A new study from MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) demonstrates a significant advancement in AI-powered robotics. Researchers have developed a system that allows robots to learn complex manipulation tasks, such as assembling a chair or cooking a meal, from a relatively small number of human demonstrations. The system, which combines computer vision with a new machine learning framework, enables robots to generalize from the demonstrations to handle objects they haven’t seen before and adapt to minor variations in the environment. This approach addresses a key challenge in robotics by reducing the need for vast, task-specific datasets and extensive reprogramming. The technology has potential applications in manufacturing, home assistance, and other fields where adaptable automation is valuable. Read the full article at https://technologyreview.com/2024/03/14/robots-learn-tasks-from-few-demos.