AirSim Development Environment on Azure#

This document explains how to automate the creation of a development environment on Azure and code and debug a Python application connected to AirSim using Visual Studio Code

Automatically Deploy Your Azure VM#

Click the blue button to start the Azure deployment (The template is pre-filled with the recommended virtual machine size for the use cases of the following two tutorials)


Note: the VM deployment and configuration process may take 20+ minutes to complete

Regarding the deployment of the Azure VM#

  • When using an Azure Trial account, the default vCPU quota is not enough to allocate the required VM to run AirSim. If that's the case, you will see an error when trying to create the VM and will have to submit a request for Quota increase. Be sure to understand how and how much you are going to be charged for the use of the VM

  • To avoid charges for the Virtual Machine usage while not in use, remember to deallocate its resources from the Azure Portal or use the following command from the Azure CLI: 

    az vm deallocate --resource-group MyResourceGroup --name MyVMName

Code and debug from Visual Studio Code and Remote SSH#

  • Install Visual Studio Code
  • Install the Remote - SSH extension
  • Press F1 and run the Remote - SSH: Connect to host... command
  • Add the recently create VM details. For instance, AzureUser@11.22.33.44
  • Run the Remote - SSH: Connect to host... command again, and now select the newly added connection.
  • Once connected, click on the Clone Repository button in Visual Studio Code, and either clone this repository in the remote VM and open just the azure folder, or create a brand new repository, clone it and copy the contents of the azure folder from this repository in it. It is important to open that directory so Visual Studio Code can use the specific .vscode directory for the scenario and not the general AirSim .vscode directory. It contains the recommended extensions to install, the task to start AirSim remotely and the launch configuration for the Python application.
  • Install all the recommended extensions
  • Press F1 and select the Tasks: Run Task option. Then, select the Start AirSim task from Visual Studio Code to execute the start-airsim.ps1 script from Visual Studio Code.
  • Open the multirotor.py file inside the app directory
  • Start debugging with Python
  • When finished, remember to stop an deallocate the Azure VM to avoid extra charges

Code and debug from a local Visual Studio Code and connect to AirSim via forwarded ports#

Note: this scenario, will be using two Visual Studio Code instances. The first one will be used as a bridge to forward ports via SSH to the Azure VM and execute remote processes, and the second one will be used for local Python development. To be able to reach the VM from the local Python code, it is required to keep the Remote - SSH instance of Visual Studio Code opened, while working with the local Python environment on the second instance

  • Open the first Visual Studio Code instance
  • Follow the steps in the previous section to connect via Remote - SSH
  • In the Remote Explorer, add the port 41451 as a forwarded port to localhost
  • Either run the Start AirSim task on the Visual Studio Code with the remote session as explained in the previous scenario or manually start the AirSim binary in the VM
  • Open a second Visual Studio Code instance, without disconnecting or closing the first one
  • Either clone this repository locally and open just the azure folder in Visual Studio Code, or create a brand new repository, clone it and copy the contents of the azure folder from this repository in it.
  • Run pip install -r requirements.txt inside the app directory
  • Open the multirotor.py file inside the app directory
  • Start debugging with Python
  • When finished, remember to stop an deallocate the Azure VM to avoid extra charges

Running with Docker#

Once both the AirSim environment and the Python application are ready, you can package everything as a Docker image. The sample project inside the azure directory is already prepared to run a prebuilt AirSim binary and Python code using Docker.

This would be a perfect scenario when you want to run the simulation at scale. For instance, you could have several different configurations for the same simulation and execute them in a parallel, unattended way using a Docker image on Azure Container Services

Since AirSim requires access to the host GPU, it is required to use a Docker runtime that supports it. For more information about running AirSim in Docker, click here.

When using Azure Container Services to run this image, the only extra-requirement is to add GPU support to the Container Group where it will be deployed.

It can use either public docker images from DockerHub or images deployed to a private Azure Container Registry

Building the docker image#

docker build -t <your-registry-url>/<your-image-name> -f ./docker/Dockerfile .`

Using a different AirSim binary#

To use a different AirSim binary, first check the official documentation on How to Build AirSim on Windows and How to Build AirSim on Linux if you also want to run it with Docker

Once you have a zip file with the new AirSim environment (or prefer to use one from the Official Releases), you need to modify some of the scripts in the azure directory of the repository to point to the new environment: - In azure/azure-env-creation/configure-vm.ps1, modify all the parameters starting with $airSimBinary with the new values - In azure/start-airsim.ps1, modify $airSimExecutable and $airSimProcessName with the new values

If you are using the docker image, you also need a linux binary zip file and modify the following Docker-related files: - In azure/docker/Dockerfile, modify the AIRSIM_BINARY_ZIP_URL and AIRSIM_BINARY_ZIP_FILENAME ENV declarations with the new values - In azure/docker/docker-entrypoint.sh, modify AIRSIM_EXECUTABLE with the new value

Maintaining this development environment#

Several components of this development environment (ARM templates, initialization scripts and VSCode tasks) directly depend on the current directory structures file names and repository locations. When planning to modify/fork any of those, make sure to check every script and template to make any required adjustment.