Robot Design Studio: Powerhouse Finger

Project Summary

Our capstone team designed a 3-DOF robotic finger focusing on high strength and durability, all without sacrificing practical performance. The final system combines antagonistic tendon routing, coupled finger kinematics, custom pulleys and encoders, and a modular testbed to support robust bench testing and iterative prototyping.

In software and firmware, we built a full stack spanning simulation, controls, teleoperation, and safety-limited embedded motor control. The resulting platform demonstrates repeatable high-load behavior and provides a strong base for future sensing and manipulation tasks.

Michael loses a contest of strength against the Powerhouse finger.

Alternate view of the pencil break demonstration.

Finger setup at final design showcase.

Team Powerhouse running a demo at the final design showcase.

Hardware Description

Finger mounted on custom modular testbed.

Side profile view showing resting configuration.

Design iterations used to improve force and durability.

Tendon crossover point on double groove pulley.

Modular fingertip with integrated compliant surface.

Parallel tensioning springs reduce slack and ease assembly.

Custom motor pulleys for high torque transmission.

Alternative finger configuration for testing setup.

Additional pulley stands for upright configuration.

Test Results

The validation process included impedance response tests, force step tests, and power-focused load tests to verify controller behavior, force capacity, and repeatability against project requirements.

• In the max fingertip force test, the force is gradually increased until the finger fails.
• In the impedance test, a dynamometer applies a chirp signal of increasing frequency to the fingertip while the finger commands zero torque, and the resulting force and position data are used to estimate impedance characteristics.
• In the position step test, the finger is commanded to oscillate between two positions that are 10 mm apart at 1 Hz, and the steady state error, overshoot, and settling time are calculated across all step responses.
• In the Lissajous trajectory test, the finger is commanded to follow a Lissajous curve in the XY plane at 0.25 Hz, and the integrated position error is calculated across the trajectory.
• In the power test, the finger is commanded in a force step trajectory while the fingertip is attached to a load cell, spring, and damper in series.. Using the fingertip position data from computer vision and the force data from the load cell, the mechanical power can be calculated.

The finger generally performed very well in tests of strength and basic position control, with the max fingertip force far exceeding our design goal and the position step response showing minimal steady state error and settling time. It struggled in the Lissajous trajectory test, which we attribute to a combination of unmodeled dynamics/friction, improper control tuning, and some tendon slack that caused delayed tension delivery. The differential tendon design was a major contributor to the high force output, but it also made the system very sensitive to tuning. The additional noise of the custom encoders also made it difficult to add any derivative gains to the system.

The power output test also had some issues, particularly with the velocity data. We used computer vision to record fingertip position during the test, with the intention of using a finite difference method to calculate instantaneous velocity along the recorded trajectory. However, the finger often moved too quickly for the camera to capture a smooth trajectory, which made the velocity and therefore power calculations very noisy. Despite this, general trends are still present. The power output (negative) increases in magnitude when the force and velocity increase (finger presses down). It also approaches zero when the releases the damper, sometimes even becoming positive. It is possible the positive power is caused by the spring pressing up against the finger when the downward force is released. We also were not able to reach the maximum fingertip speed while constraining the finger in the power output testing setup, meaning the power output is potentiall much lower than the maximum possible. Overall, we are very happy with the performance of the finger, particularly in terms of strength and repeatability, but there are still many opportunities for improvement in control and sensing to enable better performance in dynamic tasks.

Team Photo and Acknowledgements

CAD File: https://cad.onshape.com/documents/9008e8e00ba75713bd9082d6 Thank you to Dr. Colgate, Raphael, and Sairam Umakanth for their guidance throughout the entire design process. We also thank shop faculty, guest reviewers, and Tony Shilati, all for assisting us with design, manufacturing, and testing on their personal time. Lastly, thank you to our peers in RDS for their helpful feedback and support along the way!