State based DAgger Agent for SuperTuxKart using DNNs

Summary: In this project we explore developing a state-based SuperTuxKart hockey player. We describe our approach, some results, and a discussion of what worked and didn’t work. We also provide our insights as to why this might be the case. In this project we design and implement a Deep Neural Network controller taking in state features of the game which output acceleration, steer, and brake. This is a first draft, and requires improvement on training process and network.

Repository: https://github.com/omeedcs/state-based-dagger-agent-with-dnn