Robotic Finger Mechanism · Final Project Webpage

The Surgeon

A tendon-actuated anthropomorphic finger prototype for high-fidelity, backdrivable dexterous manipulation

Aileen Cleary Nicholas Melo Andrew Pavlovic Julia Xu Zach Xu

[ Northwestern University ]  ·  [ Robot Design Studio ]  ·  [ 2026 ]

Overview CAD Demos Code

01 — Overview

Project Goals & Motivation

Our goal is to develop a tendon-actuated anthropomorphic robotic finger capable of high-fidelity force control, low backdrive torque, and human-scale range of motion (ROM). The design prioritizes precise position control suitable for potential surgical and dexterous manipulation applications, using a 7-tendon architecture with antagonistic tendons on the MCP and PIP to achieve independent stiffness control at those joints.

Design Philosophy

  • Anthropomorphic kinematics matching human ROM
  • Tendon-actuated for lightweight, high-stiffness transmission
  • Antagonistic MCP/PIP tendon routing for independent joint stiffness control
  • Low friction throughout: ball bearings, idler pulleys,
  • Backdrivable by design for safe human interaction
  • High resolution motor and joint encoders for accurate tracking

Key Challenges

  • Achieving <0.03 Nm backdrive torque across full ROM
  • Tendon routing without bowstringing, tendon popping out, or becoming slack
  • Miniature sensing within 30 mm width envelope
  • Assembly complexity with multi-shaft, multi-pulley, multi-sensor architecture
  • Complicated electronics architecture

Actuation Architecture

Design utilizes remotized tendon driven actuation, with a 2N configuration for joint flexion and extension (antagonist tendon pair for both MCP and PIP), an internal tendon pair coupling the DIP and PIP, and an N configuration for splay (MCP).

Sensing Architecture

Custom-designed PCB with axial-mounted absolute magnetic angle encoders at each joint, 16-bit motor encoders, and a piezoelectric fingertip force sensor. Target resolution of 0.5° per joint matches human finger proprioception.

02 — Performance

Specifications: Target vs. Achieved

Parameter Target Specification Achieved / Current Status Notes
Continuous Fingertip Force 20 N — 19.7 N Near Miss This number recorded using Tony's Potentiometer. Can reach >20N (~22N) reading on our fingertip force sensor.
Joint Speed 1 s / ROM — pending test In Progress Baseline project standard
MCP Flexion ROM 90° 90° Met Mechanical stops prevent overextension
PIP Flexion ROM 90° 90° Met Tendon-coupled to DIP
DIP Flexion ROM 90° 90° Met Red coupling tendon
MCP Splay ROM 20° 30° total (±15°) Exceeded Limited by tendon pop-out beyond ±15°; no mechanical stop until +45°/-90°
Finger Length ≤150 mm 105 mm (165 mm overall) Met ~1.5× average human finger
Finger Width ≤30 mm 34.4 mm max Near Miss Increased width over expectation due to joint encoder magnet mounting changes.
Backdrive Torque 0.03 Nm — pending test In Progress Target: 30 mm × 1 N per Ornelas et al.
Angular Resolution 0.5° — Can settle around 0.3° degrees when calibrated Met Matches human finger precision per Feng et al.

03 — Design

CAD, Schematics & Assembly

All mechanical design was completed in OnShape. Below are key views including the exploded assembly, tendon routing diagram, friction reduction features, and sensing architecture.

Assembly View

Exploded CAD view showing phalanx links and joint encoders.

Tendon Routing

Seven-tendon color-coded tendon routing: MCP Splay (black), MCP Flexor (yellow), MCP Extensor (purple), PIP Extensor (blue), PIP Flexor (green), PIP-DIP Internal Coupling Pair (red and orange).

Joint Encoders

Joint encoder CAD assembly view.

MCP Assembly

MCP CAD joint exploded assembly view.

MCP and Finger Base

MCP and finger base plate CAD view.

PIP

PIP CAD exploded assembly view.

DIP

DIP CAD exploded assembly view.

04 — Bill of Materials

Components & Procurement

# Component Specification / Part # Qty Unit Cost Total Supplier Notes
Structural
1 Phalanx Links + MCP Base Aluminum 6061 1 1 $375.95 Outsourced + Custom machined JLCCNC, CNC milled; bearing seats reamed to size
Motors + Encoders + Drivers
2 Brushless DC Motors [DB42S01] 5 1 $72.90 Nanotec Rated Torque 50.13 Nmm
3 Planetary Motor Gearbox GP42-S2-26-SR 5 1 $203.80 Nanotec Reduction Ratio 25.62
4 Motor Encoder NME3-SSI-X14-12-C2 5 1 $136.50 Nanotec SSI 16-bit resolution
5 Controllers + Drivers ODrive Pros 5 1 $229.00 ODdrive Robotics Motor Controllers
Bearings & Shafts
6 Ball Bearings [ Part # — TBD ] 22 22 $275.88 McMaster-Carr Bearings Numbers: 104, 106, 685
7 Shafts 12L14Carbon steel 3 ft. 3 ft. $13.32 McMaster-Carr Pre Ground Stock, Machined into stepped shafts or tendon termination shafts
8 E-Rings E-Rings 100 1 $3.43 McMaster-Carr Axial shaft constraint
9 Carbon Steel Shim Stock Shims 1 150 x 2500 x .5 (mm) $30.93 McMaster-Carr Machined for shims for spacers
Tendons & Termination
10 Dyneema (UHMWPE) Cord .8 mm dia, 250 lb tensile 100 ft. 1 ft. $0.09 Emma's Kites - Amazon <1% stretch at working load
11 Vented Bolt (Coupling Tendon Tensioner) Vented Bolt 1 2 $17.15 McMaster-Carr Mechanical tightening to account for hysteresis
Pulleys
12 Pulley Stock Aluminum — custom machined 22 1 $0.00 Custom Idlers and Fixed pulley with groove to guide tendons. Material Provided by Ford Shop
Sensing & Electronics
13 Magnetic Angle Encoder Chips MA782 4 1 $3.73 Monolithic Power Systems Axially-mounted; 0.5° resolution target
14 Press-Fit Magnets [ N45 6 X 3 mm. Magnets] 4 1 $3.29 McMaster-Carr Housed within stepped shafts and a 3D printed holder
15 Custom PCB Custom Board for MA782 Sensors 1 4 $0.30 Custom / JLCPCB Joint Encoder signal routing and serialization
16 Fingertip Force Sensor A101 Tekscan 1 1 $9.44 Tekscan - Supplier: Digikey Piezoelectric Fingertip Force Sensor
17 DCDC Module DCDC Converter 1 1 $6.16 Digikey Changes output of (something) into somehing

* Some quantities and costs not shown, see link to complete BOM in RDS Drive. Materials 3D Printed from Northwestern's labs not included.

05 — Demonstrations

Demo Videos

Team Demos

Drawing Demo - Letter L

Trace trajectory of the letter L on the paper using both position and force control.

Drawing Demo - Letter S

Trace trajectory of the letter S on the paper using both position and force control.

Shading Demo

Shade in a square from dark to light on the paper using both position and force control.

Simulation Demos

Gravity + Feedforward Control

Using Drake, the finger takes a desired joint trajectory, calculates the required PD torques with gravity compensation, and uses linear programming to optimally distribute the forces into actual motor slack length commands.

Simulation Demos

Simulation Writing

Using Drake, we use PD Control to follow a trajectory generated that resembles a letter. The workspace in the simulation is larger than reality since the robot can move latterally, so we are able to generate sentences and larger images.

Simulation Demos

Simulation Shading

Using Drake, we use PD Control to follow a trajectory generated that resembles a letter. The workspace in the simulation is larger than reality since the robot can move latterally, so we are able to generate sentences and larger images.

Simulation Demos

Simulation Drawing

Using Drake, we use PD Control to follow a trajectory generated that resembles a letter. The workspace in the simulation is larger than reality since the robot can move latterally, so we are able to generate sentences and larger images.

06 — Experimental Results

Tests & Measurements

All tests listed below are either completed or planned. Plots, images, and videos will be added as results are collected.

Trajectory Tracking

Test 01

Trajectory Tracking Test

Measure the finger ability to trace a Lissajous figure on the main plane of motion x-z (no splay) bounded by a 30mm x 30mm region with a frequency of .5 Hz.

Force Control

Test 02

Fingertip Force Output

Measure continuous fingertip normal force at full tendon tension. Target: ≥ 20 N. Setup: load cell at distal phalanx tip, motors at rated current. Peak Force 19.7 N

Force Control

Test 03

Step Force Control Test

Track a sqaure wave reference force alternating between 1-3 N. at .5 hz. Average Overshoot: 13.64%, Average Steady State Error: .29 N, Average Rise Time: .12 s.

Impedance Test at Fully Extended Finger Position

Test 04

Fingertip Impedance Test (Fully Extended)

With zero control system, externally apply a damping chirp signal up to 50 Hz. From .5 hz. to 20 hz. such that coherence is > .9 Apparent interia = .0728 g, Apparent damping = 20.34 Ns/m, Apparent stiffness = 1687.7 N/m

Impedance Test at Half Extended Finger Position

Test 05

Fingertip Impedance Test (45 degree Flexed)

With zero control system, externally apply a damping chirp signal up to 50 Hz. From 1 hz. to 20 hz. such that coherence is > .9. Apparent interia = .0556 g, Apparent damping = 8.89 Ns/m, Apparent stiffness = 1145.7 N/m

Impedance Test at Fully Flexed Finger Position

Test 06

Fingertip Impedance Test (Fully Flexed)

With zero control system, externally apply a damping chirp signal up to 50 Hz. From .9 hz. to 35 hz. such that coherence is > .9. Apparent interia = .0402 g, Apparent damping = 4.90 Ns/m, Apparent stiffness = 1227.3 N/m

07 — Software

Code

Code

View on GitHub

08 — The Team

Team Surgeon

AC

Aileen Cleary

[ Kinematics Lead ]

NM

Nicholas Melo

[ Design / Manufacturing ]

AP

Andrew Pavlovic

[ Design / Manufacturing]

JX

Julia Xu

[ Electronics / Software ]

ZX

Zach Xu

[ Electronics / Software ]

logo

logo

09 — Acknowledgments

Acknowledgments

We are grateful to the following individuals and organizations for their support, guidance, and resources throughout this project.

[ Ed Colgate, Raphael Cherney, Sairam Umakanth ]

Course instruction, design guidance, and technical feedback throughout the design review process.

[ Shop Staff at Ford ]

Manufacturing support, machining guidance, and access to equipment for fabrication.

Northwestern University, CRB, HAND

Funding, laboratory space, and equipment access enabling this research.

[ Other Teams / Collaborators ]

Powerhourse, Speedster, Minimalist Teams for peer feedback during design reviews.
Anthony Shilati for allowing data collection on his testbed.
Soloman from ODrive Support for providing us with the firmware to use the oDrives with our enocoders

10 — References

References

  1. Feng, Yongfei, et al. "Research on Monocular-Vision-Based Finger-Joint-Angle-Measurement System." Sensors, vol. 22, no. 19, 2022. doi:10.3390/s22197276
  2. Ornelas, Ruben C., et al. "Everyday finger: a robotic finger that meets the needs of everyday interactive manipulation." 2024 IEEE ICRA, pp. 16016–16023. doi:10.1109/ICRA57147.2024.10611452
  3. [ Additional references — add as needed ]

Team Surgeon  ·  Aileen Cleary · Nicholas Melo · Andrew Pavlovic · Julia Xu · Zach Xu

[ Robot Design Studio ]  ·  [ 2026 ]

Built using the Academic Project Page Template by Eliahu Horwitz, adopted from the Nerfies project page. Licensed under CC BY-SA 4.0.