Coleman Farvolden

Research

Robotics, medicine and AI — wherever they meet.

Five areas, one habit: make the hard thing cheaper and reproducible.

Surgical robotics

The Perk Lab, Queen's University · 2024 · SPIE Medical Imaging 2025

Cancer can come back if tissue is left behind after surgery. I built low-cost, open-source robotic testbeds to scan the surgical cavity and follow it as it moves.

Positional accuracy 3.59 ± 1.4 mm · cavity-tracking latency 832.1 ± 132.6 ms

A 6-axis robotic arm holding an optical probe over a tissue phantom, with 3D Slicer showing the scan path.
Low-cost 6-axis arm, optical probe, 3D Slicer over ROS.
The two papers

An accessible 6-axis testbed for image-guided robotics research — first author. C. Farvolden, K. Hashtrudi-Zaad, L. Connolly, C. Barr, G. Fichtinger. A reusable 3D Slicer module for motion planning with an inexpensive arm, evaluated on a 16-point phantom. PDF

Robotic tracking of a resection cavity using a low-cost bench-top robotic arm and electromagnetics. K. Hashtrudi-Zaad, C. Farvolden, L. Connolly, C. Barr, G. Fichtinger. An electromagnetic sensor in a 3D-printed retractor lets the arm follow soft tissue across 36 test positions. PDF

Micro-scale manipulation

Ingenuity Labs, Small-Scale Robotics Lab · 2025 – present · NSERC CGS-M

Tumours are mechanically different from healthy tissue, and that shapes how cancer cells move. I'm building an open-source magnetic micromanipulation system to measure it.

3D-printable · fits any inverted microscope · maps 3D stiffness gradients in living cultures

Why it matters

Measuring stiffness at cell scale normally needs equipment most labs cannot justify buying. A cheap, reproducible instrument means labs without that budget can run the same experiments — and linking stiffness to cell behaviour should help explain how tumours form and spread.

Applied AI

Vector Scholarship in Artificial Intelligence · 2026

I work on AI that does useful work: agent systems, evaluation, and machine learning on real data.

What that includes

LLM agent systems and the evaluation harnesses that keep them honest; sequence models (LSTM and GRU) in PyTorch; and convolution kernels built from first principles to understand what a vision model is actually doing. No publications in this area yet — that is the next thing to change.

Autonomous systems

Robots that find their own way. LifeBot is a ROS2 rover that maps a building with SLAM and delivers an AED on its own; the autonomous loader follows a course, scoops and sorts material from sensor cues.

See the builds →

Medical devices & simulation

NeuroNeck is a sensing neck brace that gives first responders instant feedback while they practise immobilising a spinal injury. It won first place of 80 teams at the NextGen Medical Simulation Hackathon.

See the builds →

How I work

  • Baseline firstKnow what you are comparing against before building.
  • Report the spreadA number without its variance is half a result.
  • State the limitsSay plainly what the system cannot do.
  • Publish itA method nobody can reproduce is not much of a method.