3 Steps to Deploy Autoware.universe with Using Carla_0.9.15 on Jetson Orin

This repository is based on the github repository from TUM. LINK

Thanks a lot for their brilliant work, if you are interested in their paper, you can refer to this link [PDF]

Autoware.universe source codes can be found from their github repository. LINK

In this repository, we did some revisions to use REV-shuttle-bus in the Carla simulator, which has different sensors, also the autoware.universe version we used will be changed to our version later.

Requirements

Software

Carla 0.9.15 (the newest version)
Autoware.universe (V1.0 Branch, will be changed to other version)
Autoware-Carla-Bridge (Main Branch)
ROS2 (Humble version)
Ubuntu 22.04 (LTS Released version)
Docker

Hardware

Jetson Orin Developer kit is an awesome platform for robots and Autonomous Vehicle.

One all 1 Gigabyte Switcher or Router (for communication between your host Ubuntu machine with Jetson Orin).
One Jetson Orin Develop kit 32GB/64GB with nvme ssd 500G external storage.
One Ubuntu 22.04 Host Machine.
Two or more Cat 5e/6 or above level cables

All devices are required to connect with each other in a local network. All 1 Gigabyte specs are better.

Note: To illustrate the following steps easier, I use "Host Machine" stands for the machine which has a standard GPU device like RTX2070 or 30/40 series and it 's also the one which will deploy a Carla 0.9.15 simulator and Autoware-Carla-Bridge ROS2 packages, and use "Jetson Orin" stands for the embedded system which will be used for deploying the Autoware.universe stack.

Before you go through the following steps, I suggest you pull some necessary docker images using following commands, so after you finish reading this document, you can start deploying right away:

 $ docker pull carlasim/carla:0.9.15
 $ docker pull tumgeka/carla-autoware-bridge:latest
 $ docker pull 1429053840/autoware.universe-carla-0.9.15:humble-20240215-cuda-arm64-v0.1

3 Steps to deploy

STEP-1: Deploy the carla-0.9.15 simulator

This step is based on the carla official docker image, you have to pull the image and then run it on your Host machine.

You have to check the rpc-port is not used on your Host machine.

$ docker pull carlasim/carla:0.9.15
$ docker run --rm --privileged --gpus all --net=host -e DISPLAY=$DISPLAY carlasim/carla:0.9.15 /bin/bash ./CarlaUE4.sh -carla-rpc-port=1403

STEP-2: Deploy the autoware-carla-bridge

This step is also based on the docker image issued by TUM team. Setting the timeout to a larger volume is necessary on my own laptop, the default value is 5000.

$ docker pull tumgeka/carla-autoware-bridge:latest
$ docker run -it --rm -e RMW_IMPLEMENTATION=rmw_cyclonedds_cpp --network host tumgeka/carla-autoware-bridge:latest
$ ros2 launch carla_autoware_bridge carla_aw_bridge.launch.py port:=1403 town:=Town10HD timeout:=10000

STEP-3: Deploy autoware.universe stack on Jetson Orin

Compile the autoware.universe is sometimes difficult, as there are various dependencies. So, I suggest that use our pre-built docker images is easier. These following commands need to run on the Jetson Orin terminals. As we want to use a Jetson Orin device to deploy the autoware.universe stack.

Note: This docker image is a little bit large and only supported on the ARM64 platform, so Jetson series are all suitable.

$ docker pull 1429053840/autoware.universe-carla-0.9.15:humble-20240215-cuda-arm64-v0.1
$ sudo apt-get install python3-rocker
$ rocker --user --nvidia --privileged --network host --x11 --volume $HOME/Documents  --volume $HOME/autoware -- 1429053840/autoware.universe-carla-0.9.15:humble-20240215-cuda-arm64-v0.1
$ ros2 launch autoware_launch e2e_simulator.launch.xml vehicle_model:=carla_t2_vehicle sensor_model:=carla_t2_sensor_kit map_path:=/autoware1.0_ws/Town10/

Note: When you run the rocker commands, and it may shows the error like the following messages: invoking the NVIDIA Container Runtime Hook directly (e.g. specifying the docker --gpus flag) is not supported.

You merely need to change the "--gpus all" to "--runtime=nvidia". like the example commands below: "Your commands will be different with mine, you should copy the commands, after you run the rocker --user --nvidia xxxx command"

$ docker run  --rm -it --network host   --runtime=nvidia --privileged  -e DISPLAY -e TERM   -e QT_X11_NO_MITSHM=1   -e XAUTHORITY=/tmp/.dockerafc7hfmf.xauth -v /tmp/.dockerafc7hfmf.xauth:/tmp/.dockerafc7hfmf.xauth   -v /tmp/.X11-unix:/tmp/.X11-unix   -v /etc/localtime:/etc/localtime:ro  8ea8cd5cadfe

Optional

Spawn NPCs

If you want to add some NPCs in the carla you can use the python script in the autoware-carla-brdige docker container. Must set the port to the same with the front steps. This generate_traffic.py should be used in the docker container's terminal of the STEP-2(tumgeka/carla-autoware-bridge:latest).

$ python3 src/carla_autoware_bridge/utils/generate_traffic.py -p 1403

Deploy autoware.universe stack on Host Machine

Compile the autoware source codes on your host machine locally, if you do not want to deploy on the jetson embedded systems.

$ colcon build --symlink-install --cmake-args -DCMAKE_BUILD_TYPE=Release

If you want to compile the source codes on the Host Machine with a GPU card, you may need to solve some dependencies problems.

Or you can use my customized docker image for x86 system on dockerhub

$ docker pull 1429053840/autoware.universe-carla-0.9.15:humble-20240215-cuda-amd64-v0.1
$  rocker  --user --nvidia --privileged --network host --x11 --volume $HOME/CARLA --volume $HOME/Documents --  1429053840/autoware.universe-carla-0.9.15:humble-20240215-cuda-amd64-v0.1
$ ros2 launch autoware_launch e2e_simulator.launch.xml vehicle_model:=carla_t2_vehicle sensor_model:=carla_t2_sensor_kit map_path:=/autoware1.0_ws/Town10/

Local workflow

Local workflow will be added later, add a Dockerfile, released and revised source codes.

TODO List

Add another shuttle-bus model in carla simulator to use
Simplified a Dockerfile for users to build the Autoware.universe docker image locally.
Uploads the revised autoware.universe codes for solving the problems of starting the perception module in the autoware.universe stack

Bugs Shooting

docker run error, the error info is like the followings:

docker: Error response from daemon: failed to create task for container: failed to create shim task: OCI runtime create failed: runc create failed: unable to start container process: error during container init: error running hook #0: error running hook: exit status 1, stdout: , stderr: Auto-detected mode as 'csv'
invoking the NVIDIA Container Runtime Hook directly (e.g. specifying the docker --gpus flag) is not supported.Please use the NVIDIA Container Runtime (e.g. specify the --runtime=nvidia flag) instead.: unknown.

You have to use the "--runtime=nvidia " flag to run using docker run commands.

Reference

[1] Kaljavesi, Gemb, et al. "CARLA-Autoware-Bridge: Facilitating Autonomous Driving Research with a Unified Framework for Simulation and Module Development." arXiv preprint arXiv:2402.11239 (2024).

Blog link: https://www.bluenote.top/2024/03/12/008-AutonomousDriving/02-Autoware.universe/%E5%A6%82%E4%BD%953%E6%AD%A5%E5%AE%9E%E7%8E%B0%E5%9C%A8Jetson-Orin%E4%B8%8A%E8%81%94%E8%B0%83Autoware-Universe%E5%92%8CCarla-0-9-15%EF%BC%9F/index.html English-version

https://www.bluenote.top/2024/03/12/008-AutonomousDriving/02-Autoware.universe/How%20to%20Joint%20testing%20between%20Autoware-Universe%20and%20Carla-0-9-15%20on%20Jetson%20Orin?/ Chinese-version

Github repo: https://github.com/TUMFTM/Carla-Autoware-Bridge.git

LiZheng1997 / Autoware.universe-with-carla-0.9.15