A navigation system powered by machine learning is training robots to recognise objects

The robots can now focus on learning the relationships between objects and room layouts. It also enables the robot to navigate its way from point A to point B, in the quickest possible manner.

A robot being trained using SemExp.

(Subscribe to our Today's Cache newsletter for a quick snapshot of top 5 tech stories. Click hereto subscribe for free.)

Carnegie Mellon University (CMU) and Facebook AI Research (FAIR) have developed a semantic navigation system – SemExp, to train robots to recognise objects, using machine learning.

Through SemExp, a robot is trained to differentiate between a kitchen table and an end table, while it is also able to understand where these objects are likely to be found. The process allows the navigation system to think strategically about how to search for something, said Devendra S. Chaplot, a Ph.D. student in CMU's Machine Learning Department, in a release.

Classical robotic navigation systems, explore a space by building a map showing obstacles. The robot eventually gets to where it needs to go, but the route can be circuitous. The system uses its semantic insights to determine the best places to look for a specific object, Chaplot added.

By making the system modular, the overall efficiency has gone up. The robots can now focus on learning the relationships between objects and room layouts. It also enables the robot to navigate its way from point A to point B, in the quickest possible manner.

Going forward a navigation technology like this could improve the interactions between people and robots. While a robot could bring an item in a particular place or it could find its way when directed, said a CMU release.