Phase 2 for Autoware Core Implementation

[mitsudome-r]

Proposal

We have finished porting Autoware Core Phase 1 packages in March. I would like to start discussing about porting some extra packages and making improvements on Autoware Core so that we can release it under ROS Buildfarm for wider use. This includes the following working items:

• Port extra packages to ensure that the necessary topics(such as orientation inputs to EKF and PredictedObjects messages) for Autoware are available
• Separate the Autoware RViz Plugin from Universe so it can also be used in Core.
• Improve the documentation for Autoware Core
• Improve default parameters to work out of the box

Design

I have discussed with @TakaHoribe for newly ported packages to Autoware Core. Compared to the intial idea in Phase 1, we are also thinking of porting ndt_scan_matcher since it is also commonly used package by many users, and GNSS poser also isn't enough to provide accurate orientation information for stable localization.

Ported Packages

Perception

• euclidean_cluster: group obstacle points as objects information
• convert_to_prediction: convert detected objects messages into predicted object message type

Localization

• gyro_odometer: Merge Vehicle Velocity and IMU information to estimate twist information
• ndt_scan_matcher: Localize Vehicle position within a map

Map

• map_visualizer: converts map message into visualization topic

Planning

• detection_area: Stops ego vehicle whenever objects are detected within specific area defined in Lanelet2 Map.

Execution Plan

I will create GitHub Project board for phase two with the list of issues that we should resolve.
We are using this Spreadsheet to maintain assignee for each package.

[liuXinGangChina]

Hi Mits san @mitsudome-r

Great plan for Phase II porting, I have 2 comments for it

1. since ndt have multi implementation which one shall we port, NDT-OMP ?
2. How do you define simplfied autoware_euclidean_cluster?

please feel free to involve me in discussion related to previous comment

Best regards

心刚

[mitsudome-r]

1. We would like to port non-GPU implementation
2. We can port the euclidean_cluster package, but change the output message type from DetectObjectWithFeatures message to DetectedObject message

convert_to_prediction package is a newly created package that just does the conversion of the message type.