Human-centric Embodied Intelligence

Created on June 28, 2026

2026   ·   Embodied_AI

Robot Learning from Human Videos


Our research aims to advance human-centric embodied intelligence by developing AI systems that learn from human videos to interact intelligently with the physical world. We investigate how AI systems can understand human activities and physical interactions in complex environments, and how robots can acquire transferable physical knowledge and manipulation skills from human demonstrations. By bridging human understanding and robot learning, our research seeks to enable robots to generalize across different embodiments, tasks, and environments, ultimately supporting intelligent and adaptive interactions with the physical world.

    Publications
    1. Guo, W., Huang, Z., & Kong, Y. (2027). DynamicHOI: Coupled Dynamics for Physics-aware HOI Reconstruction.
      2. Project
    2. Guo, W., Pu, Y., & Kong, Y. (2026). Procedural Mistake Detection via Action Effect Modeling. International Conference on Learning Representations (ICLR).
      3. Project
    3. Guo, W., & Kong, Y. (2026). Benchmarking and Learning Object-Centric Procedural Understanding in Instructional Videos. Neural Information Processing Systems (NeurIPS). Track on Evaluations and Datasets.
      4. Project
    4. Li, Y., Zhou, X., Ge, Y., Tan, X., & Kong, Y. (2027). Spatially Prompted Visual Trajectory Prediction for Egocentric Manipulation. Winter Conference on Applications of Computer Vision (WACV).
      5. Project


Safety Reasoning in Industrial Environment

This research aims to develop human-centric AI systems that understand human activities and environments to enhance safety in complex industrial environments. We investigate how AI systems can perceive and interpret human behaviors, reason about potential hazards and safety risks, and anticipate unsafe situations before accidents occur. By integrating human activity understanding, physical interaction modeling, and safety reasoning, our research seeks to enable intelligent systems that proactively identify risks, support informed decision-making, and promote safer human interactions with industrial environments.

    Publications
    1. Li, Y., Li, L., Dao, A., Zhou, X., Huang, W., Ma, T., Qiao, Y., Mai, Z., Lee, D., Chen, Z., Wang, P., Yang, L., Wang, T., Tan, Z., Li, S., Bansal, M., Ni, Y., & Kong, Y. (2027). IndustryNav: Exploring Spatial Reasoning of Embodied Agents in Dynamic Industrial Navigation.
      2. Project
    2. Cai, Z., Du, Y., Wang, C., & Kong, Y. (2026). Vision to Geometry: 3D Spatial Memory for Sequential Embodied MLLM Reasoning and Exploration. Neural Information Processing Systems (NeurIPS).
      3. Project
    3. Li, Y., Chen, Y., Dao, A., Li, L., Cai, Z., Tan, Z., Chen, T., & Kong, Y. (2025). IndustryEQA: Pushing the Frontiers of Embodied Question Answering in Industrial Scenarios. NeurIPS DB Track.
      4. Project