Mowze mowers are operating in the field, and perception is what keeps them safe and productive. You will improve the safety and navigation of our robot. You will own perception on the mower. That means selecting the next-generation cameras and compute platform, training and deploying models, and building the pipeline that turns fleet data into training data. You'll ensure the mower sees grass, obstacles, trash, people, pets.. You'll deliver visual odometry that holds the mower's position when GNSS degrades, define and measure our perception safety metrics on real fleet data, and reduce the amount of human review the fleet needs. Hybrid, based in Gaithersburg, MD. On site 2-5 days a week, as needed.

Responsibilities Own the perception architecture:

camera selection, compute platform selection, and the sensor suite decision including a LiDAR/ToF/Stereo evaluation Own the perception pipeline. Train, quantize, and deploy the models on the chosen hardware.

Build the data engine:

ingestion from the fleet. Review stop events and teleop review outcomes. Deliver visual odometry / VIO to support navigation during GPS degradation. Define perception safety metrics and measure on fleet data. Quantify the human-review load, then improve system performance. Qualifications 5+ years architecting vision systems. Production perception on embedded hardware.

Preferred:

Outdoor/agricultural vision systems. Has designed a camera system (lens/sensor selection, HDR, exposure control, calibration, multi-camera stitching), not just used images from one. Experience with detection and segmentation training pipelines and with building a data pipeline. Real stereo or depth-sensing experience, or LiDAR, or ToF. An understanding of the outdoor failure modes of each Visual odometry / VIO / SLAM experience on a real moving platform. C++ and Python. Comfortable in code. Experience with Claude Code or alternative. Has worked on a safety-relevant detection task.

Pay:

$130,000.00 - $180,000.00 per year People with a criminal record are encouraged to apply Application Question(s): For a multicamera system in a harsh environment, what communication between cameras and central inference / controller do you suggest / have used? Each camera is 1-2m from the central inference. Have you ever deployed this method to a production environment, how did it go? Do you have real world experience with a production fleet of robots / vehicles? (How many were in the fleet? Did the robots operate outdoors? Roughly how much did the robots weigh? What was and how long was your role?) Which compute platform would you choose for an outdoor mobile robot, and why? What experience do you have with that compute system? (Or other products) For our application (3D sensing in the outdoors), succinctly compare LiDAR, ToF, and stereo vision. (Sonar also in this list, if you choose) Detail your experience with the systems. What applications have you developed that use segmentation models for navigation? What was the pipeline? The hardware it ran on?

Work Location:

Hybrid remote in Gaithersburg, MD 20879