Zach Nussbaum, Fabio von Schelling Goldman
Faculty Sponsor: Dr. Ramanujan
In this project we designed a program to generate an agent to play the Dinosaur Chrome game. To create this agent we used genetic programming to teach our agent to play the game and improve over multiple generations. The process of genetic programming combines fitter individuals to simulate evolution often resulting in an overall fitter individual, which is how the agent improves. Overall we found that already the random specimen performed decently and we were able to converge towards a bot who could successfully play the game. Furthermore we employed techniques of condensed state representation and similarity to maintain a low overhead of run time for larger populations.