Cargo Delivery ODD Kick-Off Meeting

Hello Autoware Community! I’m sure many of you have seen our media recently about the Autonomous Valet Parking demonstration that the Foundation (along with some very helpful member companies) completed a couple of months ago. Well, now it’s time for us to move on to our next Operation Design Domain (as chosen by the TSC): Cargo Delivery! This is an announcement for the kick-off meeting that we will be hosting to discuss features, architectural changes, milestones, epics, and all other planning aspects for the upcoming development period for this ODD.

Please feel free to join the meeting if you are interested in learning how the development process will proceed or you have input on the upcoming architectural changes to make Autoware.Auto more flexible and capable.

Note regarding timing: At the Autoware Foundation, we work with developers around the world to esnure the best-quality reference implementation of a self-driving stack in Autoware.Auto. Due to the scattered timezones and us heavily leaning on times that work best for United States participants recently, this kick-off meeting caters more to the European and Asian time-zones.

The call is scheduled for 2020-12-18T06:00:00Z2020-12-18T10:00:00Z. To get the call-in information, please visit the Autoware Shared Calendar at https://calendar.google.com/calendar/u/0/embed?src=autoware.org_6lol0ho5ft0217h8c60pi1fm30@group.calendar.google.com&ctz=Asia/Tokyo

We look forward to seeing many of you there!

5 Likes

Preliminary agenda:

  • (Approx. 30 minutes) Review of existing AVP architecture and unused features from AVP demo
  • (Approx. 1 hour) Discussion on standardization of existing inter-module APIs (messages/topics)
    • Sensor driver input
    • Sensor pre-processing output
    • Semantic map server output
    • Localization input/output
    • Planning input/output
    • Control input/output
    • Vehicle Interface input
  • (Approx. 2 hours) Overview and discussion of necessary feature additions to support Cargo Delivery ODD. Preliminary list:
    • Add camera-based object recognition and classification pipeline
      • Traditional image processing methods (OpenCV)
      • Neural-network-based image processing methods (utilizing TVM)
    • Add radar pipeline
      • Add radar-based object detection
      • Add multi-radar fusion
      • Discuss standard radar message (e.g.
    • Add multi-modal object fusion (camera, lidar, radar)
    • Add GNSS as localization source
    • Add IMU as odometry source
    • Add wheel encoders as odometry source
    • Test existing EKF implementation with multiple input sources
      • Consider UKF for highly discontiguous sources (GNSS)?
    • Consider implications of towing one or more trailers to navigation/planning/collision detection
    • Merge existing features of velodyne_driver and velodyne_node into community Velodyne driver and replace our driver completely

Other features for consideration/topics for discussion (as time allows):

  • Detected object path prediction
  • Static obstacle pre-filtering (e.g. map-based)
  • Using additional ROS-community-standard messages (e.g. vision_msgs, ackermann_msgs, etc.)