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 enables a robot to perform complex, dexterous tasks—like hanging a framed picture on a wall—using only visual data from a single camera mounted on its ‘wrist.’ This approach, which does not rely on depth sensors or multiple camera angles, allows the robot to perceive its environment in 3D and adjust its grip and movements with high precision. The system uses a neural network trained on a dataset of simulated interactions to understand spatial relationships and object geometry from 2D images. This breakthrough could make robots more adaptable and cost-effective for real-world applications in homes, warehouses, and other unstructured environments where precise manipulation is required. Read the full article at https://technologyreview.com/2024/07/18/109…