Computer Vision Engineer

Job description

About Tutor We believe robots will be the next chapter in making physical labor safer and more affordable. But, nearly nine out of ten American factories and warehouses don’t have a single robot. Tutor Intelligence is changing that. We build robots that go straight into real factories and warehouses, to do real work, with real impact. Our robot Cassie does back-breaking work, moving boxes, pallets, and bulk goods for family-owned manufacturers to some of the largest brands in the world. Our robot Sonny does handwork, using two arms to achieve human-like dexterity in picking and packing. Every robot learns on the job, and each one we deploy makes the next one more capable, more affordable, and faster to put to work. Founded at MIT and headquartered in Watertown, MA, we do it all in-house: the robotics hardware, the machine learning, and the field deployment and service. We’re a team of researchers, engineers, technicians, operators, and sellers who love building things and solving hard problems. Our mission is grand: robots everywhere, for the common good, raising the standard of living across North America, and then the world. The Role Build perception that makes robot manipulation experiments reliable. You will own a significant vision system from problem definition and data collection through evaluation, integration, and ongoing improvement on physical robots. The work combines camera geometry and learned models, with decisions grounded in measured performance. You will work closely with research, software, mechanical, and electrical engineers to understand failures across sensing, algorithms, and the robot environment. This role focuses on perception and its integration into manipulation systems. Responsibilities Define the technical approach and success criteria for ambiguous perception problems in robot manipulation. Build and integrate capabilities such as detection, segmentation, tracking, object pose estimation, and 3D scene understanding. Own camera calibration, coordinate transforms, synchronization assumptions, and checks that expose degraded sensor quality. Create reproducible datasets and evaluations covering changes in objects, lighting, viewpoint, and occlusion; prevent evaluation leakage. Select and adapt geometric methods and learned models based on accuracy, latency, robustness, and operational constraints. Deploy and maintain perception in the robot stack; diagnose failures using recorded data and physical experiments. Improve shared evaluation tools and documentation, contribute to design reviews, and mentor colleagues in your domain. Requirements Evidence of independently owning a substantial computer-vision system from an ambiguous problem through tested integration. Strong Python engineering and experience writing maintainable code, tests, and reproducible experiments. Practical command of camera models, calibration, coordinate frames, and 3D geometry. Experience with geometric vision or learned visual models, with the judgment to explain when each approach is appropriate. Ability to build representative evaluations, isolate failure modes, and distinguish model improvements from data or measurement artifacts. Ability to communicate technical tradeoffs and resolve interfaces with engineers in other disciplines. Nice to have Perception deployed on physical robots, especially manipulation systems. Multi-camera or depth systems, sensor fusion, SLAM, or object pose estimation. OpenCV, Open3D or PCL; a modern deep-learning framework; C++ or GPU optimization. What Success Looks Like Establish a reproducible baseline, meaningful evaluation, and documented sensing assumptions for an agreed perception problem. Deliver a measured improvement that holds up on the physical robot and meets runtime constraints. Keep the system diagnosable and usable by colleagues through reliable calibration checks, failure analysis, and documentation.