☕️I’m Junqi Jing (荆浚淇), a junior undergraduate in Software Engineering at Harbin Institute of Technology. My research lies at the intersection of Embodied AI, Physical Simulation, and Multimodal Diffusion Models.
I have worked at Westlake University (MiLab), and will be an incoming exchange student at POSTECH under Prof. Kwang In Kim. I am a co-author of a paper accepted to ICLR 2025 (MTID) and involved in multiple ongoing projects related to physics-grounded LLMs and robotic learning from videos.
I am currently seeking a 2025 winter internship. If you have any ideas or potential collaborations, feel free to contact me at jqjing@stu.hit.edu.cn or xingkong8527@gmail.com.
BE in Software Engineering
Harbin Institute of Technology
I’m interested in embodied intelligence, physics-informed AI, and generative modeling. I build systems that enable robots and agents to learn and reason in complex environments.
My work spans from visual-language-based policy learning to dynamic 3D scene understanding using GNNs. Currently, I’m exploring how we can combine large language models, physical simulations, and diffusion policies to achieve generalizable decision-making.
Always happy to collaborate or discuss ideas!
I’m thrilled to begin my visit to the Suzhou campus of Nanjing University, where I’ll be joining Prof. Kai Zhang’s lab to explore the application of generative AI and diffusion policies in embodied intelligence.
I’m excited to contribute as a PC member and reviewer for the NeurIPS 2025 Workshop on Embodied World Models for Decision Making, which focuses on world models in embodied AI. The workshop will take place in San Diego, USA.
I will begin a semester-long exchange at POSTECH in South Korea. During the exchange, I’m excited to join MLVlab and collaborate with Professor Kwang In Kim on research in Embodied AI and generative AI.
In this work, we propose a novel modeling method for heterogeneous hybrid objects, based on improvements to the Spring-Gaus physics engine. I contributed to simulation, GNN modeling, experiments, and dataset creation.
We propose a novel Masked Temporal Interpolation Diffusion (MTID) model for procedure planning in instructional videos. The paper is officially accepted at ICLR 2025.
