency, limiting the responsiveness of the grasp and hindering the robot’s ability to react to dynamic changes in the environment or object being manipulated. This dependence on external feedback loops creates a bottleneck in achieving precise and adaptive grasp control, particularly when handling delicate or irregularly shaped objects.

The HASEL actuator’s inherent electrical properties enable its functionality as a self-sensing actuator, eliminating the need for external sensors. Specifically, the actuator’s capacitance changes proportionally with its deformation; as the actuator expands or contracts, its capacitance value alters. This change in capacitance directly correlates with the actuator’s position and the force it exerts. By monitoring the actuator’s electrical current during operation, a precise and real-time measurement of deformation and applied force is achievable. This intrinsic sensing capability is a direct result of the dielectric elastomer material and the conductive electrodes within the HASEL structure, providing a self-contained sensing mechanism.

The operating current of a HASEL actuator correlates directly with its deformation and the external force it exerts. This relationship stems from the actuator’s dielectric elastomer construction; as the elastomer stretches or compresses, its capacitance changes, directly impacting the current required to maintain a given voltage. Consequently, precise measurement of the current draw provides a quantitative assessment of the actuator’s state, eliminating the need for separate position or force sensors. This intrinsic sensing capability offers a high-bandwidth, localized measurement of mechanical variables, as the electrical measurement is inherently tied to the physical deformation of the actuator material itself.

Contact-aware control and robust grasp detection are achieved through the utilization of current sensing within the HASEL actuator. The actuator’s operating current directly correlates with its deformation and the external forces it experiences; this relationship allows for real-time monitoring of interaction with the environment. Demonstrated performance includes successful grasping of fragile objects, such as a paper balloon, achieved via current thresholding; when the actuator makes contact and begins to deform under load, the current draw exceeds a pre-defined threshold, signaling contact and enabling precise control adjustments to prevent damage. This intrinsic sensing capability eliminates the need for external sensors, simplifying system design and improving responsiveness.

Real-time current sensing enables the algorithms to perform both grasp detection and contact-aware grasp control.
Real-time current sensing enables the algorithms to perform both grasp detection and contact-aware grasp control.

Towards a Symbiotic Future: Compact Design & Implications

The development prioritizes practicality through a design that seamlessly integrates actuators within a human-sized forearm structure. This approach moves beyond theoretical robotic hands by creating a system that mimics the scale and form factor of a human limb, enabling easier implementation in real-world scenarios. By housing the actuation mechanisms within the forearm, rather than as bulky external components, the device achieves both compactness and improved ergonomics. This careful consideration of form allows for more natural interaction with objects and environments, and lays the groundwork for robotic assistants that can operate comfortably alongside humans in tasks demanding dexterity and precision – ultimately bridging the gap between laboratory prototypes and genuinely useful robotic tools.

The innovative design leverages a Peano-HASEL actuator, a soft robotic component known for its high power-to-weight ratio and compliance, paired with a carefully selected liquid dielectric. This combination is critical for achieving efficient and reliable operation at the high voltages necessary to drive the actuator’s deformation. Unlike traditional rigid actuators, this system avoids the need for bulky and heavy mechanical transmissions, while the liquid dielectric ensures electrical insulation and facilitates uniform voltage distribution. The result is a compact and responsive system capable of generating substantial forces-up to 0.53 N for the index finger-despite its relatively small size and weight, representing a significant step toward more practical and adaptable soft robotic hands.

The developed robotic hand represents a notable leap forward in prosthetic and assistive technologies, successfully generating substantial fingertip forces essential for interacting with objects. Specifically, the system achieves a gripping force of 0.53 Newtons at the index finger and 0.26 Newtons at the thumb when operated at 5.5 kilovolts. These values demonstrate the actuator’s capacity to exert meaningful pressure – enough to reliably hold and manipulate a variety of items. This level of force, combined with the compact design, positions the hand as a viable solution for applications requiring both dexterity and a secure grasp, moving closer to the capabilities of a natural human hand and opening doors for more sophisticated robotic assistance.

The development of this compact robotic forearm signifies a crucial step towards more versatile and adaptable robotic assistants. Current designs, achieving a 30-degree joint angle range, are poised to extend beyond the limitations of structured factory settings and begin operating effectively in the complexities of real-world, unstructured environments. This capability opens possibilities for assistance in fields like healthcare, where delicate manipulation is paramount, and in-home support, where robots must navigate unpredictable surroundings and perform nuanced tasks. Further refinement of the actuator technology and expansion of the joint angle range promises even greater dexterity and functionality, ultimately enabling a new generation of robots capable of seamless integration into everyday life and offering support in a wide variety of challenging scenarios.

A soft muscle was created by stacking Peano-HASEL actuators-measuring [latex]210 \text{mm} \times 93 \text{mm}[/latex] individually-to achieve amplified force-displacement characteristics as demonstrated by a custom characterization setup utilizing a linear motor and load cell.
A soft muscle was created by stacking Peano-HASEL actuators-measuring [latex]210 \text{mm} \times 93 \text{mm}[/latex] individually-to achieve amplified force-displacement characteristics as demonstrated by a custom characterization setup utilizing a linear motor and load cell.

The pursuit of increasingly complex robotic systems, as demonstrated by this remotely actuated, tendon-driven hand, inevitably introduces layers of potential failure. This research, focused on compliant manipulation and self-sensing, attempts to mitigate those risks through inherent design-a strategy that acknowledges the eventual decay of any system. As Ken Thompson observed, “Debugging is twice as hard as writing the code in the first place. Therefore, if you write the code as cleverly as possible, you are, by definition, not smart enough to debug it.” This sentiment echoes the approach taken here; by prioritizing simplicity and inherent compliance-designing a system that gracefully accommodates imperfections-the team aims to reduce the complexity of future maintenance and increase the hand’s long-term resilience, acknowledging that even the most innovative designs are subject to the relentless march of time.

What’s Next?

The presented system, a remotely actuated hand, embodies a predictable trajectory: increased complexity yielding diminishing returns in uptime. The pursuit of biomimicry, while aesthetically pleasing, merely shifts the locus of failure. The inherent latency in remote actuation, the tax every request must pay, remains a fundamental constraint, especially as degrees of freedom multiply. Future iterations will undoubtedly focus on minimizing this delay, perhaps through localized processing or predictive algorithms, but the fundamental problem of information transit persists.

The claim of ‘self-sensing’ warrants scrutiny. All sensing is, ultimately, an external measurement interpreted as internal state. The system does not know its position; it infers it. The challenge lies not in detecting deformation, but in predicting its consequences – in modeling the fragility of the objects it interacts with. A more fruitful avenue may lie in embracing controlled instability, in allowing the hand to feel its way through a task, accepting that perfect knowledge is an asymptotic ideal.

Stability is an illusion cached by time. The system’s current reliance on centralized control and external actuation introduces single points of failure. A truly robust hand will not seek to avoid damage, but to tolerate it, to redistribute stress, and to adapt its behavior in real-time. The next iteration will not be about achieving greater precision, but about gracefully accepting imperfection.

Original article: https://arxiv.org/pdf/2603.24357.pdf

Contact the author: https://www.linkedin.com/in/avetisyan/

See also:

2026-03-26 17:42