Modi Shi
Ph.D. Candidate in Computer Science & Technology
Hello! I am a second year Ph.D. student at Beihang University (BUAA), supervised by Professor Di Huang. I am also a joint Ph.D. student at Shanghai Innovation Institute, supervised by Professor Hongyang Li. I received my B.Eng. degree in Computer Science and Technology from Beihang University in 2023. I worked as a research intern at AgiBot in 2024–2025 and at Kinetix AI in 2025–2026.
My research interests include Embodied AI, Humanoid Loco-manipulation, Large-scale Robot Data, and Generalist Robot Policy.
My Education
Selected Publications
AgiBot World Colosseo: A Large-Scale Manipulation Platform for Scalable and Intelligent Embodied Systems
A million-scale robot manipulation dataset spanning 217 tasks across 5 scenarios, with GO-1 generalist policy achieving 30%+ improvement over RDT.
GO-1-Pro: Is Diversity All You Need for Scalable Robotic Manipulation?
A systematic study revealing that task diversity matters more than data quantity, single-embodiment data suffices for cross-embodiment transfer, and expert diversity can be detrimental.
EgoHumanoid: Unlocking In-the-Wild Loco-Manipulation with Robot-Free Egocentric Demonstration
A framework co-training VLA policy from egocentric human demonstrations and limited robot data, with view and action alignment bridging the embodiment gap, achieving 51% improvement over robot-only baselines.
χ₀: Resource-Aware Robust Manipulation via Taming Distributional Inconsistencies
A resource-efficient dual-arm manipulation framework tackling distributional shifts via model arithmetic, stage-aware advantage, and train-deploy alignment, achieving 250% improvement over π₀.₅ with only 20h data.
Research Experience
- Research on humanoid loco-manipulation and egocentric data collection
- Research on large-scale robot data collection for general-purpose robotics