Skynet Is Coming
Were now one step closer to Skynet.
MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) have developed a way that enables robots to learn new skills and teach those skills to other robots.
The system is called C-Learn and allows non-coders to impart some information on how to conduct and demonstrate a simple activity to a robot. Some of these activities include opening doors or moving objects around.
But the amazing innovation is that the robot’s skills are transferable and possess the capacity to teach what they’ve learned to other robots.
“By combining the intuitiveness of learning from demonstration with the precision of motion-planning algorithms, this approach can help robots do new types of tasks that they haven’t been able to learn before, like multi-step assembly using both of their arms,” says Claudia Pérez-D’Arpino,who wrote a paper on C-LEARN with MIT Professor Julie Shah.
The MIT researchers tested the idea on Optimus, their two armed robot who was programmed to open doors, transport objects and move stuff into different containers and discovered his skills were transferable to Atlas, their six foot, four hundred pound other robot.
How Does It Work?
C-Learn bestows information on how to reach and move various objects with different constraints. (The C in C-LEARN stands for “constraints.”) Even though objects such as a steering wheel or tire, different motions are required in order to when putting those parts on a car. The information from C-Learn enables the robot to know how to do it.
Using a 3D interface, the team showed the robot a single demonstration of the task, represented by a sequence of moments known as “keyframes.” Matching these keyframes to the various situations, the robot can automatically suggest motion plans for the operator to approve or edit as needed.
Existing constraints could be learned from demonstrations that weren’t accurate enough to enable robots to perform the activity precisely. The researchers developed constraints to reconcile with that, computer-aided design (CAD) programs can tell the robot if its hands should be parallel or perpendicular to the objects its manipulating.
The robot performed even better when it worked with people, successfully completing 87.5 percent of the time sole, it did so 100 percent of the time when it had an operator that could correct minor errors relating to the robot’s occasional inaccurate sensor measurements.
“Having a knowledge base is fairly common, but what’s not common is integrating it with learning from demonstration,” says Dmitry Berenson, an assistant professor at the University of Michigan’s Electrical Engineering and Computer Science Department. “Dealing with the same objects over and over again, you don’t want to then have to start from scratch to teach the robot every new task.”