When given the correct tools, humans and robots alike can accomplish a wide variety of tasks. But what if the perfect tool isn’t available? Humans tend to be fairly adept at creating substitutes— a rock becomes a hammer, or maybe a bowl is pressed into use as a spoon, perhaps even with some sort of improvised handle — but robots tend to struggle with this type of out-of-the-box thinking.
This may be changing, however, thanks to new research out of Georgia Tech’s Robot Autonomy and Interactive Learning (RAIL) lab, with the goal of giving robots “MacCyvering” capabilities. This research, led by Professor Sonia Chernova, presents robots with a number of optional parts, and are told simply to make a specific tool. The robot then inspects the shapes of the parts that are available, and uses machine learning to construct the same type of device. It can even figuring out the best method to affix things together, whether through piercing the material, grasping it, or using magnetic attachment.
One example came when the robot was tasked with creating a screwdriver, reasoning that since pliers can grasp something, and a coin is similar to a screwdriver head, the two can be combined to form this basic tool. While it appears that tool designers’ jobs are safe for the time being, it will be interesting to see how this technology progresses in the coming years. For more information, the project’s research paper is available here.