Zeyuan Gao’s Robots That Have to Work in the Physical World

Field notes

The deployment work spans synchronized multi-view capture, teleoperation, LeRobot-compatible data preparation, inference, robot communication, and physical execution. The documented setup includes two wrist cameras, a panoramic camera, and a separate VIVE-tracked UMI arrangement with four Lighthouse base stations. In the perception study, segmentation and stereo depth feed Any6D candidate generation before CAD geometry and object symmetry rerank the results; 970 stereo frames were evaluated, with 968 valid candidate outputs.