Teaching AI to Play Tennis

Overview

Artificial intelligence will be the next great leap for mankind, and it will influence all of our lives. My objective was to create an artificial intelligence neural network that is capable of learning to play the sport Tennis through deep reinforcement learning. First a tennis game environment in Unity was created. Using an artificial intelligence neural network, agents would train against a hard coded program, as well as against themselves through Self-Play. The results of training were recorded, what went well, and what could be improved upon. After implementing the improvements, the process was repeated. Some hyperparameters would purposefully be changed to see how the agents reacted and how the overall end product would be affected. In the end, the neural networks that performed the best were the ones that trained the longest. An interesting result that was observed was that you could bias the agent to do certain things more often by giving it different rewards for doing certain actions. For example, you could bias the agent to always hit down the line if you gave it a reward every time it did so. The agents could play against each other reasonably well, with a few mistakes here and there. It would be interesting to see this idea implemented into virtual reality so that athletes could practice their technique against different types of agents, like one that never misses, or one that always hits flat, aggressive shots.

Awards (1)

• Category Award: HM

Competition history

• CSEF 2023 Mathematical Sciences · Entry J1408

Resources

• CSEF published award list ↗
• CSEF published award list ↗
• CSEF published award list ↗
• CSEF public showcase (Internet Archive capture) ↗

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Mathematics projects
• CSEF Mathematics projects
• Every CSEF 2023 project