Robots Play Minecraft to Learn to Make Logical Decisions

Friday, 26 June 2015 - 3:47PM
Weird Science
Robotics
Friday, 26 June 2015 - 3:47PM
Robots Play Minecraft to Learn to Make Logical Decisions
Robots are getting more and more complex and advanced every day, but there are still some seemingly simple functions of the infinitely complex human brain that robots struggle with. Decision-making, for example, is very difficult for robots, as they can conjure up so many possibilities that they have trouble narrowing it down. Now, researchers at Brown University are trying to help robots learn to make logical decisions- using Minecraft.


"We're at a really exciting time in robotics, because they're moving out of constrained environments like factories, and into our homes and hospitals, and they have the potential to profoundly impact people in every aspect of our daily lives," said Professor Stephanie Tellex of Brown's Humans to Robots lab.

Tellex explains that robots' algorithms have hundreds of thousands of possible sequences to respond to any one problem, and they have an enormous amount of difficulty sorting through the various factors and deciding which sequence will solve the problem. They found that playing video games, particularly Minecraft, helped robots learn to sift through many different options in order to choose the most expedient course of action. 

Opening quote
"We turned to Minecraft because it takes a really long time for a robot to try out different possibilities in the real world. It can only move so quickly, it takes a lot of computational power... Using a simulated domain like Minecraft allows our agent to very quickly try out different possibilities."
Closing quote


While the robots play Minecraft, they learn to "prune" their actions, or narrow down the many possibilities. In the study, the robots randomly generated examples of solutions to various problems encountered in Minecraft and chose between them to accomplish increasingly complex goals within the game. It then took the strategies it learned during the game and extrapolated to solve bigger problems than it had ever faced in training.

"The ultimate goal of our algorithm is to help the robot make the world less overwhelming, so it can narrow down the space of possibilities and quickly infer actions that meet people's needs," said Tellex.

Via Business Insider.
Science
Technology
Weird Science
Robotics

Load Comments