Tendon-Driven Humanoid Hand

Problem

Most robotic hands either oversimplify motion with single-actuator underactuation or become overly complex with full actuation. Underactuated fingers can’t reproduce realistic human curling paths, while fully actuated fingers add unnecessary DOA and make coordinated control difficult. I needed a design that balances realistic biomechanics with practical control complexity.

Approach

I developed a hybrid architecture using two actuators per finger, capturing natural human joint coupling while avoiding the mechanical and control overhead of full actuation. I removed ab/adduction at this stage to prioritize stable, repeatable power-grip behavior.

The goal was to make it lightweight, mechanically efficient, and easier to control.
Three design iterations (V1→V3) progressively refined joint coupling, tendon routing, and thumb opposability.

Results

• Achieved ~75% fidelity to human finger curling motion using only 2 actuators/finger

• Improved thumb opposability and object-manipulation range through redesigned CMC geometry (V3)

• Increased force transmission and reduced friction through optimized tendon routing

• Demonstrated stable power-grip performance with simplified control relative to fully actuated hands