Ants secrete pheromones, spit out flesh tongues, gazelles predict danger from shocks... from low organisms to higher organisms, “sensing†is everywhere.
“The people who have evolved for a thousand years need only a few points of information to be able to tell if this is an acquaintance.†When Philip Slusalik, Scientific Director of the German Artificial Intelligence Research Center, mentioned this ordinary ability of humans, he felt that It is mysterious. As an artificial intelligence scientist, he has not yet been able to find an exact model for the ancient biological instinct.
Where does the "sensing" of artificial intelligence come closer to the subtle feeling of "spirituality"? Recently, Apple has published a new artificial intelligence paper that converts the raw data collected by optical radar sensors into 3D mapping maps. This has led to a step forward from pure data to three-dimensional stereoscopic information. Although there is still a considerable distance from “spirituality,†this study can still inspire people to focus on the end of information acquisition and processing in human-computer interaction.
Equipped with "early brain" sensors can be smarter
Two distant relatives of the "Alfa Dog" also fired recently: One was Sogou Chinese "Wang Tsai" which showed "Lip Phrase Recognition" at the Internet conference; the other was "Alfaba" which realized unmanned public transport in Shenzhen.
The former breaks the stereotyped thinking, changes the sensor of semantic recognition to optical sensing for the first time, and judges the communication of the language with the information captured by the image. The latter's flat round sensor is very eye-catching, "Alphaba" rely on it to perceive the road and avoid obstacles.
80% of humans obtain information through the eyes; in the process of artificial intelligence capturing information, the visual sensor also occupies a very important position - currently there are two radar and video methods. Compared with radar, the video is an overall display, and its presentation is not susceptible to interference. However, the radar performs 3D modeling of the surrounding environment, which can contain more depth information than a normal camera.
“The currently used obstacle sensing devices include microwave radars, ultrasonic radars, etc., as well as methods for capturing video images,†said Bi Chao, technical staff at the Beijing Intelligent Vehicle Industry Innovation Center. The sensor devices will be installed on the vehicle side and will also be installed. On the road side, seamless coverage is required, "just like the relationship between mobile phones and base stations to ensure that the signal is unobstructed."
Smoothness is the foundation, and judgment is the key. "The speed of a car requires no one to judge and operate between milliseconds, so calculating the probability of a collision between two cars, between cars and between people must be quick." Bichao said that information is acquired, transmitted, Calculations, feedbacks, and other processes need to be done at one go.
Specifically, there are many calculations before the brakes or alarms—the radar acquires 3D point data, identifies obstacles, compares the obstacles before and after the two frequencies, and identifies static or dynamic. Dynamic obstacles calculate the speed of movement, combine the position information of the driverless car, calculate the minimum safety distance required for obstacle avoidance, and determine the countermeasures.
How can I be faster? The industry has attempted to move the "pre-processing" part of the data center to the sensor side. “We have tried intelligent sensors,†said Liu Zhengzhong, senior engineer at Shengzhe Technology. “The original sensor had a single function of collecting signals, but the signals need to be transmitted back and forth between the sensor and the processing terminal. These need time and energy, if the sensor Can filter out some unnecessary information in advance, the process will be streamlined a lot."
In other words, the previous sensors were strict hand, eyes, ears, nose and other sensory organs, and now they are equipped with a small "initial brain." "This is very effective for video-based sensors because the amount of video data is too large," said Liu Zhengzhong.
"The camera we have studied is very thorough. The radar also needs to continue to study. We also need to explore further, what kind of sensor can be perceived, these subtle changes that humans feel." Philip said. The demand for sensors in the AI ​​market is rapidly changing, and it is urgent to update the new sensors.
New materials and quantum technology allow the sensor to break through the limits
Also looking forward to is the academician of the National Academy of Engineering and Professor Stanford University’s Bao Zheernan. She once said, “We are looking for new materials to bring about disruptive technologies. At present, the carrier that transforms external signals into electrical signals is brittle. We hope to find new materials that can be easily compressed and stretched. We also hope that it will be self-healing and biodegradable."
With new materials as the substrate, the integration of electronic devices will become flexible electronics and artificial skin. The “tough†image of the sensor will undergo a radical change. Not only will it be free to stretch, bend, and rotate, but it will even be able to sweat while accurately acquiring the sense of touch.
According to statistics, the Institute of Semiconductors, Chinese Academy of Sciences, has recently developed an ultra-thin high-resolution flexible electronic skin array. By introducing polymer hollow sphere nanostructures, the sensor exhibits superior sensitivity to ambient pressure and can detect low pressures of 0.6 Pa. “After pulling down and rubbing in different environments, we can still feel the changes in external pressure and temperature. In order to avoid the influence of body surface temperature changes on the device during human physiological signal monitoring, researchers also compensated the temperature of the sensor and improved it. The accuracy of the device in practical applications." This research result was recently published in "Nano Energy."
The transformation and breakthrough that Baozhan Nan looks forward to, also come from another dimension - quantum science and technology to introduce optical sensor research can already reach the application level.
"Can break through the speed and distance limits of existing sensing," MIT conducted a study on single-photon imaging, was published in the "Science" magazine, related researchers, the traditional optical three-dimensional imaging requires high-intensity light While the combination of quantum technology and 3D imaging can achieve an average of 1 photon per pixel, the efficiency of the conventional active light imaging method is several orders of magnitude higher.
Quantum technology is disruptive to the accuracy, speed, and distance of optical imaging. In theory, driverless cars and even unmanned aircraft can be achieved. Because it is sensitive to detection, it can even put sensors on the organelles in the cells to sense cell activity.
If we say that the scale of the telecommunications interconnection is on the order of billions of people-supported carriers, and the interconnection of goods and materials is on the scale of trillions of trillions of yuan, the interconnection of cells as carriers will have a potential beyond human imagination.
Exploring the potential of the market requires the release of the industry
Although the sensor has demonstrated the potential to lead the AI ​​industry into the Blue Ocean at the research level, it is tepid in the industry.
Some analysis articles pointed out that in the field of artificial intelligence hardware, there are chips and sensors in both directions, there are several companies in the chip field domestically, and the sensors are almost all dependent on imports.
"Electroencephalogram measuring instruments are basically imported products, and the accuracy of the sensors is relatively stable," said Zhang Yunhong, a researcher at the China National Institute of Standardization. The laboratory data of Southeast University's relevant disciplines record that a six-dimensional force sensor is about 100,000 yuan, which basically comes from imports.
Compared with the application level and the system level, this “bullet and beam†component, which plays a supporting role, has not reached the eyes of most investors and industrialists. “No storytelling, no investment,†said analysts.
Dai Guozhong, a researcher at the Institute of Software of the Chinese Academy of Sciences, has a more systematic view. "Sensors are an important part of human-computer interaction. Human-computer interaction and artificial intelligence are different directions of effort." He analyzed the history of these two fields for 60 years. The law is drawn: When artificial intelligence is hot, the development of human-computer interaction will be at a low point. When the human-computer interaction is concerned, the artificial intelligence boom will fall back.
This shift can be traced back to when they were born, "Stanford University artificial intelligence laboratory director John McCarthy and others proposed artificial intelligence when they tried to build a computer system that can reproduce human behavior, and MIT psychology and artificial When intelligence expert Joseph Lickard proposed human-computer interaction, it was hoped that the machine could fulfill the task given by mankind.†Dai Guozhong said that the pursuit of the goal is the same, but the research ideas and methods are completely different.
Although in the actual industrial development process, the two are indistinguishable from each other, this academic distinction will, to a great extent, help clarify the research objectives and direction, guide the balanced development of the industry, and avoid "battle to the top". A hot one."
Sensor cold is attracting the attention of industry observers. It is hoped that those unpopular but important AI essential areas will not become shortcomings in the development of the industry, but can promote AI to achieve an overall evolution.
Dongguan Guancheng Precision Plastic Manufacturing Co., Ltd. , https://www.dpowergo.com