What’s the hardest video game you’ve ever played? If it were not QWOP then let me know you right, know it you do not know how really a match can be difficult. The deceptively simple running game is so challenging to master that even a AI trained using machine learning still only sampled a top 10 score instead of crushing the record.
If you have never played QWOP before, you owe it to yourself try it and see if you can not even get your sprinter off the starting line. Developed by Bennett Foddy back in 2008, QWOP was inspired by an arcade game from the 80’s called Track & Field it requires players on crunching cleverly without winning a race. QWOP take a different approach and use players instead four keys to control the individual movements of a runner’s thighs and calves—A runner who behaves like a floppy rag doll and is subject toworld physics, including the effects of gravity. It may sound simple, but mastering the timing and cadence of the key presses needed to get the sprinter moving just awkwardly forward can be incredibly frustrating.
Wesley Liao was curious how good an instrument like AI is trained to do things like realistic animation old photos of deceased loved ones, would play QWOP. After Liao first created a Javascript adapter that enables an AI tool to play the game and communicate with it, the KIA made its first attempt at machine learning simply to play the game on its own and learn which actions resulted in positive outcomes (the sprinter moves forward. and its velocity increases) and what results resulted in negative results (the sprint of the sprinter) too close to the ground.) Through this approach, the AI has learned a “knee-scratching” technique to successfully do it over the 100-meter finish line, but not at record speed.
Liao’s next attempt to train an AI model has taken up the recording of game videos in which they try to succeed, including the use of longer legs, which is important to increase speed and with a decent amount of time over the finish line. to come. The approach was slightly more successful, but the AI could not master a special technique used by advanced QWOP players involving an upward, forward swing of the legs to generate extra momentum.
G / O Media can get a commission
Eventually, Liao reaches out to a veteran player known as Kurodo (@cld_el on Twitter), one of the best QWOP speedsters in the world, who recorded 50 videos of themselves playing the game at expert level. But even with access to the best possible playing techniques, Liao found that the best results came from a machine learning training program that played 25 hours of AI on its own, learned 15 hours from the data Kurodo’s expert runs gained , and another 25 hours of self-play.
But even with all the effort, the QWOPplay AI’s best 100-meter point due to the finish line in 1 minute and 8 seconds –a top 10 finish. According to Speedrun.com, is the current world record of 100 meters just 48 seconds set just a month ago. Liao is confident with more training and a different reward system (how the AI learns that something was done correctly), setting up a QWOP world record can finally happen, even though it is a computer playing the game, the record can never be officially recognized.