In this series, I'd like to cover Optimal Frenet Path Planning.
Especially on this post, I'll write about how we decided to take this way, the history.
Last year, We had a frequency issue when we operated our car with all nodes on including LiDAR, Camera, and YOLO v4. Its hz didn't go down until we turned on Yolo v4, and it led the car to an unstable path tracking. Especially the speed of Waypoint_loader (which is a publisher) plummeted sharply from 10 hz to around 4hz.
So we first tried to tackle this issue with MultiThreadedSpinner and AsyncSpinner, but as you can see, it is a band-aid. We changed our direction to a heavy csv file that has over 1500 waypoint nodes. And if we have more sparse waypoint nodes and make a local path between them, then we could make the file and the entire system lighter other than Yolo. Overall, we were luckily told this path planning concept by a well experienced and rising star mate, and came to a conclusion to settle down on this !
Long story short, we decided to get in this path planning for lighter system and better performance.
However, the idea of this was very novel to me, i took quite a long time to just understand its basic concept.
This series would have a couple of posts less than 4 probably, the order is its basic concept, understanding of open source codes, and implementing our own frenet path planning.
Thank you guys! and always welcome any corrections and questions!
Vielen Dank, 감사합니다 ^_^
