A 3D-printed prosthetic hand with myoelectric control and force sensing
Commercial myoelectric hand prostheses cost tens of thousands of dollars. This project brought together the group's earlier developments in mechanics, force sensing and control into a single device and tested it with an amputee. The user learned to open and close the hand in about 15 minutes and could grasp and release objects at will. Manufacturing cost was around 330 US dollars. The prosthesis cannot yet be worn, and it closes slowly.
Context
The low-cost 3D-printed hands that circulate in open communities are usually body-powered: they close when the wrist or elbow flexes. Myoelectric hands are controlled by the electrical signal of the remaining muscles and allow a more natural grasp, at a price beyond the reach of most users in Colombia.
The project builds on earlier work in the lab: a first myoelectric prosthesis prototype, the mechanical redesign of the hand and a force sensor with grasp control.
Mechanics
Each finger has phalanges articulated with pulleys and closes by means of a tendon. Instead of one motor per finger, a floating-body mechanism shares the tension of a single actuator among several fingers. When one finger touches the object and stops, the others keep closing until they touch it too. The grasp thus adapts to the object's shape without per-finger sensing or control.
The floating-body mechanism distributes motion among the fingers according to the object
Force sensor
Tendon tension is measured with a deformable part made of U-shaped beams and an infrared optical sensor that reads its deflection. Several geometries were evaluated in simulation and the selected one was printed by stereolithography.
Design iterations of the sensor's deformable part
Sensor response against applied load. The response is linear above a load threshold
The characterization shows a dead zone at low loads and a linear response above it (R² = 0.99). That measurement closes the control loop: a PID controller stops the motor when the grasp force is reached and avoids overloading the mechanism.
Control loop of the prosthesis, with the user inside the loop
Electromyographic control
The signal is taken with surface electrodes on the forearm. The algorithm interprets the signal, from its instantaneous value and its average, as a command to open or to close, and responds to changes in the user's intention.
Five consecutive closures: EMG signal and interpreted state
Results
In the test with the amputee user, the prosthesis followed their intention to open and close and held objects in different orientations without dropping them. Closing time went from 7 s down to between 3 and 4 s after adjustments made between tests. A human hand closes in under a second, so the device is only useful for objects at rest.
All the electronics were integrated into a 2 × 7 × 5 cm board that fits inside the socket.
Board with the EMG, power, force sensor and control modules
What is missing
The user test was run with external electronics; the integrated board was not tested. The coupling between socket and residual limb was not designed, and it is specific to each person and determines whether a prosthesis is used or abandoned. Closing speed is limited by motor voltage and by the transmission reductions. And "force feedback" is for now internal to the controller: the vibration motor meant to inform the user is provided for on the board but was not evaluated.
How it fits in Robiolab
Prostheses and exoskeletons are one of the lab's founding lines, and this project shows its way of working: each thesis inherits the hardware of the previous one and tackles the next bottleneck. Underactuation through the floating mechanism is also an example of mechanical intelligence, in which adaptation to the object is solved by the mechanism rather than the controller.