Humans develop models of our bodies as infants, and robots are beginning to do the same thing. a team at Columbia Engineering has revealed that they have developed a robot that, for the first time, can learn a model of the entire body from scratch without any human help. In a recent paper published in Science Robotics, the researchers explain how their robot built a kinematic model for itself and how it uses this model to plan movements, complete goals and avoid obstacles in a range of situations – even damage to its body is automatically detected and corrected.

The researchers placed a robotic arm inside a circle made up of five streaming video cameras. The robot watches itself through the cameras as it rises and falls freely. Like a baby exploring itself for the first time in a hall of mirrors, the robot twists and turns to understand how its body moves in response to various motor commands. After about three hours, the robot came to a halt. Its internal deep neural network had finished learning the relationship between the robot’s motor movements and the volume it occupies in its environment.

“We were really curious to see how the robot imagined itself,” said Hod Lipson, professor of mechanical engineering and director of Columbia’s Creative Machines Lab, “but you can’t just peek into a neural network; it’s a black box. ” After researchers struggled with various visualization techniques, the self-image gradually emerged. “It is a gently flickering cloud that seems to engulf the robot’s three-dimensional body. As the robot moved, the shimmering cloud gently followed it. The robot’s self-model is accurate to about 1 percent of its workspace,” Lipson said.

Self-modeling robots will lead to more self-reliant, autonomous systems

A robot’s ability to model itself without the assistance of engineers is important for many reasons. It not only saves labor but also allows the robot to keep up with its own wear and tear and even detect and compensate for damage. Researchers believe that this capability is important because we need autonomous systems to be more self-reliant. For example, a factory robot could detect when something is moving incorrectly and compensate or call for assistance.

“We humans obviously have a concept of self,” said Boyuan Chen, the study’s first author on the paper, “Close your eyes and try to imagine how your own body would move if you were to take some action such as extending your arm forward or taking a step backward. Somewhere in our brain, we have a concept of self, a model of self, that tells us what volume of our surroundings we occupy and how that volume changes as we move.”

Self-awareness in robots

This work is part of Lipson’s decades-long search for ways to give robots some form of self-awareness.” Self-modeling is a primitive form of self-awareness. If a robot, animal or human has an accurate model of itself, then it can function better in the world, can make better decisions and has an evolutionary advantage.”

The researchers are aware of the limitations, risks and controversies surrounding giving machines greater autonomy through self-awareness, and Lipson is quick to acknowledge that the kind of self-awareness demonstrated in this study “is insignificant compared to humans, but you have to start somewhere. We have to go slowly and carefully so that we can reap the benefits while minimizing the risks.

The post Colombian engineers develop a robot that has learned to understand itself rather than its surroundings appeared first on TechGoing.