In Singapore have created a robot that is 1000 times more sensitive cheloveyniks Wolf
With a glass of water – a task trivial for humans, is a complex multistage quest for any robot. It is necessary to locate the desired subject, appreciate it
form, calculate required to maintain effort, not allowing the object to slip away, but not destroying it.
Most modern robots make their decisions based on the processing of visual data, which severely limits their capabilities. The designers came to the conclusion that in order for robots
able to effectively perform more complex tasks, they need to have the sense of touch, i.e. the ability in real time to correctly process information coming from
the respective sensors.
Serious progress in this direction has been achieved by the team from the National University of Singapore (National University of Singapore, NUS). In their decision they used the sensors
faux leather and vision, and neuromorphic chips Intel Loihi. On this basis, scientists have tried to create an artificial intelligence that simulates biological neural networks – to make operational and
accurate insights about the properties of different objects.
“Over the past few years there have been a lot of useful information in the field of robotic manipulation, however, the Association of tactile and video data to obtain an accurate answer about the properties
the object remains a difficult technological problem,” said Professor of faculty of science NUS Benjamin T. (Benjamin Tee), which is engaged in research in the field
robotics: together with Professor Harold (Harold Soh) of the faculty of computer science and computer engineering, NUS. Scientists have managed to make serious progress, using one
side the newest materials, and on the other the modern computer technology. The results of their joint work was presented at the conference Robotics: Science and Systems Conference in July 2020
year.
Scientists from NUS has applied advanced artificial skin, known as Asynchronous Coded Electronic Skin (ACES), developed by the specialists of the Singapore Institute in 2019.
Based on this material the sensor is able to detect touch is more than 1000 times faster than the sensory nervous system. That is, it can determine the shape, texture and
the hardness of objects up to 10 times faster than the blink of a human eye.
The creation of superfast sensor-based artificial leather has solved about half the puzzles. In addition to required the artificial brain, which ultimately can achieve the desired
the level of perception and training the whole system. Just for this, and were involved in the neuromorphic chip Intel available science institution as part of the research community Intel
Neuromorphic Research Community (INRC).
In their initial experiments, the researchers evaluated the effectiveness of reading Braille artificial hand. Data from its sensors is transmitted to the cloud where microdry transformed into
the semantic value. In the end, managed to achieve accuracy of 92%, while the energy consumption of a neuromorphic chip was 20 times less energy that a similar calculation would have spent the usual
microprocessor.
The next challenge was to classify the robot, connected to the neural network, the properties of various opaque containers containing different amount of liquid. The system with tactile sensors
classification quality was 10% better than the one that focused only on the computer vision. The performance of the neuromorphic processor was 21% higher than
the graphics chip.
The study was supported by the National office of robotics Singapore (NR2PO), an institution that supports the ecosystem of robotics in the country, sponsoring development, which can
use in the public sector and business. In NUS believe that the basis of the research will be able to create solutions for logistics and food industry, where there is a high demand for
robotic automation, especially in the era after COVID.