As a Freshman at Wachusett Regional High Shool in Holden MA, I became interested in trying to build a car that would drive itself (using computer control) through a maze. I've worked on building this (and changed the goals) for all my high school years. The car uses a technique in artificial intelligence called neural networking to learn to steer around any course. So, no program was written or used that tells the car when to turn which way. Instead, programs were written to help the computer learn to do steering choices that are "like" the ones it was trained on.The car communicates by radio with a computer that runs my programs. I enterred my work in 3 years of science fairs. It has been very interesting talking to people about this work and getting fresh ideas. I've also won $1,700 and a first place in the State of Massachusetts award. (At the State Fair, the top 15 or so projects out of the hundreds that are there get first place awards. You get to the state fair by winning locally and regionally.) The biggest thrill though, was the first time it actually all worked, driving around a course laid out in the cafeteria of Dawson Elementary School (where my Mom taught) on a cold winter night in 1999.
Check references listed on the menu for documents and sites that were helpful. The biggest inspiration from this project was the ALVINN project at Carnegie Mellon where a van is driven by a neural network. Click here for some discussion that project and the parts of it that I used in my work.
My high school hosts a Science Seminar program which I participated in for these 3 years. The seminar meets weekly. For each seminar, a speaker talks about his/her research and for the rest of the session we met with an advisor (a local scientist volunteer) to talk about our project. My advisor was Dr. Strock. He was very helpful to me!
This site has some graphics and videos as well as the programs (written in Microsoft C++) used in the project. You may need to update your Quicktime Player if its earlier than Quicktime 3. Click the link on the left to go to Apple's site.
Let me know what you think (or if you have questions) at email@example.com. Click on thumbnail pictures to see enlarged views. The movies are quite large files Watch out!
The car is steered remotely by a computer that radios steering and throttle commands. The computer gets a video image by radio from a camera mounted on the car. It feeds the image as digital data into a neural network (a kind of computer program and data structure) which chooses the next steering command. The car "learns" to steer by a technique which trains the network. This training gives the network "intelligence."
Over the 3 years, I've changed the design of the problem, the neural network, the programming language, the interfaces for both input and output radio signals etc. The car began tgethered by wire to the computer and could only move about 3 feet before the wire was fully extended. After 3 years, it was completely untethered and can handle most any closed course that fits certain restrictions. It has gone around a small gym course some 5 loops steering itself. It is great to watch! It is a bit of a problem to show it in the little space of a science fair. Click here to see what I did about that.