Before the detail, here's the challenge you'd help us solve.
We build the embodied intelligence that moves real vehicles safely, and the ecosystem a billion machines will run on in the future.
Very few people in AI can say this.
Every role here, whatever the team, plugs into that.
Here’s what this particular role covers.
We work end-to-end, from creating high-quality training data, to developing and evaluating CV/3D perception models, to iterating quickly based on performance gaps.
The team mixes "online" (on-car, latency/compute constrained) and "offline" (heavier, large-scale data generation) work, with a strong focus on measurable impact and shipping.
🧠 Your day-to-dayTrain, debug, and improve computer vision and 3D perception models, iterating based on clear evaluation signalsWork across the full ML lifecycle: data, training, evaluation, and iterationPartner with the team to decide what to tackle next, based on where the system is underperformingBuild scalable data pipelines, including auto-labelling and pseudo-labelling, to accelerate model development 🧩 What you'll be working on:Core ADAS perception capabilities such as detection, classification, and instance segmentation, spanning lanes, objects, traffic signs, and traffic lightsEnd-to-end approaches that let the model act on the world directly, rather than relying on a chain of hand-coded intermediate steps (lane assist without ever explicitly detecting lane lines is the kind of outcome we're after)Online perception that runs fast in-car as part of our production stackOffline perception work using additional inputs not viable for on-road production, including extra sensors like lidar, future frames, maps, and extra compute, to improve data quality and coverage at scaleBuilding the "data flywheel": using offline models and 3D tracking/reconstruction to auto-generate la.