Can Robots Learn Like Human Babies? Researchers Trying To Make It Happen
The debate over whether robots will eventually replace humans in day-to-day tasks is a never ending one. With several advancements in technology, researchers are now attempting to gauge whether bots have the same learning capabilities as human beings.
Researchers from Carnegie Mellon University are trying to demystify this conundrum and are taking a new approach with bots. They are allowing the robots to interact and explore with regular physical objects that form a part of one's daily life, in a bid to help them learn and adapt — just as babies would.
The research team led by Abhinav Gupta, assistant professor at the university, is attempting to enhance the learning abilities of robots.
Despite several developments taking place in robotics, they still find it difficult to climb up a flight of stairs, open doors, or use a fork. Bots are still not able to interact like humans and seem to struggle while performing the basic tasks.
Gupta and his team, however, are trying to fix this challenging problem.
According to the research methodology, the robots will be allowed to play with physical objects of everyday use. This will be done with the sole motive that the bots shall be able to explore the world. Also they can learn a lot from these interactions just as a human baby would try by absorbing from the immediate environment.
While robots are technologically efficient at perceiving the world around them, they are still inept at carrying out the daily activities and struggle to open doors, walk, and more.
"Their action and manipulation capabilities pale in comparison to those of a two-year-old." says Gupta.
Ph.D. students who are part of this team reveal that human interaction is important as it simply reveals the visual dynamics of the real world. So the interaction forms the crux of the research work.
The group showcased its first demo last year. This demo assisted the team to land a three-year award from Google worth $1.5 million. This money shall be used to expand the number of robots, which will be used to gather more data and information. The team also looks forward to conducting research using more advanced robots.
The researchers are also teaching the robots to learn other skills. The adversarial learning method used by the team is akin to the teaching methodology of parents tutoring their child in activities such as how to catch a ball, by slowly increasing the difficulty level. Apparently, researchers have discovered that this method is much faster compared to any other system.
Checkout the video below to see the robots trying to grasp the learning process