This project is a submission for the Flipkart Grid 4.0 competition, which involved building an autonomous drone for indoor delivery. The drone is able to recognize and pick up packages using an electromagnet, and drop them off at a desired location. This README provides an overview of the project and instructions for using it.
The following hardware and software are required to run this project:
- Pixhawk F450 drone with PX4 firmware
- Jetson Nano
- Intel Realsense D435i camera
- Electromagnet
- ROS Noetic
- Mavros
- AprilTag ROS package
- Intel Realsense ROS package
Install ROS Noetic and Mavros on your computer. Install the Intel Realsense ROS package by following the instructions here. Clone this repository to your computer.
Connect the Intel Realsense camera to the Jetson Nano and the Jetson Nano to the Pixhawk F450 drone.
Launch the following ROS launch files:
rs_camera.launch: This launches the Intel Realsense camera and publishes the camera image to a ROS topic. px4.launch: This launches Mavros and connects to the Pixhawk F450 drone. continues_detection.launch: This launches the AprilTag ROS package for detecting the 4x4 grid of AprilTags on the floor.
Once you have installed and set up the necessary hardware and software, you can use the drone for autonomous delivery. Here are the steps to follow:
Power on the drone and wait for it to initialize.
Launch the rs_camera.launch, px4.launch, and continues_detection.launch files.
Terminal 2:-
cd fastplanner_ws/
source devel/setup.bash
rosrun FastPlannerOctomap Planner
Place the packages to be delivered within the view of the Intel Realsense camera. Make sure the packages are either marked with an Aruco code or have a distinct color on top.
After initializing the octomap planner, it maps the world or arena and updates it as the drone moves. After some instant, the simulation map might look like this:
Launch the fast planning with dynamic octomapping algorithm by running the following command:
roslaunch dynamic_fast_planner dynamic_fast_planner.launch
The drone will begin to plan a path to pick up the package using the shortest path possible. Once the package has been picked up, the drone will plan a path to the desired delivery location and drop the package using the electromagnet.
On running the simulation on gazebo, it should look similar to this at some instant:
This project demonstrates the use of ROS and Intel Realsense for building an autonomous drone for indoor delivery. The project includes obstacle avoidance and motion planning using dynamic octomapping, localization using AprilTags and the Intel Realsense camera, and package recognition using Aruco codes or color detection. We hope that this README provides a useful guide for others who are interested in building similar projects.