Project Overview
Project Goal
We set out to retrofit an existing remote controlled vehicle into an autonomous outdoor navigation system using an Arduino, Raspberry Pi, smartphone running an AR app, and ROS2. Our hope was to design a system which enabled our vehicle to autonomously navigate forest trails using machine learning.
System OverviewLearning Goals
We chose this project as it had a large systems component which aligned well with our team's individual learning goals. We were also excited about working in a non-controlled environment (outdoors) and moving beyond the less-variable locations and Neatos that we had worked with in prior projects.
Team Learning GoalsPotential Concerns
Any autonomous vehicle has potential impact for harm - both in if it is poorly implemented and if it is used in a manner that has high potential for harm. To reduce this potential, we have documented intended use cases for our final result as it is open source and freely available to anyone looking to recreate it.
Ethics StatementProject Status
Currently, we have sucessfully integrated all of the components of our system,
and have run two field tests with our vehicle. This video shows a sucessful run
of the system - in which the robot navigates along a trail until it starts drifting,
and then corrects its path to better follow the trail. While the system worked in some
cases, there are still many improvements that we believe would be valuable to implement.
Our control model, which given the output weights of the machine learning model and determines
the tread velocities, could be improved to better guide the robot. We also could make
better use of our visual odometry to introduce more thorough path planning.