AUTOWARE-INTEGRATION

A repository for prediction integration in EDGAR project. containing the tum_prediction package and the runtime evaluation results.

Recommand environment: Ubuntu 22.04 + ROS2 Humble

Directory Structure

├── ...  
├── runtime evaluation results 
       ├── ...  
├── src/tum_prediction                              # tum_prediction package
       ├── config
            ├── map_and_node_params.yaml            # Yaml file for ROS2 node
       ├── launch                                   # Contains two launch files for prediction nodes
            ├── parallel_map_based_prediction.launch.xml
            ├── routing_based_prediction.launch.xml
       ├── encoder_model_weights                    # Contains weights for encoder model
            ├── ...
       ├── prediction_model_weights                 # Contains weights for prediction model
            ├── ...
       ├── norms                                    # Contains weights for normalization
            ├── ...
       ├── sample_map                               # Contains two version of test maps
            ├── DEU_GarchingCampus-1
            ├── old_DEU_GarchingCampus-1
       ├── test                                     # Contains pytest files
            ├── ...
       ├── tum_prediction                           # Contains all nodes and utilities
            ├── __init__.py
            ├── image_path_generator_pil.py
            ├── image_path_generator.py
            ├── map_loader.py
            ├── nn_model.py
            ├── node_map_based_prediction.py        # Key node for integration testing
            ├── node_methods_test.py
            ├── node_path_evaluation.py
            ├── node_routing_based_prediction.py    # Key node for NN-based prediction
            ├── node_visualization.py
            ├── original_path_generator.py
            ├── utils_interpolation.py
            ├── utils_nn_model.py
            └── utils_tier4.py
       ├── package.xml                              # ROS package and dependencies definition.
       └── setup.py                                 # setup.py for ROS package
├── build.sh                                        # sh file for building the package
├── requirements.txt                                # Pip requirements
└── Readme.md 

Two options for package development

Following will be an instruction about how to set up local environment.

1. If you want to further develop the package, you could start from Setup locally and you may not need Run with docker and microservices.
2. But if you only want to test or see the result of the neural network-based trajectory prediction, you can jump to Run with docker and microservices.

1.0 Setup locally

1.1 Install Autoware

If the environment of Planning simulation is already there, then pass this step.

https://autowarefoundation.github.io/autoware-documentation/main/installation/

1.2 Git clone the autoware-integration repository to local directory

https://gitlab.lrz.de/prediction/autoware-integration

1.3 Build package

Open a terminal, and run the following command. It's recommanded to build like this, or you can also use pip install all requirements and then use colcon build.

cd ~/autoware-integration
chmod +x build.sh
./build.sh

1.4 Run the simulation

In one terminal: run the planning simulation.

ros2 launch autoware_launch planning_simulator.launch.xml map_path:=$HOME/autoware-integration/src/tum_prediction/sample_map/DEU_GarchingCampus-1 vehicle_model:=sample_vehicle sensor_model:=sample_sensor_kit

In another terminal: run the routing_based_prediction_node.

ros2 launch tum_prediction routing_based_prediction.launch.xml map_path:=$HOME/autoware-integration/src/tum_prediction/sample_map/DEU_GarchingCampus-1

You can change the map_path here to your map to test another map.

2.0 Run with docker and microservices

When you only want to see the capability of the neural network-based prediction and don't want to develop it, a good choice is to launch them with docker. But to do so, you need to make sure your computer has at least 80GB free space!

2.1 Git clone the repo tum_launch to local directory

https://gitlab.lrz.de/av2.0/tum_launch

And choose branch prediction_integration.

After this, you need to clone the repository autoware-integration repository. And change the map_path parameter in /tum_launch/compose_launch/.env to the map in this repositories, which is:

map_path=/home/#{user}/autoware-integration/src/tum_prediction/sample_map/DEU_GarchingCampus-1

2.2 Install Docker

If the environment of Docker is already there, then pass this step.

https://docs.docker.com/engine/install/ubuntu/

2.3 Download dockers to local machine using docker compose

Run in a terminal:

sudo docker login gitlab.lrz.de:5005
Username: ge89yes
Password: [Use your Username and Password here!]
cd ~/tum_launch/compose_launch
sudo docker compose -f rviz.yml up

After this, open another terminal and run

cd ~/tum_launch/compose_launch
sudo docker compose --env-file .env -f tum_prediction_sim.yml up

2.3.1 Troubleshooting for rviz

If nothing show up after you running sudo docker compose -f rviz.yml up ,you can:

Execute xhost +"local:docker@" on your host machine. Or startx before run sudo docker compose -f rviz.yml up

2.4 Run the simulation by microservices

Docker files should be run automatically after step 2.3.

3.0 Check the prediction results in autoware planning simulation

After the step 1 or step 2, click on the RVIZ software opened by Planning Simulation.

Then find the Add button of the Displays part at the top left corner, -> ckick it and slect By topic -> scroll down that small page to the bottom -> click MarkerArray in /nn\_pred\_path\_marker topic -> click OK button

There should be a new display in "Displays" area.

Then click the 2D Pose Estimate to add a car on the map, and click 2D Dummy Car to add an object to be followed. Instructions about how to do this step: https://autowarefoundation.github.io/autoware-documentation/main/tutorials/ad-hoc-simulation/planning-simulation/

4.0 Further Test with another map

First, please make sure your map directory looks similar to this:

├── NAME_OF_MAP 
       ├── lanelet2_map.osm         # OSM map that can be loaded by UTM projector
       ├── map_config.yaml          # Contains map_origin parameters
       ├── map_projector_info.yaml  # Might not be needed for old Autoware version
       ├── ...

4.1 If you followed Step 1

1. Change the map_path parameter for two launch files at Run the simulation.
2. Run two launch files again.

4.2 If you followed Step 2

1. Change the map_path parameter in /tum_launch/compose_launch/.env to your map path.
2. Run dockers again.

New and old version of autoware map

Since the map loader in Autoware is changed in the latest version(23.11.2023), one file is added to the map directory: map_projector_info.yaml. To use the old version map, we should modify this file ourselves, more information on website: https://autowarefoundation.github.io/autoware.universe/main/map/map_projection_loader/.