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 perform complex, deformable object manipulation—specifically, rolling dough to match a target shape—using a dual neural network approach. The system combines a ‘diffusion policy’ for high-level planning with a ‘transformer’ model for fine-grained, real-time adjustments based on visual feedback. This two-stage method enables the robot to correct errors and adapt to the dough’s changing consistency during the task, a challenge that has long stumped rigid robotic systems. The research indicates progress toward robots capable of more nuanced, human-like tasks in unstructured environments like kitchens. Read the full article at https://technologyreview.com/2024/07/12/1094755/robot-dough-rolling-mit-csail/