The recently released annual list of the top innovators under the age of 35 by the MIT Technology Review showcases some remarkably accomplished individuals. One notable inclusion is Lerrel Pinto, an associate professor of computer science at NY University, whose work revolves around teaching robots how to perform tasks in a household by learning from failure.
To accomplish this, Pinto created a simulated home environment in his lab, complete with miniature versions of appliances like ovens. He then tasked a robot with performing various household chores, such as warming his lunch. Pinto equipped the robot with a machine learning model derived from 30-second videos of his students successfully completing the same tasks using similar tools. The robot was left to independently attempt the tasks around the clock, and Pinto evaluated its performance.
This approach is known as reinforcement learning, where the machine learning system tries different approaches, assesses its proximity to the desired outcome, and adjusts its strategy accordingly. Pinto documented his findings in a paper titled “Teach a Robot to FISH,” a play on the adage about giving someone a fish versus teaching them how to fish. The paper offers detailed insights into the breakdown and training process of the tasks. Pinto also delivered a compelling presentation on the subject at the CMU Robotics Initiative.
The work of Lerrel Pinto and others on the MIT Technology Review’s list highlights the potential of artificial intelligence in the domestic sphere. By teaching robots how to adapt and learn from their mistakes, researchers like Pinto are paving the way for more capable and useful household companions. As further advancements and applications are explored, the integration of robots into our daily lives may become more seamless and efficient.
– MIT Technology Review
– Lerrel Pinto’s paper “Teach a Robot to FISH”