Research Projects
I am passionate about developing algorithms for robots to perform complex tasks in the real world. Here are the projects I have worked on:
Proposed a global-map-free navigation framework for efficient autonomous robot navigation in complex terrains. The framework employs Sparse Gaussian Processes (SGP) to extract geometric features—such as curvature, gradient, and elevation—from point cloud data. GPU acceleration is used during feature extraction to ensure real-time performance. These features are integrated into a high-resolution local traversability map using a spatial-temporal Bayesian Gaussian Kernel Inference method that fuses historical and real-time data with slope, flatness, gradient, and uncertainty metrics. Extensive simulations demonstrate significant improvements in accuracy and computational efficiency compared to traditional methods. The framework is integrated into a planner and validated through autonomous navigation experiments on a real robot. Source code available at github.com/ZJU-FAST-Lab/FSGP_BGK.
This paper presents a lightweight, high-order–smooth trajectory representation and a coupled space–time optimization framework for articulated tractor-trailer planning. By deforming trajectories directly in continuous free space, the method eliminates repeated safe-corridor construction, greatly accelerates computation, and reduces both curvature and travel time. Building on this framework, I conducted full-scale trials with a 7 m tractor-trailer (1 : 1 real vehicle). Apollo’s native lateral LQR and longitudinal PID controllers, together with the PIX-series CAN-bus protocol, were re-engineered as ROS nodes; for comparison, I also implemented an MPC controller. All components were seamlessly integrated with the planner, yielding a complete perception-mapping-planning-control loop. Real-world tests—spanning indoor/outdoor transport, loading/unloading, and narrow-aisle navigation—achieved a mean tracking error of ≈ 0.11 m, satisfying industrial-grade precision requirements..
Tested the FastSim simulator, integrating it with drone localization (using Fast-LIO and LOAM) and autonomous exploration (using FUEL) algorithms for both structured and unstructured dark environments.
Developed an aerial-ground autonomous navigation system for drones using FAST-LIO and ego-planner. Achieved point cloud map relay sharing during exploration. Designed and built a quad-fisheye-based localization drone inspired by OmniNxt, implementing VIO-based localization using OpenVINS.
Deployed lightweight drones equipped with VIO (VINS-fusion) and autonomous flight using ego-planner. Conducted multi-drone VIO experiments using COVINS-G and tested DM-VIO, ORB SLAM, and ALOAM. Constructed an autonomous drone swarm employing a vision-based ego-swarm algorithm, enabling independent flight in both outdoor sparse-feature environments and indoor structured settings.
Developed a WiFi signal detection software for aerial-ground robots using PyQt and Linux network interfaces. Aerial-Ground Robot Communication Planning Mesh Networking Problem (Transmission of odometer information as well as map packets) with bandwidth, communication conditions and UAV energy as constraints for FUEL and ego-planner based UAV cluster exploration problem. (This idea is inspired by MRS Group of CTU, from Professor Saska's paper)
The future of unmanned transport technology is rapidly developing, and the scale of unmanned devices is expanding,
while labour costs are rising and demand for logistics is increasing. Against this background, a project was proposed to achieve
efficient logistics and distribution by integrating multiple unmanned devices so that they can operate in concert and relay distribution under different demands.
Responsible for:
Implemented RRT and RRT* path planning and navigation for a quadruped robot in Gazebo; performed rendering simulations for both drones and quadruped robots using Blender.
Conducted GPS spoofing experiments using HackRF, validated spoofing on simulated target robots, and performed market research on potential solutions.
Also responsible for technical feasibility research and organisational design.
Awards:
2023 Outstanding Camper, Redbird Challenge Camp, HKUST
Project Introduction: National University Robotics Competition RoboMaster is a robotics event jointly sponsored by several prestigious organizations and hosted by DJI.
Responsible for:
1. Laboratory creation: Negotiated with leaders and secured funding to build the robotics lab from scratch (collaboration with multiple schools, ~40 people).
2. Team Management: Managed the project using Coding and Feishu; established lab safety systems.
3. Financial management: Established a triple approval process, obtaining motor sponsorship from DirectDrive.
4. Robot control debugging: Tuned PID parameters under freeRTOS, applied cascaded PID control and inverse kinematics for mecanum/omni wheels.
5. Wiring design: Completed design and actual wiring for Infantry, Hero, and Sentry robots.
6. Algorithm debugging: Assisted in camera selection, calibration, LIDAR and multi-camera fusion, radar calibration, YOLO detection on ROS, 3D printing, and Gazebo simulation.
Awards:
3rd place in the National RoboMaster Championship, 2023 (supported by DJI)
2nd place in the National RoboMaster 3v3 League (Shanxi Division), 2023 (supported by DJI)
3rd place in the National RoboMaster 3v3 League, 2022 (supported by DJI)
Project Description: A youth robotics event under FIRST, supported by NASA and Qualcomm, to develop robots capable of fetching and grasping specific targets under both automatic and manual control.
Responsibilities:
1. External liaison and engineering documentation.
2. Participation in mechanical construction and SolidWorks simulation design.
Awards:
Individual Award & Incentive Award, 2018 FIRST Robotics Competition (Suzhou)