I set out to build a tendon-driven humanoid hand (from scratch without copying any existing designs) that balances human-like motion with practical control complexity. Fully underactuated fingers with a single actuator were too limited to reproduce realistic human curling paths and joint sequencing, while fully actuated fingers introduced unnecessary degrees of actuation and made coordinated control overly complex.
To address this, I designed a hybrid architecture using two actuators per finger, which captures the natural coupled curling motion of the human hand while remaining mechanically efficient and easier to control. At this stage, I removed finger abduction and adduction to focus on stable grasping performance and consistent power-grip behavior. This approach allowed me to achieve 65% human-like motion fidelity without the mechanical overhead of full actuation or the oversimplification of single-degree underactuation.
Even though following someone else's design would greatly accelerate the pace of this project, I decided to do everything from scratch (tendon physics and paths, joint coupling rations and ranges, degrees of freedom for optimized control, etc.) to learn all the design methodology from a first-principles approach.
Humanoid Hand V2 - 15 Degrees of Freedom
Researched human hand biomechanics and designed a 15-DOF tendon-driven humanoid hand that replicates natural joint coupling and human-like range of motion.
Modeled joint coupling and ROM in Simulink, and verified tendon-driven actuation and control dynamics through a MuJoCo simulation of the hand.
Optimized tendon routing and actuator placement through iterative prototyping to improve force transmission efficiency and reduce friction.
Developed a hybrid tendon control strategy that preserves realistic finger motion while reducing complexity compared to fully actuated designs.
Writing a research paper documenting the control methodology and comparing underactuated and coupled-actuation approaches. (Manuscript in progress)
Humanoid Hand V3 - 16 DOF: Added 1 more joint to the thumb and repositioned the carpometacarpal joint to increase opposability and grasping range for object manipulation
MATLAB Simulink Simscape Multibody block diagram simulating joint ranges and limits, validating 65% human motion capabilities