Autonomous Delivery Robot
Team
- Sam Petersen SamuelCPetersen@gmail.com
- Nathan Haws NathanRHaws@gmail.com
Project
We designed and assembled a self-driving delivery vehicle.
- People and companies spend a lot of time delivering food, packages, and other items.
- A self-driving delivery vehicle could save these people time and money.
- We set out to design and assemble an autonomous vehicle that could safely and securely navigate around its environment to deliver a package to a user-defined destination.
System
Hardware
Our processor takes inputs from phones, a GPS sensor, a proximity sensor, and a camera to navigate around its environment. The processor drives DC motors and a lock on the box.
Software
Our vehicle navigates based on inputs from the GPS and TensorFlow model. The TensorFlow model scans images from a forward-facing camera for paths and branches in these paths.
Methods
- We assembled the frame of the vehicle out of aluminum angles for their price and strength.
- We chose a Raspberry Pi to drive the vehicle since it could run Tensorflow Lite, a deep learning image processing program.
- The vehicle takes images from a forward-facing camera and scans them for the sidewalk path in front of the vehicle and for branches from the path.
- We drive two geared DC motors through a PID controller. We adjust our target speed whenever our proximity sensor detects an obstacle in front of the vehicle.
- We built an android app for our project’s UI. The app sends signals to a web server hosted by the Raspberry Pi.
Conclusion
The vehicle and its peripherals all function as intended. However, we have not yet succeeded getting the vehicle to navigate to a location. This is due to difficulties and inconsistencies in our turning and branch recognition algorithms. However, we did accomplish the following:
- Users can communicate with the vehicle through a custom android app
- The vehicle knows its location at all times through a GPS sensor
- The vehicle recognizes and avoids obstacles
- The vehicle recognizes feasible paths through a custom Tensorflow Lite model
- The vehicle met our price goal, costing less than $300
Once we improve our navigation algorithms, we will have a safe and functional self-driving vehicle that can deliver packages for people.