Role Purpose
Covers the full perception stack: sensor calibration and multi-sensor processing, and the vision models that run on the robot and in our data pipeline.
A hands-on role on a small team. The exact scope is shaped around your strengths at onboarding, with room to grow into technical ownership of the stack.
Key Responsibilities
- Sensor calibration across depth, 3D LiDAR, RGB, ToF and sonar, including on-field calibration.
- Multi-sensor processing and fusion, and keeping the pointcloud pipeline stable and inside its CPU budget.
- Onboard models: detection and segmentation, including zone proposals on the map.
- Offboard models: auto-labelling, data mining and synthetic data generation.
- Research into newer models: vision-language models, monocular depth, and 3D reconstruction / world modelling for simulation and synthetic data.
- Data pipeline and model lifecycle: dataset and model versioning, experiment tracking, and gated model OTA.
- Automated testing: benchmarks and software-in-the-loop regression as a release gate.
Required Qualifications
- Bachelor’s or Master’s degree in Computer Science, Electrical Engineering, Robotics or a related field.
- Minimum 4 years in perception or computer vision for a real product, with strong Python and PyTorch, plus C++ for onboard code.
- Required: hands-on experience on both sides – sensor processing (calibration, pointclouds, depth or LiDAR) and training vision models on your own collected data. One side only is not enough for this role.
- Deployed models to embedded compute (OpenVINO, TensorRT, ONNX, RKNN, Hailo or similar). ROS 2 experience is a strong plus.