Educational SoftHand-A: LEGO Robotic Hand
- Educational SoftHand-A is an anthropomorphic LEGO-built robotic hand that uses tendon underactuation and soft synergies to achieve adaptive grasping.
- It employs a dual-motor antagonistic design with clutch gears, enabling coordinated global motion and reactive fine control.
- The system is designed for educational use, offering a low-cost, easily assembled platform for hands-on exploration of modern robotic hand concepts.
Educational SoftHand-A is an anthropomorphic robot hand built entirely using LEGO MINDSTORMS and conceived as an educational implementation of a modern tendon-driven, highly underactuated robotic hand. It is directly based on the Pisa/IIT SoftHand design philosophy of soft synergies, but reformulates that philosophy under explicit educational constraints: only standard LEGO pieces, low-cost and accessible construction, and testing with common equipment available at home or school. Mechanically, it combines anthropomorphic morphology, tendon underactuation, synergy-based coordinated motion, adaptive grasping, and a dual-motor agonist/antagonist architecture intended to provide reactive fine control while preserving a simple actuation and control mechanism (Lepora et al., 17 Oct 2025).
1. Concept, purpose, and design lineage
The hand was developed as an easy-to-assemble educational robot hand that exposes learners to the frontiers of robotic manipulation. Its stated motivation lies in two broader trends in robotic hands: simplification of anthropomorphic designs without losing functionality, and increasing use of compliance and underactuation for adaptive grasping. Within that context, the Educational SoftHand-A is presented as a pedagogical descendant of the Pisa/IIT SoftHand lineage, preserving the essential research ideas of soft synergies and adaptive underactuation while translating them into a buildable LEGO MINDSTORMS format (Lepora et al., 17 Oct 2025).
The lineage described for the design is explicit.
| Stage | Characterization | Distinguishing feature |
|---|---|---|
| Pisa/IIT SoftHand | One-motor anthropomorphic hand | Adaptive synergies, rigid phalanges, elastic joints/ligaments, common tendon transmission |
| BRL/Pisa/IIT 3D-printed SoftHand | Lower-cost reinterpretation | Simplified joints and added soft synergies using springs |
| SoftHand-A | Two-motor extension | Agonist/antagonist tendon pair on each finger |
| Educational SoftHand-A | LEGO MINDSTORMS educational version | Standard LEGO pieces and household-style testing |
Relative to the original Pisa/IIT SoftHand, the Educational SoftHand-A preserves anthropomorphic morphology, high underactuation, and synergy-based coordinated closure, but replaces elastic-joint/passive reopening with a dual-motor antagonistic tendon drive. It also replaces spring-based soft synergy elements from earlier 3D-printed designs with LEGO clutch gears. The educational constraint is therefore not peripheral; it is the organizing principle that determines embodiment, manufacturability, and evaluation.
A common misunderstanding is to read “soft” as a statement about soft-material construction. The hand is instead built entirely from standard LEGO pieces, while its “soft” character is realized through soft synergies implemented mechanically by a differential mechanism and clutch gears. This suggests that the term primarily denotes adaptive transmission and coordinated compliance rather than a soft-bodied morphology.
2. Mechanical architecture and embodiment
The Educational SoftHand-A has four digits: index, middle, pinkie, and thumb. This is anthropomorphic but simplified relative to a five-finger human hand. The authors explicitly chose four digits to simplify construction and improve aesthetics without greatly affecting function. Each digit has three rotary joints corresponding to metacarpophalangeal (MCP), proximal interphalangeal (PIP), and distal interphalangeal (DIP), yielding 12 joint degrees of freedom driven by only 2 degrees of actuation. The thumb is further simplified by replacing the human carpometacarpal joint with a standard MCP-like arrangement to reduce design and control complexity (Lepora et al., 17 Oct 2025).
All fingers are modular and identical in design, which is central to educational usability. Each finger consists of a base phalanx, two medial phalanges, and a distal phalanx. Each finger is long from base to fingertip and wide. Internally, each finger contains 22 axle rods, 20 bearings guiding tendons, and two tendons, one on each side. The phalanges are held together by left and right phalanx covers, and joints are formed using connecting beams and axle rods that enforce rotational motion. Hyperextension is limited by mechanical stops on the distal side of each joint.
The hand is fully tendon driven. Across the full hand there are 8 tendons total: 4 agonist tendons for closing and 4 antagonist tendons for opening. Each tendon starts from a spool in the drive unit and ends at a terminating beam in a fingertip. Agonist tendons pass down the front of the palm and along the left side of the fingers, whereas antagonist tendons pass down the back of the palm and along the right side. Each finger therefore receives an agonist/antagonist pair analogous to biological flexor/extensor action.
The tendon routing through the medial phalanges follows the SoftHand tradition of using guide points to induce rotational torques as the tendon path changes around each joint. Each medial phalanx has three grooved bearings per side. On one side they form an upward isosceles triangle to favor flexion via the driving tendon; on the other side they form an inverted isosceles triangle to support extension. The distal phalanx uses a reduced routing arrangement, with only one bearing pair leading to terminating beams at the fingertip. A rubber beam at the fingertip acts as a high-friction finger pad.
The base phalanx anchors each finger to the palm and directs tendon entry and exit into the palm. The palm itself has three roles: holding the four digits in an anthropomorphic arrangement, supporting bearings for tendon routing from fingers into the base, and providing structural support during grasping. The fingers and palm are purely mechanical, while the actuation hardware is placed below the wrist in a separate base unit.
3. Soft synergies, differential coupling, and antagonistic actuation
The hand’s synergy mechanism is realized through a differential drive unit. Each motor drives four spools—one per finger tendon in that motor’s tendon group—through a common axle and gear train. The four agonist tendon spools are placed at the front of the drive unit and the four antagonist tendon spools at the back. One motor command therefore broadly closes or opens all fingers together, while contact-dependent deviations are enabled mechanically (Lepora et al., 17 Oct 2025).
The key enabling element is the clutch gear. The clutch gears have a maximum torque of before disengaging, whereas the motors can output up to . This establishes limited torque transmission to each spool. If a finger moves freely, its spool turns normally; if that finger encounters an object and the required torque rises above the clutch threshold, the clutch slips or disengages, while the other fingers continue moving. In this way the hand implements synchronized global motion with local adaptability. In the authors’ framing, the clutch gears functionally replace the spring-based soft synergy elements used in earlier 3D-printed SoftHand variants.
The Educational SoftHand-A differs from the original one-motor SoftHand in its use of two motors: one agonist motor to close the hand and one antagonist motor to open it. This dual-motor arrangement underpins the claim of reactive fine control. In many underactuated tendon-driven hands, reopening depends on passive springs or ligaments, so extension is not actively controlled. Here, the antagonist motor actively generates opening. The benefits emphasized for this arrangement are the absence of passive elastic ligaments that constantly oppose flexion, better force transmission, active opening as well as active closing, improved reactivity, and better controllability.
During grasping, the agonist motor closes the hand while the antagonist motor loosens its tendons “to synchronize motion and maintain low stiffness of the fingers.” This point is mechanically important. The architecture is not simply bidirectional drive; it is bidirectional drive combined with deliberate relaxation of the non-driving tendon group to preserve compliance and adaptive behavior. A plausible implication is that the hand can modulate responsiveness and apparent stiffness through relative tendon tension, even though the paper does not formalize this with a full stiffness model.
4. Control scheme and kinematic interpretation
The control approach is deliberately simple and educationally accessible. The actuators are LEGO MINDSTORMS EV3 Large Servo Motors (45502), controlled by a programmable EV3 brick (part 95646). Because the motors include encoders, rotation can be measured. The control scheme is mostly motor-command based rather than optimization-based or sensor-rich: to close the hand, the agonist motor reels in the agonist tendons while the antagonist motor loosens or allows release; to open the hand, the antagonist motor reels in the antagonist tendons while the agonist side releases (Lepora et al., 17 Oct 2025).
The design therefore places much of the “intelligence” in the transmission rather than in a high-dimensional controller. Adaptation to objects emerges from three coupled mechanisms: underactuated tendon coupling across fingers, clutch-based differential action, and low finger stiffness when the opposite tendon set is loosened. This is consistent with the underlying SoftHand philosophy that coordinated grasping can arise from mechanical organization and compliance rather than explicit independent finger control.
The paper is also explicit about what it does not provide. It does not present an advanced closed-loop controller, optimization-based grasp controller, tactile feedback law, explicit hand kinematic equations, tendon Jacobians, differential equations, optimization problems, or synergy matrices for the LEGO hand. No full mathematical model is given for joint angle to tendon displacement mapping, tendon force distribution, clutch slip dynamics, synergy subspace mapping, compliance matrix, or grasp force optimization. The result is conceptually grounded in soft synergy theory but positioned primarily as a mechanical implementation and educational platform rather than as a mathematical modeling study.
Several kinematic and compliance principles are nevertheless stated qualitatively. Underactuation follows directly from the mismatch between 12 joint DOFs and 2 actuators. Synergy means that motor commands define a low-dimensional global hand motion in which fingers tend to move together in a natural grasp-closing pattern. Adaptive synergy means that actual posture departs from nominal synchronized motion when some fingers contact an object earlier than others. Compliance arises from tendon looseness on the non-driving side and from clutch-mediated limited torque transfer. Antagonistic actuation provides bidirectional active control while preserving adaptive behavior.
5. Empirical evaluation and demonstrated capabilities
The evaluation is divided into performance tests and grasping adaptivity tests, both designed to be measurable in an educational setting with simple methods and common objects rather than laboratory instrumentation (Lepora et al., 17 Oct 2025).
| Test | Educational SoftHand-A | 3D-printed SoftHand-A |
|---|---|---|
| Single-finger closing time | ||
| Single-finger opening time | ||
| Single-finger bearing capacity | load | load |
| Single-finger pushing capacity | 0 weight | 1 weight |
| Single-finger closing force | 2 | 3 |
| Full-hand closure time | 4 | 5 |
| Full-hand opening time | 6 | 7 |
The paper interprets the lower speed and somewhat reduced strength of the LEGO version mainly as consequences of lower-quality motors, greater bulk, and higher tendon friction from plastic LEGO bearings instead of metal bearings. Even so, it concludes that performance is still good and entirely sufficient for educational use. The antagonistic tendon arrangement is singled out as preserving relatively strong force transmission despite modest actuator power.
The grasping adaptivity experiments mount the hand vertically and use the agonist motor for closure while loosening the antagonist side to keep stiffness low and permit adaptation. Household objects include a fan, plastic cup, bowl, tape, shoe, wheel, small ball, large ball, and soft toy, with weights ranging from 8--9. The reported outcome is qualitative rather than success-rate based. The hand successfully enclosed and held a range of objects with visibly different finger postures, and several objects were grasped in more than one way: the fan by handle or head, the cup by side or base, and the bowl from back or front. Within the paper’s framework, this variability is taken as evidence of adaptivity because finger configurations changed with shape, size, and contact pattern.
The evaluation therefore supports four specific conclusions advanced by the authors: grasp adaptability across a broad variety of object shapes and sizes, control simplicity through a two-motor actuation structure, decent robustness and force transmission despite all-LEGO construction, and strong educational value as a platform that makes advanced robotic-hand ideas physically accessible.
6. Educational role, limitations, and relation to adjacent research platforms
The most distinctive feature of Educational SoftHand-A is not merely that it is a robotic hand made from LEGO, but that it preserves recognizable research-level concepts—anthropomorphic structure, tendon underactuation, synergy-based coordinated motion, adaptive grasping, and antagonistic actuation—within the constraints of standard LEGO pieces and home or school testing. The educational context also shaped implementation choices: modular identical fingers to reduce design complexity, use of Bricklink Studio V3 to support reproducibility and sharing of build instructions, and experiments based on video timing, household weights, and everyday objects rather than specialized instrumentation (Lepora et al., 17 Oct 2025).
The paper is equally explicit about trade-offs. The hand has four digits rather than five because LEGO construction is bulkier. The 3D-printed SoftHand-A used about 150 metal bearings, while the LEGO version uses over 100 plastic LEGO bearings, increasing tendon friction. LEGO bricks make the structure bulkier than a custom-fabricated design. The use of clutch gears instead of springs preserves adaptive synchronization but introduces a trade-off: the lack of restoring tension can create tendon slack, which lowers reaction speed. The platform also remains limited in sensing, advanced control, and analytical modeling. These are not incidental omissions; they delineate the scope of the platform as an educational embodiment rather than a precision dexterous manipulator.
Within the broader SoftHand family, adjacent systems illustrate how the same underactuated philosophy can be extended when the educational accessibility constraint is relaxed. The “Tactile SoftHand-A” is a five-finger, 0-DOF, 1-DOA 3D-printed hand that retains the antagonistic tendon architecture while adding fully 3D-printed tactile fingertips and a human-hand-guided tactile feedback grasping system capable of contact-triggered stabilization and slip-triggered gesture adjustment (Li et al., 2024). A different branch of related work, “A Soft Humanoid Hand with In-Finger Visual Perception,” integrates five fingertip cameras, an FPGA-plus-microcontroller embedded architecture, and CNN-based visual object segmentation into a five-finger underactuated humanoid soft hand, reporting a cylindrical grasp force of 2, 3 grasp success on 60 YCB objects, and mechanical durability beyond 15,000 closing cycles (Hundhausen et al., 2020).
Seen against these adjacent platforms, Educational SoftHand-A occupies a specific position in the design space. It gives up tactile sensing, embedded fingertip vision, and research-grade fabrication in order to show that a hand built entirely from standard LEGO components can still instantiate soft synergies, antagonistic tendon routing, and adaptive grasping. This suggests that its principal significance lies in translation: it brings concepts that are usually studied in research laboratories into a form that can be assembled, inspected, and experimentally interrogated in educational environments without departing from the core logic of contemporary underactuated anthropomorphic hand design.
