A team of researchers from UC Berkeley developed an artificial intelligence algorithm modeled after human babies, allowing four-legged robots to rapidly adapt to their environment.
Rapid motor adaptation, or RMA, allows robots to perceive surrounding terrain through sensory joints located in their legs. RMA gives robots the ability to adapt to unforeseen environments and situations in a fraction of a second, according to Deepak Pathak, assistant professor of computer science at Carnegie Mellon University and co-author of the study.
“This adaptation model takes into account the history of the robot, like the force that is being imparted onto your legs when walking,” Pathak said. “When you walk on sand, your legs sink into the sand. You can know it’s sand by observing the forces on your own leg as it gets depressed into the sand.”
Previous efforts to improve robot locomotion typically involved preprogramming robots to match their gait to a wide array of set terrains, Pathak noted. However, problems arose when the robots encountered environments they had not been programmed to react to.
Campus researchers set out to solve the problem a year and a half ago, drawing inspiration from how babies learn to walk and adapt to their environment during the first few years of life, according to Ashish Kumar, campus graduate student and co-author of the project.
“We are inspired by how children learn,” said Jitendra Malik, campus professor in the department of electrical engineering and computer sciences and project co-author. “What’s the secret? The secret is that they are observing the world, they are interacting with the world, they are experimenting with the world.”
In the future, the researchers plan to advance RMA’s capabilities by adding vision sensors and extending RMA to robots’ fingers so that they can grasp objects and even open bottle caps, Pathak added.
According to Malik, RMA would allow these advanced robots to play vital roles in elder care.
“They need things fetched, they need to be watched over in case someone falls and you need a 911 call,” Malik said. “Pets serve some of those roles, but pets cannot open the fridge and get you something from the fridge, whereas I can imagine this robot doing that.”
Pathak noted that the biggest scientific contribution of their work is the element of adaptation.
Applications of their adaptation model can extend past robotics into areas including intelligence infrastructure, Malik added. Artificial intelligence-programmed ramps could moderate traffic.
“Adaptation is the hallmark of intelligence,” Malik said. “There are various ways of adapting to changes, but what we are adding is one powerful new technique.”
Contact Rachel Raps at