Machine Learns How to Play Pitfall On Its Own
Humanity has taken another step toward making Skynet a reality, as researchers at Rutgers University have managed to design a robot that can play videogames - Pitfall, namely - on its own, and is even human enough to celebrate clearing a tricky level.
I've never beaten a single screen of Pitfall, which is a shameful fact in and of itself, but maybe even moreso now that I know that even a brainless, heartless, and, dammit, handless machine can do it.
Researchers at Rutgers University in New Jersey (which sucks, by the way, yeah, you heard me Funk) have managed to design an algorithm in machine learning that utilizes something called Object-Oriented Markov Decisions [http://en.wikipedia.org/wiki/Markov_decision_process#Reinforcement_Learning]. Nevermind the esoteric terminology - the researchers demonstrated their findings by having a computer beat the first screen of Pitfall.
As you can tell from the video, this isn't just a case of a machine being programmed to do the right things to get past the obstacles in the game. No, on its first run it tries to go down a ladder, finds that it doesn't work, and then realizes it needs to jump over a pit and a log in order to get across. It makes mistakes, does trial and error and then wins. To celebrate its victory, it seems to do a "happy dance" by jumping up and down at the end of the video.
In college I once had to write a program in Java that would teach itself how to find its way through a maze using trial and error of sorts, but this thing doesn't just know how to beat a videogame, it knows how to feel victory. God help us all. Read the Rutgers study here [http://paul.rutgers.edu/~cdiuk/papers/OORL.pdf].
[Via GameSetWatch [http://www.gamesetwatch.com/2009/06/machine_plays_pitfall_dances.php]]
Permalink
Humanity has taken another step toward making Skynet a reality, as researchers at Rutgers University have managed to design a robot that can play videogames - Pitfall, namely - on its own, and is even human enough to celebrate clearing a tricky level.
I've never beaten a single screen of Pitfall, which is a shameful fact in and of itself, but maybe even moreso now that I know that even a brainless, heartless, and, dammit, handless machine can do it.
Researchers at Rutgers University in New Jersey (which sucks, by the way, yeah, you heard me Funk) have managed to design an algorithm in machine learning that utilizes something called Object-Oriented Markov Decisions [http://en.wikipedia.org/wiki/Markov_decision_process#Reinforcement_Learning]. Nevermind the esoteric terminology - the researchers demonstrated their findings by having a computer beat the first screen of Pitfall.
As you can tell from the video, this isn't just a case of a machine being programmed to do the right things to get past the obstacles in the game. No, on its first run it tries to go down a ladder, finds that it doesn't work, and then realizes it needs to jump over a pit and a log in order to get across. It makes mistakes, does trial and error and then wins. To celebrate its victory, it seems to do a "happy dance" by jumping up and down at the end of the video.
In college I once had to write a program in Java that would teach itself how to find its way through a maze using trial and error of sorts, but this thing doesn't just know how to beat a videogame, it knows how to feel victory. God help us all. Read the Rutgers study here [http://paul.rutgers.edu/~cdiuk/papers/OORL.pdf].
[Via GameSetWatch [http://www.gamesetwatch.com/2009/06/machine_plays_pitfall_dances.php]]
Permalink
