Autonomous Penalty Kick Goalie

MAE148 Final Project

Team 1 Fall 2024

Table of Contents

1. Overview
   • Key Features
2. Team Members
3. Project Goals
   • Core Objectives
   • Nice-to-Have Features
4. System Architecture
   • Node Descriptions
5. Technologies Used
6. How to Run
   • Prerequisites
   • Steps
7. Future Improvements
8. Final Project Documentation
9. Robot Design
   • CAD Parts
     • Custom Designed Parts
     • Open Source Parts
   • Electronic Hardware
   • Software
     • Embedded Systems
     • ROS2
     • DonkeyCar AI
10. Acknowledgments
11. Contact

Note: this document is intended as a high level overview of our project. For more technical/instructional documentation, see ball_vision_info.md

Overview

This project focuses on developing a robot car capable of acting as an autonomous penalty kick goalie. Leveraging computer vision (CV) and a DepthAI framework, the robot detects and tracks a soccer ball, calculates its spatial coordinates, and responds to intercept the ball. Using a YOLO object detection model, the system calculates steering and throttle control inputs, and publishes them to the vehicle's VESC (Variable Electronic Speed Controller) using ROS 2.

Key Features

• Ball Tracking: Detects and tracks the ball in real time using DepthAI and OpenCV algorithms.
• Depth and Angle Estimation: Computes the ball's depth (distance) and horizontal angle relative to the robot.
• ROS2 Framework: Publishes and subscribes to relevant data across custom ROS2 nodes for modular functionality.
• Proportional and Bang-Bang Control: Implements a proportional controller for accurate steering adjustments and bang-bang control for velocity inputs.
• Goalkeeper Behavior: Waits for ball movement before executing intercept maneuvers, mimicking real penalty-kick rules.

---

Team Members

Name	Major	Class
Abhi Sachdeva	Electrical Engineering	Class of 2025
Charles Lahey	Mechanical Engineering	Class of 2025
Evan Gibson	Mechanical Engineering, Ctrls & Robotics	Class of 2025
Gautam Ganesh	Mechanical Engineering	Class of 2025

---

Project Goals

Core Objectives

1. Ball Tracking and Control:

   • Implement a tracking node that uses DepthAI to:
     • Deploy a YOLO model directly to OAK-D Lite camera, detecting a soccer ball in frame.
     • Measure:
       • Ball Depth: Average distance between the ball and the robot in millimeters.
       • Ball Angle: Horizontal offset angle from the robot's center.
   • Publish data to VESC BLDC Motor Controller:
     • throttle variable via bang-bang controller.
     • angle variable, represent's angular offset between ball and car, normalized between -1 and 1 using camera's horizontal FOV.

2. Goalkeeper Rules:

   • Ensure the robot remains stationary until the ball begins moving (mimicking real penalty-kick rules).
   • React swiftly to block the ball once it starts moving.

Nice-to-Have Features

• Path Prediction:
  • Use ball position and velocity data to predict the trajectory and intercept the ball optimally.
  • Incorporate robot dynamics to determine the ideal intercept point.

---

System Architecture

The project leverages a modular architecture, where each node in the ROS2 framework is responsible for specific tasks.

Node Descriptions

1. yolo_node.py

   • Inputs: Camera feed from OAK-D Lite.
   • Measures:
     • Ball Depth: (distance in mm).
     • Ball Angle: (horizontal angle between ball and camera in radians or degrees).
   • Outputs:
     • Twist messages to the \cmd_vel topic
     • linear.x messages represent throttle commands
     • angular.z messages represent servo commands. These message are Ball Angle data normalized between -1 and 1.

2. vesc_twist_node.py

   • Inputs:
     • linear.x and angular.z messages from the yolo_node.py.
   • Outputs:
     • Commands to the VESC for motor control.

---

Technologies Used

• DepthAI: For object detection, depth estimation, and spatial tracking of the soccer ball.
• OpenCV: To process image frames and detect circular objects.
• ROS2: Middleware framework for data publishing and subscribing between nodes.
• VESC: Controls the robot's drivetrain, providing precise throttle and steering.

---

How to Run

Detailed instructions can be found in ball_vision_info.md

Prerequisites

• Install ROS2 (Foxy recommended).
• Set up the DepthAI SDK.
  • In Docker container projects directory

  git clone https://github.com/luxonis/depthai-python

• Ensure the VESC is configured and calibrated.

Steps

1. In Docker container, run source_ros2
2. Enter the src directory and clone the repository:

   cd src
   git clone https://github.com/UCSD-ECEMAE-148/fall-2024-final-project-team-1/tree/main
   cd ..

3. Build the ROS2 workspace:

   colcon build --packages-select ball_vision_package

4. Launch the system:

   ros2 launch ball_vision_package ball_tracking.launch.py

Future Improvements

• Integrate a path prediction algorithm for smarter ball interception.
• Enhance the PID controller for faster and smoother responses.
• Explore deep learning-based ball detection for improved accuracy in varying lighting conditions.

Final Project Documentation

• Final Project Presentation
• Project Proposal

Robot Design

CAD Parts

Custom Designed Parts

Part	CAD Model	Designer
Camera Mount		Evan
LiDAR Case		Gautam

Open Source Parts

Part	CAD Model	Source
Jetson Nano Case		Thingiverse
Oak-D Lite Case		Thingiverse

Electronic Hardware

Circuit Diagram of the electronic hardware setup for the car.

Software

Embedded Systems

Our team utilized a wirless SSH to a Jetson Nano that contained a docker container with all the necessary packages and dependecies used to run our program in a ROS2 workspace. SSH was done via both Mac terminal and Windows PC with Virtual Machine.

ROS2

The Docker Images, which were provided to us and pulled from the Docker Hub, contained the UCSD Robocar Module along with the ROS/ROS2 submodules that we utilized during project prototyping and lane following. The UCSD Robocar Module, running on Linux OS (Ubuntu 20.04), was initially developed by Dominic Nightingale, a graduate student at the University of California, San Diego

Acknowledgments

Thank you to Professor Jack Silberman and our incredible TA's Alexander, Winston, and Vivek for an amazing Fall 2024 class! Thank you also to @kiers-neely for the amazing README.md template.

Contact

• Abhi Sachdeva | [email protected]
• Evan Gibson | [email protected]
• Charlie Lahey | [email protected]
• Guatam Ganesh | [email protected]