I am a junior undergraduate pursuing a B.S.E. in Computer Science at the University of Michigan, Ann Arbor and a B.E. in Electrical and Computer Engineering at Shanghai Jiao Tong University. I am currently a research intern in the ARM Lab at Michigan, advised by Prof. Dmitry Berenson.

Research interest: real-world reinforcement learning for robot foundation models — post-training generalist (VLA) policies directly on physical robots to make them reliable at precise, contact-rich, and long-horizon manipulation. 🤖

• [09/2026] Joined the ARM Lab at the University of Michigan as a research intern.
• [09/2026] Started my B.S.E. in Computer Science at the University of Michigan, Ann Arbor.
• [06/2026] Q-VGM, our offline-to-online RL method for flow-matching VLAs, is on arXiv.
• [2026] Scene2Gym is accepted to IROS 2026.
• [05/2026] Started a research internship at AgiBot on real-world RL post-training of VLA policies.

• Q-VGM: Q-Guided Value-Gradient Matching for Offline-to-Online RL of Flow-Matching VLA
  Ziqian Wang, R. Zhang, Y. Liu, X. Mao, M. Wang, Y. Mu† — Under Review, 2026 · [arXiv]
• Scene2Gym: Transforming Real-World Scenes into Interactive Training Environments for Robot Learning
  Z. Ji*, F. Chen*, Ziqian Wang, et al. — IROS 2026

_{* Equal contribution · † Corresponding author}

• ARM Lab, University of Michigan (Sep 2026 – Present): extending Q-VGM with a learned latent world model for online real-world RL fine-tuning of VLA policies.
• AgiBot (May 2026 – Aug 2026): post-trained π_0.5 with real-world RL for cloth folding and phone-charger plug insertion; built an end-to-end real-robot SFT + RL post-training system.
• ScaleLab, SJTU (Apr 2025 – Present), advised by Prof. Yao Mu: Q-VGM (first author) and Scene2Gym (IROS 2026).

• PPO RL with 3DGS rendering: PPO software suite for manipulation in simulation with high-fidelity 3DGS rendering, plus a sim2real policy transfer pipeline (part of UniRoboSim / Scene2Gym).
• MMLU-PRO-Leveled-TinyBench & Jailbreak testing: a leveled, data-efficient benchmark for LLMs.
• RoboPicker: my first robotics project, from high school.

• Programming: Python, C++ (NOIP First Prize 🏆), MATLAB
• Frameworks: PyTorch, MuJoCo, Isaac Gym, ROS, Weights & Biases
• Robots: AgiBot G2, Franka, AgileX Piper, KUKA

• To the best girlfriend in the world, ddx-cs ❤️