Add a new tool to measure end-to-end delay of the Autoware for sudden obstacles

[brkay54]

Checklist

• I've read the contribution guidelines.
• I've searched other issues and no duplicate issues were found.
• I've agreed with the maintainers that I can plan this task.

Description

In Autoware, accurately detecting the first response to sudden obstacles in various topics is challenging. There's a need for a tool that can create a test environment, listen to critical topics, and measure the system's reaction across different pipelines including Sensing, Perception, Planning, and Control.

This is a follow-up from:

• Investigate end-to-end delay in sudden obstacle braking response #5540

Purpose

The purpose of this tool is to enhance the reliability and responsiveness of Autoware by providing a means to measure and analyze the delay in reacting to sudden obstacles. This will help in optimizing the system's performance across all stages of its pipeline, ensuring safer autonomous driving capabilities.

Possible approaches

1. Develop a simulation environment within Autoware that can dynamically spawn obstacles.
2. Implement a listener module that subscribes to key topics (the topics such messages have been publishing: PredictedObjects, DetectedObjects, TrackedObjects, PointCloud2, Trajectory, AckermannControlCommand) to monitor and log reactions.
3. Design a measurement tool that calculates the time delay from obstacle-spawned time to the generation of a control command.
4. Incorporate a reporting mechanism to visualize and analyze the reaction times across different modules.

Definition of done

• A simulation environment capable of spawning sudden obstacles is implemented.
• A module for subscribing and listening to the necessary topics is developed.
• The tool can accurately measure reaction times across Sensing, Perception, Planning, and Control pipelines.
• A comprehensive report detailing the reaction times and suggesting areas for optimization is produced. #7051
• Merge the PRs that solve the issue: Make it possible to detect publishing times of the messages in the pipelines #6255
  • feat(autoware.repos): add autoware_internal_msgs repository autoware#4212
  • feat(autoware_internal_msgs): add PublishedTime debug info message autoware_internal_msgs#1
  • feat(tier4_autoware_utils): add published time debug class into utils #6440
  • feat: add published_time publisher debug to packages #6490
• Merge the PRs to close this issue.
  • feat(autoware_launch): add argument to enable/disable simulation time autoware_launch#886
  • feat(tier4_simulator_launch): add option to disable all perception related modules #6382
  • feat(reaction_analyzer): add reaction anaylzer tool to measure end-to-end delay in sudden obstacle braking response #5954

[doganulus]

create a test environment, listen to critical topics, and measure the system's reaction

It would be nice to write down these topics of interest, valid test conditions, and acceptable/unacceptable values from such measurements. For example,

• If the value in /namespace/topic_name is greater than 0, then /namespace/another_topic_name will be greater than 5 within 10 milliseconds.
• Given /namespace/topic_name equals to ENUM_VALUE and /namespace/another_topic_name is greater than 20, then /namespace/topic3 is always true.

What could be other things/constructs you find useful in writing such sentences for such analysis?

[brkay54]

Hi @doganulus thank you for the comment. Currently, we are working on merging this package. However, in the coming days, I am planning to add a detailed report on this issue or show and tell (TBD).

Actually, in this report, I am planning to show 2 different kinds of tests for each node:

• 1 - Default Time Consumption: In this test, I am going to show the processing times of each message in each pipeline by using published_time_publisher.

• For more information please see: Make it possible to detect publishing times of the messages in the pipelines #6255

This will show us how much time each node takes when it is executed. I am going to list the average calculation times of each node in the Sensing, Perception, and Planning pipelines.

• 2 - Reaction Response Time: In this test, I will measure the reaction times when an obstacle is spawned suddenly in the Perception, Planning, and Control pipelines.
  Reaction Time = (published time of the first reacted message) - (spawn time of the object)

Here are the topics of interest for Reaction Response Time tests:

• For Planning & Control pipeline:

        obstacle_cruise_planner (or obstacle_stop_planner):
          topic_name: /planning/scenario_planning/lane_driving/trajectory
          message_type: autoware_auto_planning_msgs/msg/Trajectory
        scenario_selector:
          topic_name: /planning/scenario_planning/scenario_selector/trajectory
          message_type: autoware_auto_planning_msgs/msg/Trajectory
        motion_velocity_smoother:
          topic_name: /planning/scenario_planning/motion_velocity_smoother/trajectory
          message_type: autoware_auto_planning_msgs/msg/Trajectory
        planning_validator:
          topic_name: /planning/scenario_planning/trajectory
          message_type: autoware_auto_planning_msgs/msg/Trajectory
        trajectory_follower:
          topic_name: /control/trajectory_follower/control_cmd
          message_type: autoware_auto_control_msgs/msg/AckermannControlCommand
        vehicle_cmd_gate:
          topic_name: /control/command/control_cmd
          message_type: autoware_auto_control_msgs/msg/AckermannControlCommand

• For Perception pipeline:

        common_ground_filter:
          topic_name: /perception/obstacle_segmentation/single_frame/pointcloud_raw
          message_type: sensor_msgs/msg/PointCloud2
        occupancy_grid_map_outlier:
          topic_name: /perception/obstacle_segmentation/pointcloud
          message_type: sensor_msgs/msg/PointCloud2
        multi_object_tracker:
          topic_name: /perception/object_recognition/tracking/near_objects
          message_type: autoware_auto_perception_msgs/msg/TrackedObjects
        lidar_centerpoint:
          topic_name: /perception/object_recognition/detection/centerpoint/objects
          message_type: autoware_auto_perception_msgs/msg/DetectedObjects
        obstacle_pointcloud_based_validator:
          topic_name: /perception/object_recognition/detection/centerpoint/validation/objects
          message_type: autoware_auto_perception_msgs/msg/DetectedObjects
        decorative_tracker_merger:
          topic_name: /perception/object_recognition/tracking/objects
          message_type: autoware_auto_perception_msgs/msg/TrackedObjects
        detected_object_feature_remover:
          topic_name: /perception/object_recognition/detection/clustering/objects
          message_type: autoware_auto_perception_msgs/msg/DetectedObjects
        detection_by_tracker:
          topic_name: /perception/object_recognition/detection/detection_by_tracker/objects
          message_type: autoware_auto_perception_msgs/msg/DetectedObjects
        object_lanelet_filter:
          topic_name: /perception/object_recognition/detection/objects
          message_type: autoware_auto_perception_msgs/msg/DetectedObjects
        map_based_prediction:
          topic_name: /perception/object_recognition/objects
          message_type: autoware_auto_perception_msgs/msg/PredictedObjects

These are the topics I am going to measure the reaction times. Please feel free to share any suggestions or comments with me!

[doganulus]

Thank you @brkay54. I will wait for your report. Looks very useful.

[stale]

This pull request has been automatically marked as stale because it has not had recent activity.

[brkay54]

I created the following document by using the reaction_analyzer before.
The tests marked by yellow color indicate more delay caused by this issue.

• Document: GoogleDocs