In the rapidly evolving field of humanoid robots, the dexterous hand serves as a critical functional endpoint, enabling complex interactions and tasks. As industries push for advancements in embodied intelligence, the performance capabilities of these hands have seen significant breakthroughs. However, the morphological expression of dexterous hands, which plays a vital role in human-robot interaction, has not received the attention it deserves relative to its importance. This article presents an innovative morphological design method for dexterous hands in humanoid robots, leveraging principles from artistic anatomy. By employing quantitative analysis of perceptual descriptors through Kansei Engineering, establishing aesthetic templates, and facilitating engineering transformations, this method uncovers the connection between artistic anatomical principles and the design of dexterous hand forms. It is applied to key morphological elements such as shape, proportion, and structure, resulting in a dexterous hand that maintains functional reliability while achieving a remarkable enhancement in visual aesthetics.
The development of dexterous hands for humanoid robots has gained momentum in recent years, driven by their role in precise manipulation and interaction. These hands are essential for tasks ranging from industrial assembly to service applications, where their ability to grasp and manipulate objects mimics human dexterity. Despite this focus on functional performance, the morphological aspects—such as visual appeal and emotional resonance—are often overlooked. In humanoid robots, the hand is not just a tool but a medium for communication; its form can evoke positive psychological responses and foster natural emotional bonds with users. This underscores the need for a design approach that balances technical specifications with aesthetic qualities, addressing the gap in current research that prioritizes functionality over form.
Current dexterous hands in humanoid robots frequently suffer from morphological issues, including bulky shapes, disproportionate dimensions, oversized scales, and structural inaccuracies. For instance, when compared to a typical male hand measuring 190 mm in length, many commercially available hands exhibit discrepancies in proportion, size, and posture. This misalignment stems from a broader disconnect in understanding morphology: it encompasses not only external造型 but also internal structure, as the two are inherently linked. In robotics design, efforts often concentrate on superficial aspects like form, color, and texture, neglecting the integral relationship between structure and appearance. This has led to a theoretical void in morphological design methods for humanoid robotic hands, where functional requirements dominate, resulting in aesthetic compromises. Industrial design principles advocate for the fusion of art and science, yet dexterous hands have struggled to achieve this balance due to the subjective nature of aesthetics and the challenges of quantifying it for engineering applications.
The core challenge lies in translating感性美学 into quantifiable engineering parameters. Hand morphology in art is often described with subjective,感性 terms—for example, in classical literature and sculpture, hands are depicted as “slender” or “powerful” based on artistic exaggeration. This contrasts with the scientific objectivity required in engineering, which relies on data, models, and verification. Technical constraints, such as space limitations and material properties, further complicate this translation, demanding optimized solutions under multiple restrictions. The key is to identify the major obstacles to this transformation and achieve a high evaluation in both aesthetic and performance metrics. For humanoid robots, this means developing a method that integrates perceptual analysis with rigorous engineering, ensuring that the hands not only function effectively but also resonate emotionally with users.
To address this, we adopted a Kansei Engineering approach to quantitatively analyze perceptual descriptors related to dexterous hand morphology. We began by selecting a sample of existing dexterous hands and conducting surveys to assess their “beauty and affinity.” Using a semantic differential method with a five-point Likert scale ranging from -10 (no beauty or affinity) to 10 (extremely high beauty and affinity), 30 participants evaluated the samples. The overall scores were low, indicating general dissatisfaction, but we identified higher-scoring samples for further analysis. Through open-ended questions, we extracted key factors influencing perceptions, which were consolidated into four elements: proportion, appearance shape, construction, and motion bionics. These elements collectively point to a “high human-like bionic degree,” reflecting user expectations that dexterous hands should closely resemble human hands in both form and function. This aligns with trends in humanoid robots, where increased degrees of freedom, smaller sizes, and higher integration are driving more anthropomorphic designs.
The four elements were analyzed using an Analytic Hierarchy Process (AHP) to derive quantifiable engineering indicators. We constructed a hierarchical model with layers including the base layer (beauty and affinity, high human-like bionic degree), indicator layer (high integration), decomposition layer (proportion and appearance, structure), element layer (proportion, appearance shape, construction, motion bionics), cognition layer (methods from artistic anatomy abstraction), guidance layer (proportional relationships and whole-hand aesthetic template, hard materials, non-clashing colors, human-like forms and joint structures, “dexterous” performance, direct force transmission, built-in actuation), expansion layer (golden ratios for whole hand and parts, balanced male and female hand characteristics, abstract translation of geometric joint axes, simulation of muscle and tendon stretching, simulation of bone joint movements), and technical layer (detailed data extraction, verification, and calibration). This framework allowed us to translate subjective perceptions into objective metrics, with the main challenge identified as high-density integration of actuation systems in small spaces.
Next, we developed a standard hand aesthetic template based on artistic anatomical principles, serving as a 3D digital model for data and造型 reference. This template balanced male and female hand characteristics—incorporating the strong bone structure of male hands and the slender uniformity of female hands—to achieve a universal form that ensures structural stability and operational precision. The size was set at 190 mm in length, equivalent to an average adult male hand, based on anthropometric data. Using proportional segmentation principles from artistic anatomy, we defined dimensions for all hand parts, with hand length as the primary基准. The aesthetic template not only summarized visually appealing hand forms but also coupled with engineering needs, reducing the difficulty of coordinating technology and aesthetics in humanoid robots.
The design and fabrication phase focused on solving the high-density integration problem in small spaces, applying abstract归纳 from artistic anatomy to structural design. The process involved several key steps:
Step 1: Translation of complex motions and geometric reconstruction of complex joints. Human hand movements are intricate, involving “rolling + sliding” motions at joints like the metacarpophalangeal (MCP) joints. Engineering constraints made it impractical to fully replicate this, so we abstracted these motions into rotational movements using axis-based mechanisms. For example, the non-orthogonal axes of MCP joints—which allow flexion, extension, abduction, adduction, and rotation—were simplified to orthogonal geometric states, focusing on the primary movements of flexion-extension and lateral swing. This transformation reduced complexity while preserving essential functionality for humanoid robots, as described by the equation for joint motion conversion: $$ \theta_{\text{engineered}} = f(\theta_{\text{biological}}) $$ where biological joint angles are mapped to engineered equivalents.
Step 2: Joint mapping guided by distributed actuation. The human hand relies on distributed muscles, tendons, and ligaments for coordinated movements. To mimic this, we mapped hand joint characteristics to 22 degrees of freedom (DoFs), including 6 orthogonal joint axes for 6 dual-DoF joints and 10 axis-rotation joints for 10 single-DoF joints. This was achieved with 22 motor modules in a fully active, distributed actuation system, outperforming existing hands like the Shadow Hand (20 active DoFs) and Tesla Optimus (17 active DoFs). The distribution follows the formula for DoF allocation: $$ \text{Total DoFs} = \sum (\text{joint types} \times \text{DoF per joint}) $$ ensuring high dexterity in humanoid robots.
Step 3: Clearance processing inspired by joint decoupling characteristics. Human hand joints operate independently due to decoupled actuation, enabling precise motions. In our design, we used multiple motors for decoupled joints and implemented clearance in hard materials to avoid interference during movement. For instance, the fifth metacarpal was isolated to allow inward flexion during gripping, with corresponding clearance in the palm area. This enhanced the bionic posture, a feature often missing in current products. The clearance design can be modeled as: $$ d_{\text{clearance}} = \max(\Delta x, \Delta y, \Delta z) $$ where $\Delta x$, $\Delta y$, and $\Delta z$ represent spatial tolerances.
Step 4: Mechanism integration and mechatronic assembly. We designed bionic joint motion chains using a “mortise-tenon” configuration to map joint morphology and function, improving visual coherence. Actuation units—integrating custom servos and reduction mechanisms—were topologically optimized for compact arrangement, simulating the dynamics of muscle-tendon systems. Electrical wiring was distributed based on human joint ligament patterns, increasing space utilization and visual harmony. This addressed the small-space integration challenge, resulting in a hand size smaller than counterparts like DexHand and Shadow Hand.
Step 5: Final adjustments and定型. We refined the hand through topological checks for surface continuity, motion simulations to prevent interference, and material adaptations like electronic skin fitting that accounts for curvature and strain thresholds. After validation, all functional metrics were unified in the assembly, leading to the final production.
The resulting dexterous hand, named “Hand A,” was designed for fine manipulation scenarios in humanoid robots, with applications in research, education, service, industrial manufacturing, healthcare, space exploration, rescue operations, and art. It maintains power capabilities while emphasizing precision. Upon completion, we conducted another survey with 20 participants using the same Likert scale, comparing Hand A to previously high-rated samples. Hand A scored significantly higher, confirming its superior morphological表现. It was officially launched at the IEEE International Conference on Robotics and Automation (ICRA 2025), where it received positive feedback for its form and performance.
A comparative analysis of Hand A with existing dexterous hands highlights its advanced features. The table below summarizes key performance indicators, demonstrating Hand A’s advantages in size, full active DoFs, dynamic tactile sensing, and distributed built-in actuation.
| Product | DoFs | Actuation Method | Sensing | Output Force | Speed | Precision | Sensitivity |
|---|---|---|---|---|---|---|---|
| Shadow Hand | 20 active, 4 passive | Distributed, external | Tactile | N/A | N/A | N/A | N/A |
| DH2012 | 12 active, 8 passive | Centralized + distributed, internal | Tactile | 15 kg | N/A | N/A | N/A |
| Unitree Dex5 | 16 active, 4 passive | Centralized + distributed, internal | Tactile | N/A | N/A | N/A | N/A |
| MagicHand S01 | 11 active | Centralized, internal | Current, tactile | 5 kg | N/A | N/A | N/A |
| Hand A | 22 active | Distributed, internal | Dynamic Tactile Array (DTA) | 20 N per fingertip | 4 Hz/s | 1 mm spatial resolution | 6,000 pressure levels, 180 Hz frame rate |
Hand A achieves a balance of aesthetic and functional high evaluations, meeting the预设 goals for humanoid robots. The integration of artistic anatomy principles into the morphological design process represents a parallel, cross-disciplinary approach where engineering and art are nested. This method elevates aesthetic cognition beyond mere imitation, transforming formal aesthetic rules into core engineering logic. It finds an equilibrium between rigorous technology and the感性 expectations of beauty, offering a replicable framework for future developments in humanoid robots. Further research should expand the scope, deepen the methodology, and involve long-term dynamic follow-ups to refine this approach.
In summary, the morphological design of dexterous hands for humanoid robots benefits greatly from the fusion of artistic anatomy and engineering. By quantifying perceptual elements, building standardized templates, and overcoming integration challenges, we have demonstrated that it is possible to enhance both visual appeal and performance. This advancement not only improves user interaction with humanoid robots but also sets a new standard for the integration of aesthetics in robotic design, paving the way for more intuitive and emotionally engaging human-robot interfaces.