SoftHand Model-W: A 3D-Printed, Anthropomorphic, Underactuated Robot Hand with Integrated Wrist and Carpal Tunnel

1. Problem Statement

While anthropomorphic robot hands have advanced significantly, a critical gap remains in replicating full human manipulation capabilities. Most existing designs focus on increasing finger actuation while leaving the wrist either fixed or passive.

• The Limitation: The human wrist contributes three additional Degrees of Freedom (DoF), which are essential for efficient in-hand manipulation and reaching diverse orientations. Without an active wrist, robot hands mounted on arms must perform large, inefficient compensatory movements to achieve desired poses. This is impractical in confined spaces and increases the risk of singularities or collisions.
• The Challenge: Integrating an active wrist into an underactuated, tendon-driven hand (like the Pisa/IIT SoftHand) is difficult because it risks coupling wrist motion with finger motion (e.g., wrist extension causing unintended finger flexion) and increases distal inertia if motors are placed at the hand.

2. Methodology and Design

The authors propose the SoftHand Model-W, a 3D-printed, underactuated, tendon-driven robot hand featuring an integrated 2-DoF active wrist and a biomimetic carpal tunnel.

A. Mechanical Design

• Underactuated Fingers: Based on the BPI SoftHand-A design, the hand features five fingers with 15 DoFs total. It uses antagonistic tendon pairs (flexor and extensor) driven by a single motor per pair, preserving the "adaptive synergy" that allows the hand to self-adapt to object shapes.
• 2-DoF Wrist: The wrist provides flexion/extension and radial/ulnar deviation. It uses a serial rotational-rotational (RR) configuration.
  • Note: Pronation/supination is omitted as it is typically handled by the robot arm's final joint.
• Carpal Tunnel & Tendon Routing: A key innovation is the "carpal tunnel" design. Finger and wrist tendons are routed through central channels in the wrist using PTFE tendon sheaths (similar to Bowden cables).
  • Function: This decouples finger motion from wrist motion. The sheaths bend with the wrist without changing length, allowing the tendons to slide freely. This prevents the wrist's rotation from inducing unwanted finger flexion/extension.
• Remote Actuation: All four servos (two for fingers, two for wrist) are mounted in a "forearm" box proximal to the hand. This reduces distal inertia and maintains a compact form factor.
• Construction: The hand is primarily 3D-printed in PLA (with resin-printed high-precision components), using aluminum links, bearings, and standard M2 hardware. It is human-scale (164.6 mm height), fitting within the 5th–95th percentile of adult hand dimensions.

B. Control System

• Actuation: Four Feetech STS3215 servos drive the system.
• Finger Control: Uses open/close commands based on calibrated positions. Tensioning is managed via M2 bolts and washers rather than knots to ensure uniform tension across the antagonistic tendons.
• Wrist Control: Utilizes an open-loop proportional controller. Because the wrist cams and spools have a constant radius, there is a linear relationship between servo position and wrist angle.
• Kinematics: The system accounts for the changing Tool Center Point (TCP) caused by wrist movement by updating forward kinematics in real-time.

C. Experimental Setup

The SoftHand-W was mounted on a UR5 6-axis industrial robot arm. Two manipulation tasks were performed to evaluate performance with and without wrist actuation:

1. In-Hand Rotation: Rotating a 90mm disc by 90°.
2. Cube Stacking: Stacking six cubes, requiring specific reorientations (yaw, roll, and pitch) for each layer.

3. Key Contributions

1. Miniaturized Anthropomorphic Design: A low-cost, 3D-printed version of the BRL SoftHand that matches human dimensions, making it accessible for research and assistive applications.
2. Integrated 2-DoF Wrist with Carpal Tunnel: The first integration of an active wrist into this specific underactuated lineage, featuring a novel tendon routing system that decouples finger and wrist actuation while allowing remote motor placement.
3. Validation of Wrist Utility: Empirical demonstration that adding proximal DoFs (the wrist) significantly reduces the need for compensatory arm movements, improving task efficiency and success rates in reorientation tasks.

4. Results

The experiments compared the performance of the hand with and without wrist actuation.

• Rotation Task: The wrist allowed the hand to rotate the disc continuously. Without the wrist, the arm had to release and re-grasp the object to avoid joint limits, doubling the configuration changes and time.
• Stacking Task: Without the wrist, the arm had to make extreme, large-arc movements to pitch cubes upward, leading to a failed placement of the 5th cube. With the wrist, the hand could pitch the object directly, resulting in a 36% reduction in maximum joint travel for the arm's pitch/yaw joints (Joints 3 and 4).
• Decoupling: The PTFE sheaths successfully prevented involuntary finger motion during extreme wrist poses.

5. Significance and Conclusion

The SoftHand Model-W demonstrates that proximal degrees of freedom (the wrist) are critical for achieving versatile, human-like manipulation in real-world robotic applications.

• Efficiency: By offloading orientation tasks to the wrist, the robot arm performs fewer compensatory movements, reducing energy consumption and task time.
• Feasibility: The design proves that complex, dexterous manipulation can be achieved with a simple, low-cost, 4-servo control system and a compact, 3D-printable form factor.
• Future Impact: This work provides a reproducible benchmark for accessible dexterous robotics, paving the way for better integration of robots in unstructured environments, service industries, and assistive settings. The authors suggest future improvements could include an actuated thumb, individual tendon tensioning, and tactile sensing.