2024-04-25 Racing Working Group Meeting

[pojenwang]

Administrative

Previous Meeting Minutes

Attendees

• Po-Jen Wang (AWF)
• Mitsudome-san (Tier.IV)
• David Walmroth (Open AD Kit / Pix-Moving)

Minutes: Po-Jen Wang

Topics

Autonomous Karting Series (AKS) Competition

• AKS competition website
• The competition will take place at the Purdue Grand Prix Track at Purdue University between May 20th - 24th, 2024
• The track length is 434 meters. The track width is 5 meters in most sections of the track. A small section of the track (between the first and second turn) the track width is 6 meters
• 6 teams registered this year (+1 compared to last year)

Competition Rules

• Two race categories:
  1. Prior-mapping allowed category: track data (map, centerline, etc) before the race is allowed. No cone on the sides of the track
  2. Reactive required category: No track data (map, centerline, etc) before the start of the race is allowed. With cones on the sides of the track.
• Two obstacles (white pyramid boxes) will be randomly placed with the left-right or right-left sequence at the straightaway. The sequence will be swapped between each consecutive lap. The go-kart must avoid obstacles or a penalty will be given
• Team rankings will be by the total time taken to complete 5 laps

UPenn Team Strategy

• Prior-mapping allowed category:
  -- Since there will be no cone on the side of the track. Lidar-based maps and localization will not be useful. It will be featureless
  -- Localization may be achieved with GNSS or vision-based localization
  -- Vision-based localization can be achieved by first generating a map based on the track outline from the satellite image. At run time, perform grass boundary detection and convert them into 2D Lidar format, and utilize particle filter to achieve localization. Image wrapping from uneven ground can affect the localization accuracy
  -- We will generate an optimized centerline and run a trajectory follower controller (RecordReplay_planner, pure-pursuit) to follow the trajectory. Will switch to gap follow when a collision with an obstacle is identified

• Reactive category:
  -- Cones will be placed on the sides of the track 3 meters apart from each other
  -- Track boundaries can be identified with grass boundary detection or (vision/Lidar) cone detection and connect nearby cones to close the gap
  -- Gap-follow type planners will be used to navigate through the track

• All of the above packages will be ported and run within the Autoware framework. Will run the test in real life and in the simulator

• A binary version of the simulator will be released along with the editor version of the simulator

Other topics:

• AWF Work Group overview: https://autoware.org/join-a-work-group/