Pokuang Zhou

Third-year Ph.D. student in Robotics at Purdue University, advised by Professor Yu She.

Previously earned Bachelor's and Master's degrees in Engineering from Huazhong University of Science and Technology, advised by Academician Han Ding and Prof. Huan Zhao.

email
G. scholar
github

Research during Ph.D.

Research focuses on embodied AI, robotic in-hand manipulation, foundation model, visuotactile representations, control theory, and multi-modal robotic systems.

* indicates equal contribution.

Built an open-source platform for vision- and tactile-based robotic task execution in both Isaac Gym and the real world, enabling benchmarking of state-of-the-art machine learning methods.

We developed a custom 5-DOF gripper that integrates both hardware design and control algorithms, enabling in-hand singulation, scooping, and cable untangling.

A Control Barrier Function based framework for Safe Human-Robot Collaboration that enables adjustable risk modulation to trade off between safety and task efficiency.

Developed a reinforcement-learning whole-body controller with visuotactile fusion for a quadruped-arm, enabling diverse loco-manipulation tasks; trained in NVIDIA Isaac Gym and successfully transferred to real-world deployment via sim2real.

Designed a vision- and tactile-based reinforcement learning system enabling a UR5 robot to throw deformable darts overhand and accurately hit the target.

TacScope is a compact, low-cost vision-based tactile sensor with a spherical elastomer that, after single-image calibration, captures high-resolution 3D contact geometry.

StickRoller achieves precise in-hand stick repositioning through a sample-efficient tactile model. It enables the stick to be rolled to the center of the fingers using only a few well-planned two-finger manipulations.

A compact gripper with integrated GelSight tactile sensing that simplifies dexterous manipulation, demonstrating high success on granular singulation/classification and precise credit-card scooping/insertion tasks.

An autonomous robotic system that uses SAM-guided RGB-D perception, in-hand leaf manipulation, and a line-scan hyperspectral camera with a vision-based tactile tracker to scan individual corn leaves at high resolution.

Miscellanea

Feel free to steal this website's source code. Do not scrape the HTML from this page itself, as it includes analytics tags that you do not want on your own website — use the github code instead. Also, consider using Leonid Keselman's Jekyll fork of this page.