A Chinese startup called Feagine Robotics claims to have created a system that, when installed in a robot, does more than just learn how to move a specific arm; it learns the task's objective and automatically adapts its movements to any compatible mechanical structure. To demonstrate this capability, the researchers used three completely different robotic arms.
Another interesting feature is that training may require only a single human demonstration; if a novel task arises, an operator simply needs to perform it. Once the system observes the sequence of movements, it identifies the ultimate goal and calculates how to replicate that result using any other compatible robot—even one with a completely different anatomy. This approach effectively decouples intelligence from hardware.
Posted Using INLEO
