preprint · arXiv (2025) · Dec 16, 2025
Towards Interactive Intelligence for Digital Humans
Yiyi Cai, Xuangeng Chu, Xiwei Gao, Sitong Gong, Yifei Huang, Caixin Kang, Kunhang Li, Haiyang Liu, Ruicong Liu, Yun Liu, Dianwen NG, Zixiong Su, Erwin Wu, Yuhan Wu, Dingkun Yan, Tianyu Yan, Chang Zeng, Bo Zheng, You Zhou
In plain English
A large team proposes treating a digital human as one end-to-end system rather than a chain of parts, and builds Mio, a framework of five modules — a reasoning "Thinker", a speech "Talker", face and body animators, and a renderer — sitting on a layered memory so the character stays consistent across a long interaction. They also publish benchmarks for measuring it. Worth reading as a reference architecture if you are assembling an avatar stack from separate vendors, since it names the seams where consistency usually breaks; the reported wins are self-evaluated against the authors' own benchmark.