GoAnywhere: Towards a Whole-Body Teleoperation System Breaking the Spatial Limitation

Xiaojie Niu∗ (to be added)

[Video] [Project website] [Paper]

We present GoAnywhere, a whole-body teleoperation system using an omnidirectional treadmill to break spatial limitations. Integrating motion capture, motion editing, and a trained tracking policy, it enables long-range task execution and has the potential to serve as a scalable data collection framework for humanoid autonomy.

Feel Robot Feels: Tactile Feedback Array Glove for Dexterous Manipulation

Feiyu Jia∗, Xiaojie Niu∗, Sizhe Yang∗, Qingwei Ben, Tao Huang, Feng Zhao†, Jingbo Wang†, Jiangmiao Pang†

[Project website] [Paper]

We present TAG, a low-cost glove system with magnetic 21-DoF hand tracking (<1° error) and high‑resolution tactile feedback. It enables real‑time perception of contact geometry and force, improving success rates in contact‑rich teleoperation and demonstration quality.

Behavior Foundation Model for Humanoid Robots

Weishuai Zeng, Shunlin Lu, Kangning Yin, Xiaojie Niu, Minyue Dai, Jingbo Wang, Jiangmiao Pang

[Project website] [Paper]

We propose BFM, a generative model using masked online distillation and CVAE. Pretrained on large-scale behavioral data, it generalizes across whole-body control tasks and adapts to new behaviors without retraining.

PhysHSI: Towards a Real-World Generalizable and Natural Humanoid-Scene Interaction System

Huayi Wang∗, Wentao Zhang∗, Runyi Yu∗, Tao Huang, Junli Ren, Feiyu Jia, Zirui Wang, Xiaojie Niu, Xiao Chen, Jiahe Chen, Qifeng Chen†, Jingbo Wang†, Jiangmiao Pang†

[Project] [Paper]

We present PhysHSI, a humanoid-scene interaction system using adversarial motion prior-based policy learning and coarse-to-fine LiDAR-camera fusion. It achieves high success rates, strong generalization, and natural motions across four tasks in simulation and real world.

JAEGER Dual-Level Humanoid Whole-Body Controller

Ziluo Ding∗, Haobin Jiang∗, Yuxuan Wang∗, Zhenguo Sun, Yu Zhang, Xiaojie Niu, Ming Yang, Weishuai Zeng, Xinrun Xu, Zongqing Lu†

[Project website] [Paper]

We propose JAEGER, a dual-level whole-body controller separating upper and lower bodies. Using retargeting and curriculum learning (supervised + RL), it outperforms SOTA methods on two humanoid platforms in simulation and real world.

Design and Control of Bio-Inspired Joints for Legged Robots Driven by Shape Memory Alloy Wires

Xiaojie Niu, Xiang Yao, Erbao Dong†

[Paper]

We propose a bio-inspired joint with antagonistic SMA actuators and differential dual-diameter pulleys, achieving ±60° stroke and bidirectional control. An SMA-driven leg and motion retargeting framework validate its motion performance via human gait tracking.

Design and Position Control of a Bionic Joint Actuated by Shape Memory Alloy Wires

Yida Zhu, Zhikun Jia, Xiaojie Niu, Erbao Dong†

[Paper]

We propose a bionic joint using SMA wires and a differential pulley set. A PI controller achieves good position control with ±30° rotation, offering an innovative solution for enhancing SMA actuated joint range of motion.

Whole-body control system based on multi-view vision pose estimation

Project Leader

[Project website]

We propose a whole-body control system based on multi-view human pose estimation. Using multiple RGB cameras, it eliminates the inaccurate root position estimation common in traditional methods, producing coordinated and executable human motions. These motions are retargeted to the robot and combined with a downstream reinforcement learning-based motion tracking policy, achieving real-time, robust, and natural whole-body control.

Universal Data Pipeline for Humanoid Robots

Project Leader

Universal Data Pipeline for Humanoid Robots, adaptable to various types of data interfaces and hardware, enabling real-time, robust human motion capture through a processing toolkit, and supporting a wide range of humanoid robots and dexterous hands.