Researchers from the Massachusetts Institute of Technology (MIT) have developed a new way to program robots to do certain tasks, endowing them with skills that can be automatically transferred to other robots with different configurations.
C-LEARN, where C stands for “constraints,” is a system created by Claudia Perez-D’Arpino and Julie Shah, doctoral candidates within MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL). It combines two different methods of programming robots to do simple tasks, resulting in an approach that is quite similar to human learning processes. The first method involves learning from demonstration, in which robots watch a task being done and then try to replicate it, while the second one relies on motion-planning techniques, which require a programmer to specify a task’s goals and constraints.
For example, in order for the robot to complete a task involving a certain object, the user would provide it with a knowledge base of information on how to reach and grasp objects that have different constraints. Then, the robot is shown a single demonstration of the specific task using a 3-D interface. By matching the “keyframes” from the demo to different situations in the knowledge base, the robot can automatically suggest motion plans for the operator to approve or edit as needed.
One of the biggest advantages of C-LEARN is that it does not rely on directly imitating motion, but on trying to infer the principles behind it so that the robot learns more than one way to perform a certain action. This flexibility could prove crucial in teaching robots to perform complex multiarm and multistep tasks necessary for assembly manufacturing and ship or aircraft maintenance, but also for bomb disposal and disaster response.