Software Engineer, Perception

We're looking for a Robot Perception Engineer to develop the systems that give our robots a real-time understanding of the parts and workcells in front of them. You'll own the software that turns raw sensor data into the structured observations our controllers depend on, from camera integration and calibration through segmentation, tracking, and the 3D monitoring that lets our robots run alongside people.

• Develop and maintain real-time perception pipelines, integrating and calibrating cameras, depth sensors, and LiDAR with our robot platforms
• Build segmentation, tracking, and pose estimation that holds up under part-to-part variation, deformable materials, and unreliable depth on thin or reflective features
• Design the structured observation spaces our learned controllers consume, along with the labeling pipelines that make training them cheap
• Own the 3D safety perception stack: occupancy prediction from point clouds, speed modulation, and protective stops that let our robots operate without fencing
• Work with the controls and policy teams to connect perception outputs to policy inputs at control frequency
• Support bring-up and field testing of new camera and sensor configurations on live production lines
• Contribute to fault detection and recovery logic so the system survives an eight-hour shift

• 4+ years of experience in robotics perception, computer vision, or a closely related field
• Strong software engineering fundamentals in Python and C++
• Hands‑on experience with sensor integration and calibration (cameras, depth sensors, LiDAR, hand‑eye)
• Experience building real‑time perception pipelines that ran on real hardware
• Fluency with both learned and classical CV, and the judgment to know which one a problem calls for
• Experience with ROS/ROS2 or similar robotics middleware
• Ability to debug across the full stack, from driver‑level sensor issues to model behavior on live hardware

• Production experience with foundation‑model perception (SAM2, DINOv2, open‑vocabulary detectors), including where they break
• Point cloud and 3D work: voxelization, occupancy networks, LiDAR processing
• Manipulation background: visual servoing, imitation learning, ACT or diffusion policies
• Sim‑to‑real experience, including domain randomization and sim/real co‑training
• Deploying and optimizing models on edge or onboard compute (quantization, TensorRT)
• Exposure to safety standards for collaborative robotics (ISO/TS 15066, speed and separation monitoring, power and force limiting)
• Prior work on early‑stage hardware programs

• Perception is the constraint. What our controllers can see sets the ceiling on what they can do, and the observation design you own moves task success more than the size of the model behind it.
• You’ll work at the boundary of hardware and AI on live production lines, where success is measured by a QC pass rate rather than a benchmark score.
• High ownership on a small team. The decisions you make will shape how our robots perceive every site we deploy to.