Autonomy Droid Perception SWE - Onboard Systems
$200,000–$240,000 year
On-siteSouth San Francisco, California, United States
Job Summary
Implement, train, and evaluate real-time 3D perception models using multiple camera inputs across timesteps, then deploy and optimize these systems on resource-constrained onboard hardware to minimize compute and memory footprints. Build visualization and introspection tooling to assess model performance on test datasets and in the wild, while collaborating with the motion planning team to define compact interfaces and track system-level metrics. Ship production-grade features or new models to production every six weeks, driving experimentation on CNN and transformer architectures while maintaining alignment with state-of-the-art research in real-time compute-constrained 3D perception.
Required Qualifications
- At least 5+ years of experience (Senior) or 8+ years of experience (Staff) building and deploying deep learning-based perception systems
- Experience in 3D geometry, semantic segmentation, or learned multi-view stereo
- Strong understanding of classical computer vision (e.g. camera calibration, epipolar geometry, structure-from-motion, SGBM stereo)
- Ability to blend classical computer vision with modern ML approaches
- Expertise and depth with robotics fundamentals: reasoning about reference frames, matrix math, SE(3) manifolds and probabilistic sensor fusion
- Hands-on experience training, iterating on, and optimizing CNN and transformer architectures on target hardware (NVIDIA Jetson sized compute)
- Engineering mindset focused on outcomes over experimentation
- Experience with the full ML lifecycle (training, data annotation, evaluation, hard mining)
- Experience influencing design across systems: data pipelines, training infrastructure, compute requirements
Desired Qualifications
- ML experience on hardware applications (robotics)
Hiring someone like this?
Get your role in front of qualified candidates on Sorce.