.png&blockId=245cc7c3-f3fb-80a5-b36a-e92fb05407bb&width=3600)
Auto valet parking demo
•
(0:00~0:45) In the first drive, the driver presses the Map training button and drives manually to learn the map of the parking environment.
•
(0:48~2:00) In the second drive, the driver presses the localization button and gets out, and the car drives automatically to park at the specified location.
Purpose of the project
•
Develop an ultra-lightweight visual SLAM algorithm for autonomous parking software
•
Position estimation and mapping that runs in real-time on embedded platforms for automotive with limited computational resources
What we contributed
•
Lightweighting of non-linear optimization used by frontend and backend and porting to embedded platforms
•
Developed LiDAR GT data generation algorithm for Visual SLAM algorithm evaluation
•
Built automated pipeline for Visual SLAM algorithm evaluation and created tools to analyze evaluation results