Introduction
I’ve always been fascinated by autonomous vehicles, from watching DARPA Grand Challenge videos on the internet to reading articles on their workings. I took an AI class at Air University in the sixth semester, which exposed me to Reinforcement Learning and piqued my attention. I had to submit a proposal for my Final-Year Project that same semester. Despite the fact that I was leaning toward building a webapp in Django and had half-completed my proposal, I decided to switch gears and combine my enthusiasm for AVs with my newfound interest in Reinforcement Learning, resulting in this project. My grade suffered that semester, however, for Final-Year Project II and III, my team scored the highest grade and got selected for presentation at Open House 2019 infront of the Vice-Chancellor and various companies.
Highlights
- Led a group of three in assessing and evaluating performance of Deep Q-Network and Double Deep Q-Network in OpenAI gym’s CarRacing-v0 environment.
- Responsible for conducting research on relevant subject matter, setting timelines, and handing out reading material.
- Implemented using Python and TensorFlow, measured and observed performance using average Q-values and average reward values.
- Wrote a report detailing background, findings, performance, implementation details and presented findings to a panel of professors.
- Recognized and selected among several projects for presentation at Air University Open House 2019.
Acknowledgements
Special thanks to our supervisor Sir Dr. Shafi for guidance, Sir Dr. Kamal for support, and my group partners Ramin and Junaid.