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 transfer complex manipulation skills from simulations to the real world with high accuracy, a challenge known as the ‘sim-to-real’ gap. The method uses a technique called ‘residual physics learning,’ where the robot learns a base policy in simulation and then a secondary model to correct for the inevitable differences between the simulated and physical environments. In tests, a robotic hand successfully performed delicate tasks like rotating a wooden egg within its fingers—a skill requiring precise, continuous adjustment—after minimal real-world training. This approach reduces the need for extensive and costly real-world data collection, potentially accelerating the deployment of adaptable robots in unstructured settings like homes and warehouses. For the full details, read the complete article at https://technologyreview.com/2024/05/15/robotic-hand-simulation-real-world-transfer.