Quadruped robot navigation
EKF state estimation, A* planning and LiDAR processing for a legged platform at Geely Research Institute.
Stack: ROS · EKF · A* · LiDAR · C++ / Python Where: Geely Research Institute, 2022
Navigation for a quadruped platform. Legged locomotion breaks assumptions that wheeled navigation takes for granted — the body pitches and rolls with every step, odometry is derived from leg kinematics rather than wheel encoders, and LiDAR returns arrive from a sensor that is never level.
What I worked on
- State estimation — EKF fusing leg odometry with IMU, on a base whose motion is periodic and whose odometry drifts differently from a wheeled robot’s
- Path planning — A* over a costmap, with traversability meaning something different for a machine that can step over an obstacle a wheeled robot must go around
- Perception — LiDAR processing and ground segmentation from a continuously tilting sensor frame
- Integration — ROS nodes, message design, and the timing behaviour that determines whether the stack works on hardware or only in simulation
<Add: what the platform was, what the navigation had to achieve, and a measurable outcome — success rate, localisation error, distance covered autonomously. If a demo video exists and is not proprietary, link it; this is the single most convincing asset you have for a robotics application.>
No proprietary Geely material is reproduced here.