What algorithms are available to test the performance of an AI agent in an automated way? Maybe creating a maze with obstacles to avoid, or rendering characters with slightly random ways to check whether it detects them. Any other method?