Spatial Interaction Intelligence
ZGCA HMI LAB
The team's research goal is to explore the next stage of embodied intelligence, moving from language and action capabilities toward analyzable, predictable, and closed-loop spatial interaction among humans, machines, and environments. The team's industrial vision is to build interactive embodied intelligence that can enter human activities, understand human intentions, and act naturally together with people.
CognitionUnderstand spatial and human-interaction knowledge
ImaginationPredict world and ego-state evolution
SkillGenerate spatially consistent embodied behavior

Research Directions
Interaction Priors
- 01Scene Interaction Parsing
- 02Scene Reconstruction and Tracking
- 03Hand-Object Contact Prediction
Interaction Medium
- 01Spatiotemporal State Representation
- 02World State Rollout
- 03Dynamic State Modeling
Interaction Generation
- 01Manipulation Model
- 02Navigation Model
- 03Reaction Model
Research Updates
- 2026-07-10Two Papers Accepted by ACM MM 2026
- 2026-02-20Rex-Omni Accepted by CVPR 2026
- 2026-01-26Rex-Thinker Accepted by ICLR 2026

