Robots need AI and ‘common sense’ to navigate [video]

Researchers at the Carnegie Mellon University demonstrated how robots equipped with “common sense” will be able to navigate more easily, thanks to understanding their environment and recognising the context of objects.

Common sense for robots: A new way to navigate

The CMU research team partnered with Facebook AI Research (RAIR) to design a semantic navigation system, or SemExp system, that utilises artificial intelligence (AI) and machine learning to train robots.

To be clear, this is more complex simple superficial traits and involves more than simply teaching a robot the difference between a chair and a plate.

Instead, researchers want robots to distinguish the difference between, say, an end table and kitchen table to the extent that it will be able to extrapolate in which rooms the different tables are located.

Understanding the context of objects

A fridge is pretty stock standard in this scenario as it is usually restricting to a single room, ie, the kitchen. But by adding “common sense” to the equation, the robot will be able to understand the context of a kitchen table when compared to an end table.

The robot needs to extrapolate data to determine where it needs to go. Machine Learning PhD student Devendra S. Chaplot explains:

“Common sense says that if you’re looking for a refrigerator, you’d better go to the kitchen. Classical robotic navigation systems, by contrast, 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.”

Watch: Common sense guides robots

In essence, a robot travelling from point A to point B will get there faster if it understands that point A is an end table in the lounge and point B is a kitchen table.

Thus, the basis of the semantic navigation system is to ensure the robot knows where it needs to go, even if it is in an unfamiliar place, in the same way a person would apply common sense to navigate around a home.

The semantic navigation system uses machine learning to enable the robot to “think” strategically about how to search for something”, according to Chaplot.


According to the research team, previous attempts to utilise machine learning to improve a navigation system, failed because robots would memorise objects and their locations, without understanding the context.

Chaplot and the research team found a way around that problem by equipping the SemExp with a modular system with semantic insights. This allowed the robot to determine the best of course of action for finding a specific object.

“Once you decide where to go, you ca just use classical planning to get you there”, Chaplot said. The modular approach proved to be efficient

In the long run, semantic navigation will also make it easier for humans to interact with robots. We’d be able to tell a robot to fetch a mug from the kitchen table or direct it to a specific door.”

Also read — Watch: Here’s how Boston Dynamics’ robo-dog encourages social distancing Protection Status

Source link

Leave a Reply

Your email address will not be published. Required fields are marked *