People start to get to know their environment and themselves from the very first moment they are born. In this process, we learn that the stove is hot, the ice is cold, that hitting our little toe can hurt a lot, that some dishes are delicious and we act accordingly.
Until now, this feature was unique to living things. Robots were learning their own existence and features from outside information, that is, from developers. A new study seems to have changed that. A team from Columbia Engineering has developed a robot that can sense not only its environment but also itself, and even “imagine” itself.
Will robots have dreams too?
A study by Columbia Engineering revealed that for the first time in history, a robot that can fully learn and model its own body without human assistance has been developed. In the article published in the journal Science Robotics, the researchers also explained how the robot achieved this.
Accordingly, the researchers started by connecting the robot with five cameras that shoot the robotic arm from a 360-degree angle. Seeing the results of its own motor commands in images from these cameras, the robot stopped after spending three hours learning about its movements and itself. According to the study, the robot then used this information to overcome different obstacles, accomplish objectives or avoid obstacles. He even predicted possible damage and made adjustments in his strategies accordingly.
The researchers then began the grueling process of discovering how the robot sees itself in the neural network. At the end of this process, they found that the robotic arm could model itself in a kind of fog, with a margin of error of 1%. When it had to make a decision, the robot first imagined itself doing it. When he could achieve results in his simulations, he took action.
According to scientists, this work will be an important breakthrough for self-sufficient robots. In this way, robots will be able to solve problems on their own, understand themselves, and repair malfunctions or damage.
8122
