Exploring Efficient and Easy Ways to Teach Robots New Skills

Robots haven't just invaded our workplaces, but it has penetrated our personal lives. The advancements of robots demand expanding their skills, moving up to show more productivity and accuracy in their performance. So, apart from inventing distinct robots, scientists and researchers are also looking for various efficient and easy ways to teach robots new skills. Recently, there has been some breakthrough in this research and scientists claim that they have found the easiest way to teach robots new skills.

Reprogramming the robot involves hand-labeling several types of images that enable the robot to understand new things. But MIT researchers have found a way that would require only a handful of human testaments to reprogram the robot. This technique uses a neural network specifically designed to reconstruct the shapes of 3D objects. With just a few demonstrations, the system uses what the neural network has learned about 3D geometry to grasp new objects that are similar to those in the demos.

According to these researchers, a robot can be trained to pick up a specific thing, but if the object is lying on its side, the robot will see this as a completely new scenario. This is one of the many reasons it is so hard for machine learning systems to understand the difference between new object orientations. So, to overcome this issue, researchers came up with a new technique involving a neural network model, a Neural Descriptor Field (NDF), that learns the 3D geometry of a class of items.

Researchers are quite happy with the robot's performance, and how their method works. It seems like these kinds of innovations will only rise in the future, advancing the mainstream adoption of robots in our lives.

More Trending Stories