Off-road/Racing Working Group Meeting 2025/11/13

[pojenwang]

Administrative

Previous Meeting Minutes

Attendees

• Po-Jen Wang (AWF)
• David Walmroth (Open AD Kit / Pix-Moving)
• Jack Silberman (UCSD)

Minutes: Po-Jen Wang

SAM-based Labelling Tool

• Previously developed an interactive image labeling tool based on Meta’s Segment Anything Model (SAM) V1 and OpenCV, designed for creating and refining segmentation masks for datasets.
• The quality/accuracy of the drivable region mask largely depends on the point locations (prompt) used to define the drivable region/road.
• Use "m" key for fast-labelling: Previously proposed to apply points from the last frame for auto-label, though these points don't always land on the most optimal location. Use the cursor to adjust to the optimal location and press "m" key to generate a segmentation mask on the next image.
• Use "+" and "-" to increase/decrease brush radius.
• Finished labeling drivable regions for the CARLA_Mining datasets with 3539 images.
• Downloaded UCSD dirt-road 4k video from YouTube. Extracted the frames at 2Hz and resized them to 1080p. A total of 9283 images were extracted. Labelling these images is in progress.
• Abijit from Virginia Tech (CoE member) will be sharing their off-road dataset soon.
• Retraining the FreespaceSeg model with the updated off-road dataset will be the next step.

IV2026 Competition Update

• Rahul/David recently met with the conference chairs to plan the competition. David has identified some tracks near the Detroit hotel and MCity where we can get started and map out the track.

• Collaborating with Clemson's TinkerTwin to build off-road roboracer sim for the competition, though the initial meeting was pushed back to after Thanksgiving.

• Alex CARBALLO (Gifu University, CoE member) suggested hosting an Autoware-based go-kart competition in MCity at IV. In addition to speed, introduce

  • 1. dynamic objects on the track, such as pedestrian and car dummies
  • 2. GPS jammers to reduce the dependency on GNSS
  • 3. water sprinklers and/or smoke generators to simulate rain/fog (in case weather is clear)

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