🚁 Police Chase Drone - ROB498 Capstone Project

Overview

This repository contains the final implementation and documentation for our ROB498 Capstone Design Project as part of the Robotics Engineering undergraduate program at the University of Toronto.

Our project, Police Chase Drone, showcases the design and development of an autonomous drone system capable of pursuing a dynamic ground target. Leveraging ROS 2 and onboard computation, the drone tracks and follows a "getaway car" through a series of waypoints and behaviours designed to mimic a pursuit scenario.

Check out the project demo on YouTube!

Features

• 🔍 Real-time tracking of a ground vehicle using onboard sensors and vision using a custom MobileSSD-Net
• 🧠 Autonomous navigation with waypoint following and obstacle avoidance
• 📡 ROS 2-based modular software stack for sensor integration and control
• 🛸 Custom flight behaviors to simulate realistic police drone pursuits

System Overview

The software system is built in ROS 2 using a set of nodes that handle:

• Perception: Camera input, object detection, and tracking
• Planning: Waypoint generation and trajectory tracking using a state machine
• Control: Velocity and position control for the drone

Here's the state machine diagram that outlines the drone's behavior:

And here is it in action in the simulation:

Installation

1. Clone the repository:

git clone https://github.com/Alex-Alexiev/drone-dogs.git

2. Make dependancy folder:

mkdir -p drone_deps/src

3. Install jetson-inference:

cd ~/drone_deps/src
sudo apt-get install git cmake
git clone --recursive --depth=1 https://github.com/dusty-nv/jetson-inference && cd jetson-inference
mkdir build && cd build
cmake ../
make -j$(nproc)
sudo make install
sudo ldconfig

4. Install vision_msgs:

cd ~/drone_deps/src
git clone https://github.com/ros-perception/vision_msgs.git -b foxy
cd ~/drone_deps
colcon build --packages-select vision_msgs
export CMAKE_PREFIX_PATH=~/drone_deps/install:/opt/ros/foxy

5. Install ros_deep_learning:

cd ~/drone_deps/src
git clone https://github.com/dusty-nv/ros_deep_learning

cd ~/drone_deps/src/ros_deep_learning && mkdir build && cd build
cmake .. -DCMAKE_PREFIX_PATH=~/drone_deps/install:/opt/ros/foxy -DVPI_DIR=/usr/share/vpi1/cmake
make -j$(nproc)

cd ~/drone_deps
colcon build --packages-select ros_deep_learning

source install/setup.bash

Usage

From the drone-dogs repository:

Launch the Rviz2 'simulation' environment:

./test.sh

For hardware-in-the-loop testing:

./run.sh

Launch video source:

ros2 launch ros_deep_learning video_source.ros2.launch

Launch detectnet:

ros2 launch ros_deep_learning detectnet.ros2.launch output:=display://0 model_path:=/home/jetson/ros2_ws/src/px4_autonomy_modules/models/ssd-mobilenet-apr7.onnx input_blob:=input_0 output_cvg:=scores output_bbox:=boxes threshold:=0.2 class_labels_path:=/home/jetson/ros2_ws/src/px4_autonomy_modules/models/labels.txt image_height:=180 image_width:=320

Hardware Used

• NVIDIA Jetson Nano for onboard computation
• IMX219 Color Camera
• Orange Cube+ PX4-compatible flight controller
• Battery-powered quadcopter frame
• Wi-Fi telemetry and ground station

Team Members

This project was completed by:

• Kevin Angers
• Chris Adolphe
• Alexander Alexiev
• John Wolf

Supervised as part of the ROB498 Capstone Design Project at the University of Toronto.

Acknowledgments

We thank Professor Waslander and the teaching assistants staff for their guidance and support throughout this project, as well as the sample code found here. Special thanks to dusty-nv for the Jetson Inference and ROS Deep Learning repositories, which provided invaluable resources for our computer vision tasks.

License

This project is licensed under the MIT License.