docker

Avoidance with Docker

Prerequisites

Install docker and docker-compose:

$ apt install docker-compose

When using the X server in a docker container, you need to run the following command, which has (not necessarily big) security implications (more on this here):

$ xhost +local:root

This can be undone with:

$ xhost -local:root

Quick Start

A wrapper script has been created to help running the planners in simulation or on the drone.

Run the simulation

To run the simulation for the local planner, run:

$ ./run.py local

Similarly, for the global planner:

$ ./run.py global

Run in production on the Aero

In production mode, the simulator is not started and the modules connect to the FCU on the Aero. There are two modes:

• debug: start a vpn server, allowing you to connect your machine into the ROS network on the Aero and e.g. run RViz on your machine.
• release: just start the production nodes.

To run the local planner in debug mode:

$ ./run.py local --prod-debug

To run the local planner in release mode:

$ ./run.py local --prod-release

To run the global planner in debug mode:

$ ./run.py global --prod-debug

To run the global planner in release mode:

$ ./run.py global --prod-release

Running the demo

Find instructions here.

Deploying with Docker

Deployment configurations are located in "global_planner/global-planner-prod".

The global planner can be deployed in two flavours: debug and release. They are currently the same, but in the future, the debug flavour will allow debugging from a remote machine (i.e. one will be able to run rviz on a computer while the containers are running on the drone).

Find instructions to run the release flavour here.

Developing with Docker

Development configurations are located in "global_planner/global-planner-dev".

In order to develop in good conditions, it is always nice to compile the code on the host. But installing the whole simulation on your host requires that you install everything, including Gazebo and a whole lot of ROS packages. Setting up the right environment might be troublesome if you happen to not be developing on Ubuntu 16.04. Turns out that if you only work on the avoidance node, chances are that you don't need to modify anything in Gazebo or Mavros. In this case, you could just run Gazebo and Mavros on docker containers, and connect your avoidance node directly from the host. For this reason, different containers have been created (they are defined in components):

• sitl-avoidance-server, in sitl-server, is running gazebo in server mode (requires X server).
• sitl-avoidance-gui, in sitl-gui, runs gazebo and rviz (requires X server).
• mavros-avoidance is based on mavros and obviously runs mavros.
• avoidance-node runs the actual avoidance system, composed of different nodes (including the octomap one).

You will find here instructions on how to run the global_planner containers in dev mode.