Tong Zhang 张彤

I received my Ph.D. from the Institute for Interdisciplinary Information Sciences (IIIS), Tsinghua University, advised by Prof. Yang Gao. I previously obtained my bachelor’s degree from the Department of Electronic Engineering at Tsinghua University.

During my Ph.D., I spent six months as a visiting scholar at UC Berkeley, advised by Koushil Sreenath in the Hybrid Robotics Group (HRG) and Berkeley Artificial Intelligence Research Lab (BAIR).

My research focuses on Embodied AI, which lies at the intersection of Artificial Intelligence and Robotics.

Email: tongzhangthu [AT] gmail.com

Email  /  Google Scholar  /  Github

News

Publications

We present OpenHLM, an open-source recipe for whole-body humanoid loco-manipulation. Through controlled studies on whole-body teleoperation, VLA policy design, and heterogeneous co-training, OpenHLM outperforms GR00T N1.6 and Psi0 with less than half the demonstration time.

We present Qwen-RobotManip, a Vision-Language-Action foundation model for robotic manipulation that studies how unified alignment across heterogeneous robot data enables scalable multi-source training. The model is evaluated on a range of robotic benchmarks and ranks 1st on the RoboChallenge Table30 v1 Generalist Track, demonstrating strong generalization across unseen tasks, scenes, and robot platforms.

We present HuMI, a portable framework for humanoid whole-body learning via robot-free data collection. HuMI achieves 3x higher efficiency than teleoperation and a 70% success rate across diverse tasks and unseen environments.

We propose HuB (Humanoid Balance), a framework that enables humanoids to perform challenging quasi-static balance tasks, including extreme single-legged poses such as the Swallow Balance and Bruce Lee's Kick.

We introduce SGRv2, an imitation learning framework that enhances sample efficiency through improved visual and action representations. Central to the design of SGRv2 is the incorporation of a critical inductive bias-action locality, which posits that robot's actions are predominantly influenced by the target object and its interactions with the local environment.

We propose Reinforcement Learning with Foundation Priors (RLFP) to utilize guidance and feedback from policy, value, and success-reward foundation models. Within this framework, we introduce the Foundation-guided Actor-Critic (FAC) algorithm, which enables embodied agents to explore more efficiently with automatic reward functions.

We build a 3D flow prediction model directly from large-scale RGBD human video datasets. Based on this model, we achieve stable zero-shot human-to-robot skill transfer in the real world.

We introduce ViLa, a novel approach for long-horizon robotic planning that leverages GPT-4V to generate a sequence of actionable steps. ViLa empowers robots to execute complex tasks with a profound understanding of the visual world.

We present Semantic-Geometric Representation (SGR), a universal perception module for robotics that leverages the rich semantic information of large-scale pre-trained 2D models and inherits the merits of 3D spatial reasoning.