Projects
Industrial Projects
WeRide - Multi-sensor Calibration Algorithm
Research Intern | 2024.12-2025.6
- Multi-modal sensor calibration algorithm design and optimization without calibration objects and initial mechanical parameter requirements
- LiDAR-Camera extrinsic parameter validation, operational health monitoring, and online calibration
- Tech Stack: Multi-sensor fusion, online calibration, robust estimation
Academic Research Projects
Optimal and Reinforcement Learning Control for Parallel Multi-DOF Wheel-Legged Robot
Undergraduate Thesis Project | 2022.11-2023.05
Established a simplified second-order inverted pendulum model for a five-link parallel mechanism, derived dynamic equations using Newton-Euler method, and designed an MPC controller with validation in MATLAB Simscape Multibody simulation.
Key Contributions:
- Model Predictive Control: Implemented SQP solver through CasADi, added model-based spatial controllers to extend robot motion to 3D space, achieving terrain adaptation, dynamic suspension, and center-of-gravity adjustment
- State Estimation: Designed multi-sensor fusion state observer using dynamic equations, onboard IMU, and joint motor encoders to determine robot-ground contact state and horizontal velocity, suppressing disturbances and model mismatch effects
- Reinforcement Learning: Built multi-terrain environments using IsaacGym, designed targeted rewards for curriculum learning, employed asymmetric actor-critic training to achieve stable locomotion on slopes, stairs, and gravel surfaces
| Simulation Code | Deployment Code |
Multi-sensor Hardware Timestamp Synchronization System
Independent Project | 2023.09-2023.10
Designed LiDAR-Camera-Inertial-GNSS hardware time synchronization scheme. GPS PPS output is split into two channels: one combined with GPRMC messages inputs to LiDAR as synchronization pulse; the other inputs to STM32 external capture for frequency multiplication, outputting PWM to trigger camera acquisition.
Technical Highlights:
- Implemented camera acquisition completion and exposure event callback functions on computing platform, setting camera timestamps based on current frame exposure duration and time difference with GPRMC messages
- Added error handling for camera frame errors and PPS errors
- Achieved microsecond-level time synchronization precision exceeding PTP according to oscilloscope test results
Open Source Projects
Hunan University RoboMaster Team - Captain
Team Leader | 2022.03-2023.02
As team captain, responsible for technical direction formulation and project management, while deeply involved in algorithm development and system architecture design.
Mobile Target State Estimation Algorithm:
- Extracted target robot features through morphological operations, validated features using SVM, then extracted known relative position points for PnP to determine pose relative to camera
- Modeled target robot as rigid body with superposed uniform rotation and translation, acceleration following zero-mean Gaussian distribution (Singer model), using feature poses relative to camera as observations with particle filter for state estimation
Deep Learning Optimization:
- Modified lightweight CNN detector NanoDet for keypoint detection, added unsupervised teacher-student distillation, performed reparameterization + quantization pruning + OpenVINO deployment
- Completed overall framework for 22/23 team programs including MCU communication and camera ROS2 drivers
Educational Resource Contributions:
- Authored popular CV tutorial: Understanding CV and RoboMaster Algorithms
- Programming introduction knowledge base: Software Development Primer
Embedded Framework Development:
- Designed robot embedded framework basic_framework (600+ stars) with supporting debugging tutorials and workflows
- Established modern development paradigm using ARM-GNU toolchain, Ozone, and VSCode to replace legacy Keil
- Developed C++-based powerful_framework
- Framework deployable on all ST series MCUs, requiring only CubeMX pin-map reconfiguration
- Awarded RoboMaster 2023 Open Source Award Second Prize
Tech Stack: Computer vision, state estimation, deep learning, embedded development, ROS2, C++