Skip to content

Latest commit

 

History

History

sample_autoware

README.md

CARET Analysis for Autoware

This page shows how to analyze Autoware with CARET

  1. Get Autoware project
  2. Install CARET
  3. Build Autoware with CARET
  4. Run Autoware to record trace data
  5. Create analysis report

1. Get Autoware project

2. Install CARET

  • Follow the instruction
  • Note:
    • This explanation assumes you install CARET to ${caret_dir} (e.g. export caret_dir=~/ros2_caret_ws/ )

3. Build Autoware with CARET

Build Autoware

  • Note:
    • Before building Autoware, CARET needs to be enabled like the following commands
    • It's also important to set BUILD_TESTING=Off
cd ${autoware_dir}
rm -rf build/ install/ log/

source ${caret_dir}/install/local_setup.bash
colcon build --symlink-install --cmake-args -DCMAKE_BUILD_TYPE=Release -DBUILD_TESTING=Off -DCMAKE_CXX_FLAGS="-w"

Check compilation (optional)

The following command checks if Autoware is built with CARET. It outputs warnings if a package is built without CARET. You can ignore them unless package names which you want to analyze are included in the warning message

cd ${autoware_dir}
ros2 caret check_caret_rclcpp ./

# Expected result
INFO    : 2023-01-17 14:12:28 | All packages are built using caret-rclcpp.

# Acceptable result (these packages are not necessary to analyze performance)
WARNING : 2022-08-25 18:14:31 | The following packages have not been built using caret-rclcpp:
 tier4_calibration_rviz_plugin
 initial_pose_button_panel
 autoware_auto_perception_rviz_plugin
 tier4_vehicle_rviz_plugin
 tier4_control_rviz_plugin
 localization_error_monitor

4. Run Autoware to record trace data

  • Note:
    • Before running Autoware, some environmental settings need to be done like the following commands
    • Please modify map_path and rosbag file for your environment
    • Make sure that object detection works and path is created when you set a 2D Goal Pose, so that you can analyze end-to-end path later

Copy topic filter

  • Autoware has lots of nodes and topics. Tracing all events causes huge trace data size and load. So, it's recommended to apply a filter to ignore less important nodes and topics
cd ${autoware_dir}
cp ${path-to-this-repo}/sample_autoware/caret_topic_filter.bash .

Run Autoware and start recording via CLI

  • Recording can be started via CLI while autoware is running
  • The trace data will be created in ~/.ros/tracing/session-yyyymmddhhmmss
cd ${autoware_dir}
ulimit -n 16384
source ${caret_dir}/setenv_caret.bash
source ./install/local_setup.bash
source ./caret_topic_filter.bash

ros2 launch autoware_launch logging_simulator.launch.xml map_path:=$HOME/work/rosbag_map/universe/sample-map-rosbag vehicle_model:=sample_vehicle sensor_model:=sample_sensor_kit

# on another terminal
cd ${autoware_dir}
source ./install/local_setup.bash
ros2 bag play ~/work/rosbag_map/universe/sample-rosbag

# on another terminal
source ${caret_dir}/install/local_setup.bash
ros2 caret record -f 10000 --light
## press enter to start recording, then press enter to stop recording

Validate trace data (optional)

ros2 caret check_ctf ~/.ros/tracing/session_yyyymmddhhmmss

5. Create analysis report

source ${caret_dir}/install/local_setup.bash
cd ${path-to-this-repo}/sample_autoware
sh ./run.sh
  • Before running the script, please modify the settings in sample_autoware/run.sh
    • trace_data: Set path to trace data
    • target_path_json: Use ./target_path_latest.json if you runs the latest Autoware or modify the json file
    • Setting files in this directory are just a sample, and may not work with your trace data.
  • Reports are created in sample_autoware/output/
    • report_{trace_data_name}/index.html: general analysis report
    • val_{trace_data_name}/index.html: validation report

FAQ

General

https://tier4.github.io/caret_doc/latest/faq/faq

Build

  • Build for a package which uses pcl_ros fails (e.g. static_centerline_optimizer , map_loader )

  • Build fails due to too few arguments to function ‘void ros_trace_rclcpp_publish(const void*, const void*, uint64_t)’

Recording

  • Trace data size is extremely small
    • If you use LTTng 2.13+, run the following command before starting Autoware
      • ulimit -n 65535

Analysis report

  • Path results in a created report is blank
    • Please find Target path not found error message in script log and modify target_path.json
    • If a node in target_path.json doesn't run at all while recording, the path will be blank
      • e.g. If route (2D GOAL) is not set, a path including planning module will be blank
    • target_path.json is just a sample and path may be changed as Autoware is modified