Perception & sensor fusion
Cameras, depth, LiDAR and IMUs that agree with each other and hold up in dust, glare and darkness. Calibration, fusion, detection and segmentation that survive the field.
services
From prototype to field
I help teams put perception and robot learning onto real machines: heavy equipment, manipulators, mobile robots and edge devices. Every engagement starts with a measured baseline and ends with something your team owns.
what I help with
Robot learning
Imitation learning and reinforcement learning, from teleoperated demonstrations to policies that run on the robot: data capture, behaviour cloning with ACT, GR00T and π0, RL fine-tuning and sim-to-real in Isaac Lab, and touch where contact matters.
Simulation & synthetic data
Isaac Sim, Isaac Lab and MuJoCo twins of your machines and sites, for sensor placement, policy evaluation and data you cannot easily collect.
Edge deployment & MLOps
Models that fit the device: quantisation, TensorRT and TFLite, Jetson and phones, Triton serving, and the cloud pipeline that trains, versions and ships them.
engagements
Technical audit
2 weeksAn independent look at your perception or learning stack against real field data.
you get
- Ranked failure modes with root causes
- Latency and compute profile on target hardware
- A costed, prioritised fix plan
- Written report and debrief with your team
Build sprint
4 to 8 weeksI join your team to ship one defined capability, end to end, on your hardware.
you get
- A success metric agreed on day one
- Weekly demos on the real machine
- Code in your repository, reviewed
- Handover docs your team can own
Advisory
MonthlyA senior second pair of eyes for teams moving from prototype to product.
you get
- Architecture and design reviews
- Two calls a month, plus async review
- Hiring loops and technical interviews
- Roadmap and build-vs-buy input
how it works
- 01
Discover
A 30-minute call about the machine, the site and the one metric that matters.
- 02
Measure
Baseline it with your real logs and hardware before proposing anything.
- 03
Build
Small, reviewed increments that run on the target device, demoed weekly.
- 04
Hand over
Prove it in the field, then hand over code and docs your team owns.
results from past work
questions
What kinds of teams do you work with?
Startups and established companies putting robots, machines or vision systems into real, unstructured environments: construction and mining equipment, logistics, retail, manufacturing and teleoperation.
Do you work remotely?
Yes. I am based in London and work remote-first, with on-site visits for integration and field testing when the work needs it.
Who owns the work?
You do. Code, models, data and documentation produced in an engagement belong to your company and live in your repositories.
Can you work under NDA?
Yes. Most work is confidential. Anything shown on this site is published, public, or shared with permission.
How do we start?
Send a short note about the machine and the problem. If it looks like a fit, we have a 30-minute call, and I follow up with a written proposal and a fixed scope.