3D Modeling, Artificial Intelligence (AI), Benchmarking, C++ Programming Language, Cloud Computing, Computer Vision, Data Modeling, Debugging Tools, Device Drivers, Embedded Hardware, Embedded Systems, Experiment Design, Localization, Machine Tool, Prototyping, Robotics, Robotics Software, Software Engineering, System Architecture
Robotics Embedded Compute Computer Vision The Role
This role bridges the gap between ML research and real-world robotic systems. You'll take trained vision models and make them performant and production-ready on embedded hardware not by doing the research yourself, but by deeply understanding the path from a PyTorch checkpoint to optimized edge inference. You'll be responsible for the full perception pipeline: ingesting sensor data, executing models in real time, and feeding actionable results to downstream robotics software. Day-to-Day Responsibilities
- Optimize pre-trained computer vision models for NVIDIA embedded platforms, primarily using TensorRT and supporting toolchains
- Own the perception stack from end to end depth camera ingestion, point cloud processing, and real-time object detection leveraging commercial sensor models
- Design and run experiments to evaluate perception approaches; make data-driven trade-off decisions on what belongs in production
- Wire OpenCV- and PyTorch-based inference pipelines into the team's ROS 2 architecture
- Integrate stereo camera hardware (ZED and similar) and make pragmatic use of vendor-provided SDK models where appropriate
What You Bring
Core Requirements- At least 7 years building software for robotic systems
- Deep hands-on experience with ROS 2 authoring nodes, orchestrating topic/service/action patterns, and designing clean system architectures
- A systems-level perspective: you're comfortable at the boundary between hardware drivers, real-time control loops, and application-layer software
- Strong Python skills; working proficiency in C++ (able to read, modify, and extend existing codebases confidently)
- Prior experience deploying and debugging software on NVIDIA Jetson platforms (Orin, Xavier, or comparable)
- Self-directed and effective in early-stage, greenfield development environments
Perception & Edge Deployment- Production-level experience with PyTorch and OpenCV not just prototyping, but building pipelines that need to run reliably
- Practical experience with model optimization workflows: TensorRT, ONNX conversion, DeepStream, or equivalent tooling
- Solid grasp of perception fundamentals including depth estimation, 3D point cloud processing, and detection architectures
- Track record of independently evaluating and benchmarking technical approaches rather than solely executing against a spec
Nice to Have- Experience with visual SLAM, sensor fusion, or related localization techniques