AV Global Planner

Project Overview

Made for Queen's AutoDrive, I developed Lanelet2 based global planner implemented with Python and ROS2. This module uses the A* algorithm to return the shortest path between two lane segments, localizes the egovehicle to segment, and cancels obstructed routes in real time.

I developed a parsing script to read the HD map data and convert to Lanelet format, resolving many map conflicts/incompletions. The segment graph is searched to obtain the shortest path, and all centerline points within n+2 segments are packaged and transmitted via ROS2 to the trajectory tracking module.

The current GNSS position is converted to UTM frame, and used to determine the current occupancy of the egovehicle and all detected objects (3D position from LiDAR-camera overlay module).

Results & Analysis

With a significantly reduced timeline, we were able to first test the Autonomy stack 1 month before the competition. Fortunately, the first live test of the global planner integrated with the CAN interface was successful.

Takeaway

Modular software is good