Researchers are developing a new algorithm to help robots efficiently plan their actions in complex environments.
The work aims to make these robots more useful in real life, but is being developed in the virtual world of Minecraft, Brown University reported. The study could lead to better action-planning skills for robots, overcoming the phenomenon in which there are so many choices it "boggles" the robot's intuition, dubbed the "state-space explosion."
"It's a really tough problem," said Stefanie Tellex, assistant professor of computer science at Brown. "We want robots that have capabilities to do all kinds of different things, but then the space of possible actions becomes enormous. We don't want to limit the robot's capabilities, so we have to find ways to shrink the search space."
This new algorithm uses "goal-based action priors," which are sets of actions in a given space that are most likely to help a robot achieve its goal. These priors can be applied by an operator, or learned through trial and error. The researchers tested the effectiveness of this algorithm in the game Minecraft.
"Minecraft is a really good a model of a lot of these robot problems," Tellex said. "There's a huge space of possible actions somebody playing this game can do, and it's really cheap and easy to collect a ton of training data. It's much harder to do that in the real world."
The researchers created small domains within the game and had a character solve a task within that environment using the algorithm. Some of these tasks included finding buried gold or building a bridge. They tested out different options in the algorithm to help the character achieve its goal based on priors.
"It's able to learn that if you're standing next to a trench and you're trying to walk across, you can place blocks in the trench. Otherwise don't place blocks," Tellex said. "If you're trying to mine some gold under some blocks, destroy the blocks. Otherwise don't destroy blocks."
After the character learned the appropriate priors to solve a task in a given domain, it was moved to a domain it had never seen before to see if it could apply the same strategies. They found that with the right priors, the character could solve problems in an unfamiliar domain more quickly than if standard planning algorithms had been applied.
Once the method was tested out in the virtual world, the researchers applied it to a real robot that was employed to help a human bake brownies. The priors told the robot eggs often needed to be beaten with a whisk. Once a carton of eggs was introduced to the environment, the robot identified the need for a whisk and handed it to the robot.
"I think it's going to provide a way for very rapid iteration for algorithms that we can then run in our robots and have some confidence they're going to work," Tellex said.
WATCH:
