BikeBot: A Study in Reinforcement Learning
April Kutz, Kristen A. Brennan, Jeremy Dixon, Nathan Bogdonas · Huskie Commons (Northern Illinois University) · 2016
In the United States, there is a lack of young adults interested in studying Science, Technology, Engineering, and Math (STEM). One way to help spark interest in STEM is a classroom aid that demonstrates how STEM can be fun and how it is used to push the limits of modern technology. BikeBot was designed for just this purpose. BikeBot is a robotic bicycle that utilizes an algorithm called Q-learning. Q-Learning is a type of reinforcement machine learning program, a basic form of artificial intelligence, that will allow BikeBot to balance without any explicit controls. The algorithm was tested and proven to work through a computer simulation of BikeBot’s dynamics. Our results show that BikeBot learned to balance upright while moving forward at a constant velocity in 50,000 iterations or 16 minutes and 40 seconds of real time learning.