SJTU MVIG Lab

Cewu Lu's lab at Shanghai Jiao Tong University, one of China's earlier academic teams working on robot manipulation data and grasping.

MVIG is the Machine Vision and Intelligence Group at Shanghai Jiao Tong University, led by Professor Cewu Lu. It started out working on human pose and activity understanding, open-sourcing the pose-estimation tool AlphaPose, before turning to embodied AI and robot manipulation. Its guiding idea is to have robots learn general-purpose behavior from large amounts of human activity video and demonstrations. The lab's embodied-AI work is mostly datasets and foundational tools: the grasping benchmark GraspNet-1Billion and the grasp-perception system AnyGrasp, the real-robot manipulation dataset RH20T, the hand-object interaction dataset OakInk, the dual-arm exoskeleton AirExo, and force-aware policies such as FoAR and RDP. In 2023, Cewu Lu founded the embodied-AI company Noematrix to commercialize part of the lab's work. Readers of embodied-AI papers will often see these datasets and tools cited as benchmarks or data-collection methods.

ExampleMany grasping papers call the AnyGrasp SDK directly: feed in a depth camera's point cloud, and it returns a set of scored gripper grasp poses for the scene.

Also called

SJTU Machine Vision and Intelligence Group, Cewu Lu's Lab

Related

GraspNet-1Billion · AnyGrasp · RH20T · AirExo · OakInk · Noematrix

Sources

MVIG 实验室主页 (Chinese)
GraspNet 项目主页 (Chinese)
上海交大新跑出一家具身智能公司「穹彻智能」（雷峰网） (Chinese)

As of

2026-09