Robot able to imagine itself created by Columbia University scientists

Scientists at the University of Columbia, New York, have developed a smart robot that can imagine itself and can learn about it, from scratch, with zero prior knowledge of geometry, physics, or motor dynamics.

The research team believes creation of this self-imagining robot could be the first step towards the "self-aware" robotic machines envisaged in science fiction books and films, such as I, Robot. 

The robot was created at Columbia University' Creative Machines Lab and consists of a grasping hand and a four-degree-of-freedom articulated robotic arm. Initially, it moves randomly, with no clue about its shape or existence, but slowly, it attains the ability to "imagine itself" through a process of self-simulation. 

^{The less-than-impressive robotic arm of the 'self-imagining robot'. Image: Robert Kwiatkowski/Columba University}

To start self-simulation, it has to undergo a brief period of 'babbling', which is followed by about 35 hours of intensive computing. The robot's self-simulation model is achieved via what the researchers describe as a deep learning (machine learning) technique, which enables it to build up an internal image of its abilities and structure.

After that, that robot becomes able to use the self-simulator to imagine different situations and adapt to them, they claim.

"If we want robots to become independent, to adapt quickly to scenarios unforeseen by their creators, then it's essential that they learn to simulate themselves," says Hod Lipson, professor of mechanical engineering at Columbia University.

According to the researchers, their robot initially collected about 1,000 trajectories, each comprising of one hundred points. The first self-models of the robot were inaccurate, with robot having no idea about its shape or how its joints are connected. However, after using the technique of deep learning, the self-model became consistent with the physical robot. It became able to grasp objects from specific locations and drop them, with 100 per cent accuracy.

To test if it could detect damage in its own body, the researchers replaced a specific part of its body with a deformed version. The robot was able to recognise the damage, improve its self-model, and perform tasks with little loss of performance.

But, gifting the ability of self-awareness to machines comes with some potential dangers, according to the research team.

"Self-awareness will lead to more resilient and adaptive systems, but also implies some loss of control. It's a powerful technology, but it should be handled with care," they warn in their paper.

The findings of the study are published in journal Science Robotics.