MIT researchers develop robot arm able to pick up any object after inspection

Shane McGlaun - Sep 10, 2018, 8:02 am CDT
0
MIT researchers develop robot arm able to pick up any object after inspection

One of the biggest challenges for robots is something that humans can do so easily we don’t even see it as a challenge. When we walk through our home if we see something like a mug on the table we can pick it up by the handle no matter the orientation. Robotic hands find something that simple to be a major challenge. Human dexterity is credited in part to our eyes because we can see the object clearly making it easy to pick up.

Breakthroughs in computer vision are allowing robots to make basic distinctions between object types but those breakthroughs still don’t allow the robots to understand the shape of an object. That means there is little that a robot can do once the object is picked up. MIT’s Computer Science and Artificial Intelligence Laboratory (CSAIL) is changing that.

MIT has developed a system that allows the robot to inspect random objects and understand them enough visually to perform specific tasks with them without having seen the objects before. MIT calls its system the Dense Object Nets or DON. That system looks at objects ad collections of points that work as a visual roadmap. MIT’s tech allows robots to better understand and manipulate items and allow sit to pick up a single object amidst multiple objects.

The researchers say that this is an important skill for machines in use for picking items at companies like Amazon or Walmart. The system would allow the robot to do something like grabbing the tongue of a shoe successfully. Scientists say that other approaches to manipulation can’t specify parts of an object throughout the many orientations of the object that the robot might encounter.

DON uses coordinates on a given object to tell the robot where it needs to grasp. The system was trained to look at objects as a series of points that make up a larger coordinate system. After training, the system allows the robot to take a photo of an object and then pick the object up at a specific point. This technique could find use not only in industry but in homes as well. This could be the breakthrough we needed to get robot maids into our homes. MIT reckons the tech might one day allow you to show a robot an image of a clean home and have the robot clean while you are at work.

SOURCE: MIT


Must Read Bits & Bytes