MIT engineers are making waves with their recent advancements in robotics, called Clio, which promises to simplify the way robots interact with their environment. This new approach, as reported by MIT News, enables robots to analyze through visual information and retain only what's necessary for the tasks at hand, all the while described in natural language.
In layman's terms, Clio is designed to smartly filter out the noise in a scene, allowing robots to quickly zero in on relevant items. For example, if instructed to specifically retrieve a green book from a table stacked with various items, Clio would enable the robot to identify and focus on the singular object, while disregarding the rest. MIT engineers believe Clio can be widely beneficial, from aiding search and rescue missions to improving domestic robots, according to Luca Carlone, MIT's associate professor of the Department of Aeronautics and Astronautics (AeroAstro).
The workings of Clio leverage a combination of cutting-edge computer vision, natural language processing, and classic information theory. By implementing the information bottleneck technique, Clio can compress a scene's image segments into clusters that hold task-relevant details, as explained by Dominic Maggio, a member of the MIT SPARK Lab. This methodology is a significant departure from more traditional fixed levels of granularity, which often result in inefficient and irrelevant mapping for robotic tasks.
What sets Clio apart is its ability to dynamically adjust to varying levels of detail needed for different tasks. For instance, in an experiment run at an MIT researcher's own apartment, robots were able to single out a pile of clothes from a cluttered scene to perform tasks like "move pile of clothes." "Running Clio in real-time was a big accomplishment for the team," Maggio told MIT News. He noted that similar technologies could take hours to yield results, while Clio operates with a notable efficiency.
The practical application of Clio was further showcased using the quadruped robot, Spot, from Boston Dynamics. As Spot moved about in an office building, it was able to create an overlay map pinpointing objects that were required for given tasks. This could potentially revolutionize future search and rescue operations by enabling robots to quickly identify and act upon key objects or individuals in need amidst chaotic environments. The research team, funded by several institutions including the U.S. National Science Foundation and the U.S. Army, envisions expanding Clio's capabilities to handle even higher-level, complex tasks moving forward.
