Grasping New Advances in Robotics Programming

Grasping New Advances in Robotics Programming

Though industrial robotics is advanced enough to keep the robots on assembly lines picking up the same objects over and over, humans still remain the rulers of dexterity. I came to a recent understanding that is is just now starting to change with some new programming and analysis in robotics.

Computer Vision

I learned that there have been recent breakthroughs in computer vision and this is allowing robots to distinguish between various different objects. Still, computers may not comprehend the shapes of objects so they cannot do much more than simply pick up an object.

According to researchers from MIT’s Computer Science and Artificial Intelligence Laboratory, there are key developments being made in this area. There is now a system that allows robots to inspect various random objects to the point of understanding them visually and understand enough to accomplish tasks, all without having ever seen them previously.

robotics

Self-Programming

I found out this new system has been called “Dense Object Nets” (DON) and it perceives objects as points collected as a sort of visual road map. This allows robots to learn and understand objects to the point that they can distinguish specific objects even among clutter.

I now understand that this type of self-programming development can be useful in a variety of industries. As an example, DON can be used to teach a robot to grasp a particular spot on an object such as the handle of a cup. After that point, the robotics will be able to grab the handle on any cup since it is recognized as a similar object.

Former Application Limitations

As I did some more research, I learned that most of the approaches to robotic manipulation are not able to identify specific parts of objects. As a result, it would be impossible for existing algorithms to allow a robot to grab a cup by the handle, in particular if it is laying in different positions.

Upon learning about DON, I came to the conclusion that this will allow for the systems to learn many new applications that could be useful in industries and in homes as well. It seems to be it would be possible to teach robots what a clean home looks like and then it could tidy up while people are away.

Self-Supervised System

With this type of programming, robotic systems can literally learn how objects are oriented and potentially learn how to become dexterous with the right hardware. This seems like a futuristic advancement to me, something straight out of science fiction now becoming real.

According to what I learned, researchers say that these systems will be self-supervised and they will not require additional annotations from human input. That being stated, we are looking at some amazing advances in robotics that could help on many different levels.

Since the DON system allows robotics to create a visual road map of objects, these robots will be able to learn how to manipulate various objects that it is familiar with. I understand this is well beyond the current limitations of robotic technology.