A small "game" where you should have a maze with 2 players/agents..each one should find the exit using a different algorithm: one using A*(maybe this one tuned) and one with supervised learning. You can choose what technique you want to use for for the supervised learning and reinforcement learning(q-learning, neural networks and so on). You can also choose the programming language.
The project should be well documented,with all the design choices and decisions made explained.
A report to explain the algorithms/techniques and showing the results, also a comparison on the 3 algorithms based on the performance and how could be any of them improved.