Researchers at the Georgia Institute of Technology have developed robots that have the ability to reason about shape, function, and attachment of unrelated parts. This then lets them build basic tools by combining objects. The breakthrough is a significant step toward enabling intelligent agents to devise tools that could prove useful in hazardous situations.
In this latest work, a robot trained using the team’s novel approach is given a set of optional parts and told to make a specific tool. Much like a human, the robot first examines the shapes of each part and how one might be attached to another. Using machine learning, the robot is trained to match form to function from numerous examples of everyday objects. For example, by learning that a bowl’s concavity lets it hold liquids, it can uses this knowledge to construct a spoon. Similarly, the robot was taught how to attach objects together from examples of materials that could be pierced or grasped. In the study, robots successfully created hammers, spatulas, scoops, squeegees, and screwdrivers.
“The screwdriver was particularly interesting because the robot combined pliers and a coin,” says Lakshmi Nair, a Ph.D. computing student. “It reasoned that pliers could grasp something and that the coin matched the head of a screwdriver. Put them together, and it created an effective tool.”
Currently, the robot is limited to manipulating shapes and attachments. It cannot yet reason about particular material properties, a crucial step in advancing to real-world scenarios.
“People reason that hammers are sturdy and strong, so they wouldn’t make a hammer out of foam blocks,” Nair says. “We want to the robot to reach that level of reasoning, something we’re working on now.”