About the role
The 3D Vision group trains models that infer geometry and semantics from lidar, camera and radar: 3D detection, occupancy and flow, scene completion and online mapping. We care less about leaderboard rank than about whether a model degrades gracefully, flags its own ignorance and still runs within the vehicle's latency budget.
The fleet generates far more data than we can label, and the interesting cases are rare. Your research feeds directly into the perception stack and into the data engine that decides what to label next. Methods from Frontier AI on self-supervision, multimodal pretraining and evaluation are a regular source of ideas, and you will work with those researchers when a problem overlaps.
Research here ends in a shipped model or a documented negative result. You will own experiments from hypothesis to evaluation on held-out fleet scenarios, with ablations that others can reproduce.
What you’ll do
- Develop and train 3D perception models for detection, occupancy, motion forecasting or open-set recognition on multi-sensor fleet data.
- Study uncertainty estimation for geometric predictions, including ensembles, evidential methods and out-of-distribution detection, and measure how well it holds in the field.
- Design self-supervised and weakly supervised objectives that exploit temporal consistency and cross-sensor agreement to reduce labelling cost.
- Run careful ablations and failure analyses, and write up findings so the perception engineers can adopt or reject them quickly.
- Distil and quantise large models to meet vehicle latency targets without hiding a regression in the long tail.
- Contribute to evaluation suites that measure performance on rare, safety-relevant scenarios rather than average accuracy.
What you’ll bring
- A research record in 3D vision, point cloud learning, neural scene representations or related areas, shown through papers, code or shipped systems.
- Strong PyTorch skills, including distributed training, mixed precision and debugging data pipelines that mix lidar, images and calibration.
- Solid grounding in multi-view geometry, rigid transforms, camera models and the maths of sparse convolution or transformer attention over point sets.
- Experience designing evaluation protocols that expose failure, not just report a mean, including calibration and selective prediction.
- The ability to write clear C++ or Python that engineers on other teams can read and run.
- Comfort disagreeing with a result, including your own, and saying so in writing.
Nice to have
- Experience with neural radiance fields, Gaussian splatting or other differentiable rendering applied to driving scenes.
- Work on test-time adaptation or domain shift between cities, seasons and sensor units.
- Familiarity with TensorRT or similar toolchains for deploying 3D networks on embedded accelerators.
- Experience with foundation-model features for labelling or retrieval of driving scenes.
Work you might take on
- 01
Train an occupancy network from lidar and camera that predicts free space behind occluders, and evaluate it on near-miss scenarios from fleet logs.
- 02
Add an out-of-distribution head to the 3D detector and show it fires on unseen object types such as debris and unusual loads.
- 03
Pretrain a point cloud encoder with temporal self-supervision on unlabelled fleet drives and measure the label efficiency gain downstream.
- 04
Quantise the main detector for the vehicle accelerator and publish an audit of which classes and ranges lost accuracy.
Compensation
$240K – $380K
The US base salary range for this role. Every offer also includes equity in Clexerl and, when it spins out, in the venture you help build. We tell you the reasoning behind your number.
Benefits
Health, fully covered
Medical, dental and vision for you and your dependants, premiums paid in full.
Two kinds of equity
Equity in Clexerl, plus a stake in the venture you help build when it spins out.
22 weeks parental leave
Fully paid, for every parent, however your family arrives.
Rest that sticks
Flexible time off with a four-week minimum we actually check on, plus a company-wide week in December.
Learning and compute
A yearly budget for courses, conferences and books, and compute credits for side projects.
401(k) with a 4% match
Matched from your first paycheque and vested immediately.
How we hire
Apply 5 minutes
Send links to what you have built and a few lines on why this role. No cover letter needed.
Intro call 30 minutes
A conversation with the hiring manager about the work, the team and what you want next.
Work sample 2–4 hours
A paid take-home or a pairing session on a real-shaped problem. You pick the format.
Studio day One day
Meet the team, go deep on your work and ours. Remote or in person.
Offer Within 5 days
A written offer with the full range, the equity and the reasoning behind both.
Opens an email with a short template. Links beat résumés.