HMI, the name is short for H: Human, M: Machine, I: Intelligence, and based at the School of Computer Science at Peking University. Our lab is dedicated to exploring the dynamic interplay between humans and machines, striving to advance intelligent interaction systems that are both innovative and efficient.

Our goal is to solve real-world problems through research and development, enhancing the level of machine intelligence to better serve humanity and improve the interactive experience between humans and machines. Through these efforts, HMI Lab aims to contribute significant value to society, fostering deeper collaboration and co-evolution between humans and machines across various domains.

Machine computational power and human creativity converge at HMI Lab, where we are dedicated to unleashing the synergistic potential that heralds a new era in intelligent technology. Our lab boasts extensive computational resources, including a fleet of high-performance servers such as A100s, V100s, and RTX 4090s, which provide the necessary power for complex data processing and model training. Additionally, our arsenal of embodied AI, ranging from versatile robotic arms and agile robotic dogs to sophisticated dual-arm robots, equips us to tackle a wide array of research challenges in robotics and human-machine interaction. This powerful combination of advanced hardware and creative minds ensures we can meet and exceed the ambitious goals of cutting-edge scientific research.

Intelligence advancements at HMI Lab are redefining the boundaries of multimodal large models, with numerous top-tier international conference including CVPR, NIPS, ICML, ICLR, etc, and securing top positions in prestigious competitions like CVPR and ICCV. Our achievements have not only enhanced the generalization capabilities of systems employed by Baidu, Didi, and Panasonic but also led to the development of the X-Accessory suite. This comprehensive toolkit facilitates pre-training, fine-tuning, quantization, and deployment across various modalities. Products such as SPHINX, WeMix, OneLLM, and Mixtral-MoE have emerged from this platform, gaining significant traction on GitHub and widespread industry adoption, including by ByteDance.