When faced with a task, deciding what the next action will be comes intuitively for a human. But for a robot, that can be an almost impossible task.
That's why researchers at Brown University's Humans to Robotics Lab set out to develop a set of algorithms to help robots better plan their actions in more realistic environments, such as hospitals or homes.
What they discovered was that computer games were great at helping robots learn important decision making skills. In particular, they chose Microsoft’s Minecraft, as they believe the game’s unlimited possibilities offer a great platform for such learning.
“A challenging problem the robot is confronted with is how to decide of the thousands, or hundreds of thousands of possible sequences of actions that it could carry out, which one it should actually do so that it helps the person?” says professor Stephanie Tellex of the Humans to Robotics Lab.
“We turned to Minecraft because it takes a really long time for a robot to try out different possibilities of the real world. It can only move so quickly, it takes a lot of computational power to perceive what’s out there in the environment,” she explains.
She also adds that the whole goal of the project is to help the robot learn what the promising directions are, depending on the problem the robot is faced with.
Minecraft allows the robot to very quickly try out different possibilities, and effectively that way learn what it can and should do, with the ultimate goal of helping humans.