When humans pick an object, they don’t need to think about how to orient it in their hands. However, that’s not the case with robots. But that might change soon, as a team of scientists from MIT’s Computer Science and Artificial Intelligence Laboratory has created an AI system that could give robots the same kind of intelligence.
Using a model-free reinforcement AI algorithm, they created a simulated, anthropomorphic hand that could handle over 2,000 objects. Moreover, the system didn’t need to know what it was about to pick up.
Transferring the ability to robots
The system isn’t fully ready for use. To start, the team needs to transfer this ability to an actual robot. That might not be as much of a problem as it sounds. Previously, researchers from Zhejiang University and the University of Edinburgh managed to transfer an AI reinforcement approach to their robot dog. The system lets the robot learn how to walk and recover from falls without any external help.
As of now, the system isn’t perfect either. It can reorient many smaller objects such as tennis balls and apples, with a nearly 100 percent success rate, but tools like screwdrivers and scissors might pose a challenge.
Accuracy drops significantly in some cases
In the case of such objects, the success rate drops down to 30 percent. Still, the potential of the system is huge. The system could one day make robots better at manipulating tools and make them more efficient at tasks such as packaging boxes.
Speaking of robots taking over the world, there’s a new breed of killer robots in the offing. Military robot maker Ghost Robotics has showcased a contraption that has a sniper gun attached to it. The robot is designed to kill with precision on the battlefield. The image was filled with comments; while some thought it was cool, some think it’s a nightmare turned into reality.
