The problem with this trial is that the simulated environment provided the AI algorithms with "perfect" data that's unobtainable in the real world. Certainly, it's a benchmark in the progress of AI but the real world is messy and unpredictable. Just ask the designers of self-driving cars. They work really well when roads are clear and well-marked, but they really struggle when the real world gets in the way (e.g. snow that obscures road markings).