For most of us, it is relatively easy to follow a how-to diagram from IKEA or for a set of LEGOs. Common sense helps a lot when trying to get from point A to point B. Robots, unfortunately, have to be programmed extensively for them to move a certain way, let alone accomplish complicated tasks that require following a certain number of steps. Even when a robot is taught a particular set of motions it just copies them without understanding why the moves were performed in the first place.
This is precisely what researcher Dileep George and his team of artificial intelligence and neuroscience researchers are trying to change. To do this, they have developed an operating system that can understand basic ideas of schematic instructions and then translate them into real-world actions. These robots are eventually going to replace the current line of robots just by using their common sense. The robots that are being used in various industries are following a set of strictly-coded guidelines, while these new robots can apply newly learned schemes and actions on the fly with little assistance needed from humans.
So far, the system has mastered around 500 general concepts. These concepts range from stacking colored objects from left to right, all the way to arranging the objects in a circle. They do this by looking at before and after pictures for every action that they need to perform. When faced with a new schematic or set of instructions, the robot studies them and then tries to apply previously learned concepts which best fit the current situation.
The team of scientists applied the new OS to a pair of gripper arm robots which were then able to perform basic tasks like separating lemons from limes and grouping colored objects in a row. Even when some of the conditions were changed, the robots were still able to perform their tasks without any issues.