1. Introduction

1.1. Project Background

ENPH 353 provides an introduction to machine learning and computer vision. The course culminates in a competition in which teams design and implement algorithms for autonomous navigation of a parking lot while detecting and predict license plates along the way. Our approach to the problem allowed us to come 2nd in the competition in terms of plates detected, while completing the course in under 2 minutes.

1.2. Requirements

Before designing our solution, we began by defining our high-level requirements to ensure success during the competition, outlined below:

Detect license plates without affecting the drive capabilities of the robot.
Keep track of the plate ID for the plate detection module
Keep two wheels of the robot inside the track at all times
Avoid hitting or being hit by the truck in the inner loop of the course.
Do not hit pedestrians when driving across a pedestrian crossing.
Come to a safe stop after completing both the outer and inner loops.
Complete both the inner and outer loop in under 4 minutes.

2. Software & Algorithms

2.1. Software Architecture

To allow for responsiveness of the robot, we separated the required functionality into scripts that could communicate with each other using ROS. The individual modules that we created were as follows:

master

This module was an overarching state machine for the robot. On launch, this module would determine the required state for the robot, and communicate this to the right_lane_keeping module through the /drive_enb ROS topic.

pedes