Junsong Wu 吴俊松

I'm an M.S. student in Mechanical Engineering at Shanghai Jiao Tong University, advised by Prof. Zhenhua Xiong. I'm also a research intern at TeleAI, working with Dr. Chenjia Bai.

I aim to build general-purpose intelligence that learns and acts in the physical world. Toward this goal, my research focuses on scaling physical experience through simulation, grounded by real-world interaction. I explore how hardware, simulation, and learning can be co-designed so that physical knowledge accumulates and transfers across tasks, environments, and embodiments.

Research

^* Equal contribution ^† Corresponding author

Test-Time Compute for Humanoid Behavior Foundation Models

Weiji Xie^*, Junsong Wu^*, Zibo Zhou, Yueru Chen, Xingyi Wang, Zhenhua Xiong, Weinan Zhang, Chenjia Bai^†

Under review

project page/arXiv/code

TL;DR Test-time command optimization adjusts the inputs of a frozen humanoid behavior foundation model so the executed motion stays closer to the desired one, without retraining.

KODA: Kinematics-Observation Data Alignment for Humanoid Learning from Human Egocentric Demonstration

Zibo Zhou^*, Yueru Chen^*, Junsong Wu, Weiji Xie, Qingyao Xu, Sheng Yin, Shiqing Wang, Jiyuan Shi, Siheng Chen, Chenjia Bai^†, Xuelong Li^†

Under review

project page/arXiv/code

TL;DR KODA aligns human egocentric demonstrations with a humanoid's kinematics and observations, so those demonstrations can be used for learning.

NEXUS: Perceptive Whole-Body Control for Terrain-Adaptive Teleoperation

Xiangyu Miao, Junsong Wu, Jiyuan Shi, Weiji Xie, Jinrui Han, Xingyi Wang, Weinan Zhang^†, Chenjia Bai^†, Xuelong Li

Under review

project page/arXiv/code

TL;DR NEXUS turns flat-ground motions into terrain-consistent references and trains a perceptive whole-body controller for teleoperation across different terrains.

ECHO: In-Context Behavior Foundation Model for Humanoid Whole-Body Control

Xingyi Wang^*, Bin Li^*, Jiyuan Shi, Chenyun Zhang, Weiji Xie, Xiangyu Miao, Junsong Wu, Wei Song, Shiqiang Zhu, Chenjia Bai^†

Under review

project page/arXiv/code

TL;DR ECHO learns an in-context behavior foundation model that prompts one whole-body controller for many behaviors.

CLOT: Closed-Loop Global Motion Tracking for Whole-Body Humanoid Teleoperation

Tengjie Zhu^*, Guanyu Cai^*, Zhaohui Yang^*, Guanzhu Ren, Haohui Xie, Junsong Wu, ZiRui Wang, Jingbo Wang, Xiaokang Yang, Yao Mu^†, Yichao Yan^†

Under review

project page/arXiv/code

TL;DR CLOT closes the loop on whole-body humanoid teleoperation, correcting global position and heading during long-horizon tracking.

Miscellaneous

Beyond research, music has long been a big part of my life. I have spent countless hours somewhere between rehearsals and the stage, playing guitar with my bands. Alongside my engineering education, I studied French and developed a lasting interest in languages and the humanities, especially literature, philosophy, and history.

A few moments on stage

Playing The Smashing Pumpkins’ Siamese Dream, captured with motion blur — Playing The Smashing Pumpkins with my band.

My band at a Tatsuro Yamashita tribute show, Ride on Time — With my band at a Tatsuro Yamashita tribute — Ride on Time.

My band playing Parachute Day outdoors at night — Playing Parachute Day outdoors with my band.

Playing Radiohead with my band on an indoor stage — Playing Radiohead with my band. Video

My guitar pedalboard with effects pedals and cables — My pedalboard, inspired by Jonny Greenwood, my favorite musician.

Wealthy with all you’ve gained on the way.

— C. P. Cavafy, Ithaka