Proposed a novel method for reproducible and shareable robot learning by focusing on rich and intuitive representation, processing, and storage of manipulation data. This was achieved through the DeepClaw system that features a unified data storage format, easily accessible web-based software, and open-source, low-cost hardware for teleoperation and data collection.

• GitHub

Designed and implemented an automated robotic waste sorting line with visual recognition. Improved sorting efficiency with a throw method rather than a traditional pick-and-place approach.

Demo Video

Control robot hand with any camera by recognizing hand gestures.

Demo Video

Showed on The Mobile World Congress (MWC) 2025 with Honor(a Chinese smartphone manufacturer).

• Bilibili

• YouTube

Developed large-scale (de)palletizing application using heavy-duty KUKA robots and Mixed Integer Programs (MIPs) for industry automated warehouses, achieving a processing speed of up to 1200 boxes per hour per single robot.

Demo Video

Depalletizing by layer

Depalletizing by row