A new study from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) demonstrates a significant advancement in robotic manipulation. Researchers have developed a system that allows a robot to learn complex, dexterous manipulation tasks—such as spinning a long baton—using a relatively small amount of data collected from a single human demonstration. The key innovation is a framework that translates the visual data from the demonstration into a simplified, two-dimensional representation that the robot can more easily interpret and generalize from. This approach, which contrasts with methods requiring vast datasets or numerous physical trials, enabled the robot to successfully perform the baton-spinning trick after just 10 minutes of practice. The technique points toward more efficient ways to train robots for intricate, real-world tasks. Read the full article at https://news.mit.edu/2024/robotic-system-learns-dexterity-0724.