Wesley Liao trained an AI to learn to play QWOP using reinforcement learning and imitation learning and it managed to achieve a top 10 speedrun.
QWOP is a simple running game where the player controls a ragdoll’s lower body joints with 4 buttons. The game is surprisingly difficult and shows the complexity of human locomotion. Using machine learning techniques, I was able to train an AI bot to run like a human and achieve a finish time of 1:08, a top 10 speedrun.
If you’ve ever played QWOP (you can play it here), you know how frustratingly difficult it is. Like creator Bennett Foddy’s other game, Getting Over It With Bennet Foddy, it was designed to punish the player into wanting to kill themselves so it’s only fitting somebody designed an AI to kill it first. It’s kind of like the plot to The Terminator if The Terminator was stupid and I knew how analogies work.
Keep going for the full explanation video or read the writeup here.