Robots learn to solve problems faster by playing Minecraft
Minecraft is great for teaching young gamers key skills like problem solving and creativity, but it's not just humans who can learn a something from it. Researchers have used the computer game to teach robots these same skills in a faster, more efficient way.
Usually, robots require careful programming to perform specific tasks in a fixed working environment. But one of the main problems people have been trying to address is around how they can be taught to do more open-ended tasks in less structured settings. Stefanie Tellex, a professor at Brown University, has turned to Minecraft to explore this.
The algorithm created by researchers was tested on a robot programmed to help with cooking. This is a difficult nut to crack, as a scenario like this, alongside hypothetical examples like a housekeeping robot, there's plenty of external factors and new skills to learn and pick up along the way, requiring ridiculous amounts of computational power.
“You might tell a robot ‘Make me coffee,’ but the next minute you might say ‘Do the laundry,’” Stefanie comments. “In this context, where you don’t know the goal in advance, there’s this planning problem. Finding the sequence of actions that’s going to work in this particular environment is very challenging. Our approach is about learning that faster.”
The algorithm mentioned above, designed by the whole team of colleagues at Brown, enables a robot to look at all possible paths of actions and variations, then decide the best course for it all. With this in place, the robot can learn to understand that key things like washing clothes doesn't require kitchen utensils.
To test this, they wheeled out Minecraft. The researchers controlled the character as it put a gold block into a furnace without touching the lava. The algorithm began to learn the specifics of this through trial and error. When the same task was introduced in a more complex setting, the character went through a much smaller set of scenarios to get the job done based upon the past experience.
Fascinating research with a bright-yet-terrifying future of self-aware robots
