About Me

Incoming PhD at UMich in Fall 2024!!
Thank you to everyone who supported and helped me during the application season!

I am a PhD student of Robotics Department at University of Michigan (UMich), where I work on legged robot control and learning with professor Yanran Ding and previously collaborated with Dr. Maani Ghaffari and Dr. Elena Shrestha. Prior to that, I received my bachelor degree from Southern University of Science and Technology (SUSTech), majoring in Robotics Engineering. Advised by professor Wei Zhang, I had a great research experience on MPC and RL of quadruped robots.

I am particularly interested in agile legged locomotion in dynamic environments, with a focus on control and learning methods, and proficient in developing efficient and deployable robotic algorithms. When I'm not immersed in technology, you can find me swimming or playing badminton.

Experience

	Research Assistant Aug 2022 - Present, University of Michigan, Ann Arbor, MI, US
	Software Engineer Intern on Motion Planning May 2023 - Aug 2023, PlusAI Inc., Santa Clara, CA, US
	Visiting Student Aug 2021 - Sep 2021, National University of Singapore, Singapore
	Research Intern Jun 2021 - Jul 2021, Tsinghua University, Beijing, China

Highlight Projects

Hierarchical RL for MPC of Quadruped Locomotion [ Code ] [ Slides ] [ Parallel Training ] Fast simulation and RL training framework for a quadruped locomotion task by dynamically predicting the weight parameters of a MPC controller. The control framework is a hierarchical controller composed of a higher-level policy network and a lower-level model predictive controller.

Motion Planning Practice [ Code ] [ PDF ] Fast motion planning algorithm implementations with planar hovercraft and 7-DOF KUKA arm demos in pybullet. Algorithms including Kinodynamic RRT-Connect, RRT-Connect, A*, Iterative IK, lazy-rebuilt KD-Tree and more. Code optimized for readability and efficiency.

Bayesian Optimization for MPPI Planar Pushing [ Code ] [ PDF ] [ BOA ] Self implemented Bayesian Optimization Algorithm (BOA) in PyTorch for automatically tuning the hyperparameter of Model Predictive Path Integral (MPPI) control to solve a planar box pushing task with non-trivial obstacles.

Dynamic Object Removing SLAM with MonoRec [ Code ] [ PDF ] [ System Diagram ] We present a dynamic object removing SLAM method named MonoRec-SLAM. Our method adapts the MaskModule from MonoRec and achieves an average APE improvement by 6.03% on KITTI dataset, obtains a more static map of the scenes, and achieves a great balance between real-time capability and dynamic object masking.

Single-Image to Camera Pose with iNeRF and PoseCNN [ Code ] [ PDF ] [ System Diagram ] We present an efficient and robust system for view synthesis and pose estimation by integrating PoseCNN and iNeRF. Our method leverages the pose and object segmentation predictions from PoseCNN to improve the initial camera pose estimation and accelerate the optimization process in iNeRF.

Autonomous SLAM and Exploration with MBot [ Code ] [ PDF ] In the Botlab, movement control, obstacle detection, maze exploration, and self-localization functionality was developed on the MBot robot, a mobile robot platform. It is designed to explore the fundamentals of robot autonomy by developing MBot with autonomous mapping, localization, and exploration capabilities. Winner among 24 teams in the final competition.

Robust Detecting and Palletizing with Robot Arm [ Code ] [ CV PDF ] [ Control PDF ] In the Armlab, a 5-DOF robotic arm fully autonomously arranges blocks of different sizes, colors and positions into the desired arrangement. Analytical inverse kinematics is used to determine the appropriate waypoints. An overhead LiDAR Camera is utilized to identify blocks on the board. Winner among 24 teams in the final competition.

Agile Waste Sorting with Tossing [ Code ] [ PDF ] Implemented the automatic collection and cleaning of images based on MOG2 algorithm. Deployed and trained YOLOv5 to achieve quick waste classification. Achieved planning for robot arm to pick toss waste on dynamic conveyor belt.

Publications

Sense, Imagine, Act: Multimodal Perception Improves Model-Based Reinforcement Learning for Head-to-Head Autonomous Racing
arXiv:2305.04750
Shrestha, E., Reddy, C., Wan, H., Zhuang, Y., & Vasudevan, R.

[ Video ] [ PDF ] [ Multimodal Dreamer ] We present a multimodal perception system combining egocentric LiDAR and RGB camera improves robustness of the world model without requiring additional training data. The resulting multimodal Dreamer agent safely avoided collisions and won the most races compared to other tested baselines in zero-shot head-to-head autonomous racing.

Height Control and Optimal Torque Planning for Jumping with Wheeled-Bipedal Robots
International Conference on Advanced Robotics and Mechatronics. IEEE. 2021
Zhuang, Y., Xu, Y., ... & Fu, C.

[ Code ] [ Video ] [ PDF ] We present a wheeled-bipedal jumping dynamical (W-JBD) model to optimize the height control, and the Bayesian optimization for torque planning (BOTP) method based on a joint optimization framework for torque planning to achieve accurate height control and minimal energy cost.

Wheeled-Bipedal Robot
[ PDF ] [ View Robot ]
China Patent CN112977666A Jun 2021
W. Zhang, Y. Zhuang, et al.

Connecting Structure and Multi-Hull Ship
[ Code ] [ PDF ] [ View Robot ]
China Patent CN212637810U Mar 2021
Y. Zhuang, W. Zhang, et al.

Education

	University of Michigan, US Aug 2022 - May 2024, Master of Science in Robotics
	Southern University of Science and Technology, China Sep 2018 - Jun 2022, Bachelor of Engineering in Robotics