This project showcases real-time obstacle avoidance in UAVs by utilizing the ModalAI Voxl 2 Sentinel Drone with the PX4 flight stack and an Intel Realsense D435i camera to precisely detect obstacles in its environment. The detected object positions are incorporated as constraints in a non-linear programming-based path planning algorithm, allowing the UAV to autonomously generate safe flight paths with a defined distance threshold around these obstacles. To ensure continuous paths, reference velocities and accelerations are calculated at each waypoint. By accounting for room dimensions and drone dynamics as constraints in the optimizer, these paths ensure efficient and secure navigation, enabling the drone to reach its designated destination accurately and reliably.
This guide provides step-by-step instructions to launch the PX4 simulation, including setting up the necessary environment, starting essential services, and running trajectory and obstacle avoidance scripts.
Make sure docker is installed on the host linux pc.
- Pull the latest ubuntu 22.04 image by
docker pull ubuntu:22.04- Use the
Simulation Codesas the mounting point for your docker container so that you can access the files in docker as well as the host. A docker container can be opened by using the scriptSimulation Codes/run_simclassic_docker.sh. Make sure to edit the script for the right docker image.
The list of docker images on the host can be seen by using docker images.
- Run the following commands to install git and PX4 sim.
apt update && apt install git
apt-get install sudo # makes using sudo commands copied from internet easier to run
git clone https://github.com/PX4/PX4-Autopilot.git -b release/1.14 --recursive
./Tools/setup/ubuntu.sh # this installs prerequisties
pip uninstall em && pip install empy==3.3.4-
To verify the successful installation of the simulator, navigate to the PX4 folder and execute the command
make px4_sitl gz_x500_depth. If a window labeled "Gazebo Sim" appears, it indicates that the simulator has been installed correctly. This command will launch the drone in an empty world. To load the drone in a simulated environment, use the commandmake px4_sitl gz_x500_depth_baylands. -
Install ROS2 humble in the docker image by following instructions from ROS Docs
-
Install the apt version of Gazebo for ROS Humble by following this link.
-
Make a directory in the root of the workspace and clone ros_gz bridge.
mkdir -p ros2_gzbridge/src
cd ros2_gzbridge/src
git clone https://github.com/gazebosim/ros_gz.git -b humble
export GZ_VERSION=garden- Now go back to the root of the workspace and install dependencies for
ros_gz bridge, then execute a colcon build of the same:
cd ..
rosdep install -r --from-paths src -i -y --rosdistro humble
source /opt/ros/humble/setup.bash
colcon build- Commit the docker container by getting the name of the docker container by running
docker psand then:
docker commit name_of_docker_container name_of_image:versionThe name_of_image and version can be any desired names. Once you have specified these, exit the Docker container. This ensures that the terminal used for building is not the same one used for isolating build artifacts.
- Run the script to open the docker container again from step 2, with the required edits to change to respective image name and version from step 9. Source the built ros-gz_bridge:
source ros2_gzbridge/install/setup.bash- Make a folder called
modelsand run the script as given below to source gazebo. Make sure to edit the path in the script before it is run.
."/home/s/Downloads/Files/Simulation Codes/set_GZ_SIM_RESOURCE_PATH-2.sh"- Once it is sourced, you can run the command to open the PX4 simulator using the script provided below. This command will launch the simulator:
./Simulation Codes/run_baylands_total-1.shQGC can be used to take off and control the drone, which can be downloaded from this link.
- It is recommended to create a folder and always start the Docker container from it, as the command to run the container will mount the specific directory where the script is located. Docker containers typically sandbox themselves from the host system.
- If the PX4 simulator with baylands does not open, try running
cd px4-models && python3 simulation-gazebo --world baylandsandcd PX4-Autopilot && PX4_GZ_STANDALONE=1 make px4_sitl gz_x500_depthin separate terminals from the root folder of the workspace.
Ensure the following commands are added to your .bashrc file to set up the ROS environment:
source /opt/ros/humble/setup.bash
source /home/root/offboard_ws/install/setup.bashReload your .bashrc or source it manually:
source ~/.bashrcOpen a terminal and execute the following command to start the uXRCE and Gazebo bridge:
./start_uXRCE_and_gzbridge.shIn a new terminal, run the script to launch the PX4 autopilot and the Gazebo Baylands world:
./run_baylands_total-1.shIn another terminal, change the directory to the obstacle avoidance package:
cd final_project_ws/src/obstacle_avoidance/scripts/Before running the offboard control scripts, ensure QGroundControl is started and manually take off the drone.
Video demonstrating the tracker following a square trajectory:
To run the optimal trajectory tracker between a set of goal points, execute the following command:
python3 trajectory_tracker.pyVideo demonstrating obstacle avoidance between a set of goal points:
To run the obstacle avoidance script that maintains a 1-meter distance threshold from obstacles, use the command:
python3 obstacle_avoidance.py- Obstacle Position: For the purpose of simulation, the obstacle position is defined halfway between the drones current position and the goal position. Real-time obstacle detection and avoidance is executed in the Onboard Codes scripts
This code is designed to run on a ModalAI Voxl2-based drone, specifically tested on a Sentinel drone for this project. To ensure optimal performance, connect an Intel Realsense D435i camera to the drone. While a USB 2.0 connection was utilized in this project, it is highly recommended to use a USB 3.0 connection for the camera if available.
Install docker on the Sentinal Drone as given on ModalAI docs. Create a docker image with Ubuntu 20.04 and ROS Humble. The Intel RealSense SDK and respective ROS2 wrapper can be installed as explained in the GitHub repository for the RealSense ROS Wrapper. This also has to be installed within the docker container.
It is essential for the camera nodes, FMU nodes of the drone, and all other nodes to be on the same ROS Domain ID. This ensures that the Docker container can access all topics associated with these nodes, thereby simplifying communication with both the camera and drone nodes. To set up the ROS Domain ID Override and XRCE_DOMAIN_ID_OVERRIDE, follow these steps:
- Open a terminal and navigate to your home directory:
cd ~- Open the
.bashrcfile using a text editor:
nano .bashrc- Scroll to the bottom of the file and add the following lines:
export XRCE_DOMAIN_ID_OVERRIDE=9
export ROS_DOMAIN_ID=9- Save the changes and exit the text editor
The camera node has to be started before the obstacle avoidance nodes are started.
To start the camera node, execute the following command:
ros2 launch realsense2_camera rs_launch.py pointcloud.enable:=true pointcloud.ordered_pc:=trueThis command will enable the point cloud functionality and ensure that the point cloud data is ordered. Make sure to run this command before starting the obstacle avoidance scripts.
The ros_pointcloud_publisher publishes the filtered XYZ point cloud in a specific range. The obstacle is detected based on the average of n closest points to the drone.
To run the ros_pointcloud_publisher, execute the following command:
python3 "Onboard Codes/ros_pointcloud_publisher.py"The autonomous_obstacle_avoidance maintains a 1-meter distance threshold from detected obstacles. In the event that no obstacle is detected, the optimizer generates a straight line trajectory to the goal point.
To run the obstacle avoidance nodes, execute the following commands:
python3 "Onboard Codes/obstacle_avoidance_ws/autonomous_obstacle_avoidance.py"