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 robots to perform complex, deformable object manipulation—such as shaping dough or arranging cloth—by using a suite of specialized tools and a two-stage learning process. The system first uses a physics simulator to learn general strategies for tool use, then transfers this knowledge to a real robot, which refines its actions through trial and error. This approach overcomes the difficulty of training robots directly in the real world, where experiments are slow and prone to failure. The robots successfully completed tasks like using a rolling pin to flatten dough and a squeegee to gather scattered seeds, showing more precise and efficient performance than previous methods that attempted to use a single tool for all jobs. The research points toward more adaptable robots capable of handling varied, real-world tasks in settings like kitchens or factories. Read the full article at https://technologyreview.com/2024/07/12/1094755/robots-tools-dough-cloth-manipulation/