Shuzhe Zhang

My research focuses on robot manipulation, particularly dexterous manipulation.

Shuzhe Zhang
Shuzhe Zhang

I am an M.S. student in Electrical Engineering at Northwestern University, in the Robotics and Autonomous Systems specialization, advised by Sam Kriegman in the Xenobot Lab. I am also a research intern in the HCI Lab at SIAT, Chinese Academy of Sciences, advised by Yinling Qian.

I currently work on the co-design of dexterous robot hand morphology and control, with an emphasis on bringing these designs to real hardware. I am also studying world models for precise, long-horizon manipulation, with the goal of learning dynamics and higher-level task structures that support more effective modeling and planning of complex manipulation processes.

In the long term, I hope to further explore the relationships among robot morphology, dynamics, and control, and contribute to more general and dexterous robotic manipulation.

Research

  1. Evolving Dexterous Robots from Scratch 2026

    Robot co-design for dexterous manipulation. SPaCE, a spherical-projection autoencoder trained with contrastive learning, embeds two million voxel-grown manipulators on a unit hypersphere; hands are evolved in that latent space, trained in Isaac Lab, built on Dynamixel servos, and fine-tuned on the physical hand.

    Under reviewProject pagearXiv
  2. S²-HWM: a Hierarchical World Model for Long-Horizon Surgical Manipulation Aug 2026

    A hierarchical world model that learns a few task events from latent dynamics and plans over them rather than primitive steps. 98.7% success on SurRoL PegTransfer, against 76.0% for a flat DreamerV3.

    PreprintProject pagearXivAlso at ELSR Workshop, IROS 2026
  3. Creating Manufacturable Blueprints for Coarse-Grained Virtual Robots Mar 2026

    A four-stage pipeline that converts evolved virtual robots into manufacturable, assembly-ready blueprints. Across 1,000 sampled designs 67% passed every solver stage, and an evolved tripedal design walked zero-shot off the printer.

    Under reviewProject pagearXivAlso at Co-Design Workshop, IROS 2026
  4. Constraint-Aware Bone Tumor Osteotomy Planning 2025

    Constraint-aware deep learning that automates pelvic bone-tumor resection planning with guaranteed margins. 100% target-margin coverage on 20 clinical cases, against 93.3% for geometry-only ranking.

    Project page
  5. Shared-Control Evaluation for Powered Wheelchairs Sep – Dec 2025

    User studies of REACT, a shared-control driving-assistance system on a LUCI-equipped powered wheelchair. Ran the sessions of an IRB-approved study and kept the wheelchair's ROS 2 stack reliable across them.

    Part of Project DRIVE

Publications & Preprints

  1. Evolving Dexterous Robots from Scratch Z. Guo*, S. Zhang*, M. Li*, P. Li, S. Kriegman
    Under review · arXiv:2609.33101 · Project page · *equal contribution
  2. S²-HWM: Sparse Event-Structured Hierarchical World Model for Long-Horizon Surgical Robot Manipulation S. Zhang, X. Zhu, Y. Qian, Q. Wang
    Preprint · arXiv:2608.13103 · Project page · Also presented at the ELSR Workshop, IROS 2026 (non-archival)
  3. Creating Manufacturable Blueprints for Coarse-Grained Virtual Robots Z. Guo, M. Li, S. Zhang, S. Kriegman
    Under review · arXiv:2603.13582 · Project page · Also presented at the Co-Design Workshop, IROS 2026 (non-archival)
  4. USD-YOLO: An Enhanced YOLO Algorithm for Small Object Detection in Unmanned Systems Perception H. Deng, S. Zhang, X. Wang, T. Han, Y. Ye
    Applied Sciences 15(7):3795, 2025 · DOI

Research experience