5 trends of robotics: neural networks, speech and emotion recognition, navigation and security systems

Neural network technology

Every day there is news about new applications of neural networks. Created in the USA

neural network that animates 2D images:it processes data about the object, separates it from the background and other objects, and then creates a 3D model and the mechanism for its movement, filling in the background that was previously covered by the object. In Israel, a neural network determines intelligence by the shape of the skull - the system analyzes a person’s facial features and, based on them, determines what features are embedded in his DNA. In Russia, they are creating a neural network capable of trading cryptocurrency - analyzing the market and making forecasts.

Artificial Neural Network (INS)— mathematical model (as well as its softwareor hardware embodiment), built on the principle of organization and functioning of biological neural networks - nerve cells of a living organism. This concept arose while studying the processes occurring in the brain and trying to model these processes.

There are many applications of this technology. For example, a neural network should distinguish dogs and cats. To adjust the algorithm, a large array of signed images of cats and dogs is given. The neural network analyzes the features of the objects in these pictures and builds a recognition model that minimizes the percentage of errors relative to the reference results.

By the way, when Google asks you to confirm thatyou are not a robot, and note the traffic lights or buses, you do not pass an elementary test for cognitive abilities, but teach the neural network to distinguish objects of road infrastructure. The system will be used in drones.

The use of neural networks is unlimited.However, the most widely used in robotics neural networks found in voice assistants and interaction with people. Neural networks provide accurate answers to questions. The quality of the voice assistant depends on them. Among the voice assistants can be identified Alexa from Amazon, Cortana from Microsoft and Siri from Apple. Among the Russian voice assistants are “Alice” from Yandex.

The better a neural network is trained, thebetter, it selects answers to the interlocutor's requests: understands the reverse word order, context, and indirect query. This direction will be one of the most promising in the development of the future. As the Internet changes social processes, the speed of information transfer and the general pace of life have grown tremendously. But the man is becoming more lonely. A companion robot will be relevant, which will satisfy the need for communication, understanding and support.

Digital companions are already emerging in Japan,whose functionality goes beyond simple consultants. They become helpers, friends and even wives. The West is also not far behind: 47 million people in the United States (almost 20% of the total adult population) use smart speakers, Amazon Echo or Google Alexa. Moreover, based on recent research, they use smart speakers, not only to give them commands. 25% take them to sleep with them, 20% joke with them, 15% use as a nanny for children - the column tells fairy tales and helps distract the child.

The ability to maintain dialogue will be one of the key requirements for robotsand one of the most promising and sought-after areas of development.

Speech recognition system

It's one thing to just pick up the answer to the request, butIt is much more difficult to hear and decipher the interlocutor’s request. Accuracy is affected by all elements of the speech recognition system: the training sample and the recognition algorithms themselves.

The quality of the audio stream is affected - the ratiosignal / noise, speech intelligibility and volume. Modern systems are trying to complement the "unheard" through linguistic models - each language has its own stable expressions and stable bundles of words.

That is when they say that for recognitioncontext is used, then it is understood that there is an adjustment for recognition due to additional data, for example, specific usage phrases, as in smart columns, or the use of a specific dictionary for a specific subject area.

Accuracy or recognition quality is counted asthe ratio of correctly recognized words to the number of all spoken words, the false recognition metric is also added as the ratio of falsely recognized words to all recognized words.

LG introduced in 2018 itshome assistant. On stage, David Vanderwal, senior director of marketing, tried to demonstrate Cloi, LG's new home assistant. The size of a little drip coffee maker, Cloi should be on the table, she has no hands and wheels, her head rotates and nods during a conversation. This is a voice assistant designed to help you organize your life.

On stage, Vandervall asked Cloi whenthe washing will end - a relatively simple request had to demonstrate exactly what kind of assistant LG intends to sell. Cloi stunned in response.

“Even robots have hard days,” he triedlaugh it off Vandervol. - So, if we know when the washing will end, then we can synchronize the operation of the washing machine with the dryer and remember that we have chicken in the refrigerator, which expires after three days. Looks like we should cook it. Cloi, join the conversation: what can I cook from a chicken. ”

But even Cloi said nothing to this phrase.

It is not known what caused the failure: a large space with acoustics, the quality of the Internet connection or the flaws in the software. Anyway, the robot-conductor between man and equipment simply did not work.

Google is considered the flagship of the industry - already a lotFor years, he has been developing in this area, achieving the highest accuracy and stability of recognition. In the summer, he presented an updated voice assistant - and argues that the recognition accuracy (English) is comparable to human. To process a request and issue a relevant response, Google system takes no more than a second.


However, to use the recognition system fromGoogle needs to acquire licenses, and it is quite expensive. In addition, a speech recognition device can be used in a wide variety of environments. The recognition parameters for the home assistant and the robot are different in places with high concentrations of people. And the specifics are being tried by the developers of Promobot - they are developing a system of microphone arrays and offline recognition. This will allow robots to be less dependent on the quality of the Internet connection and remain a good conversationalist, both with unstable connections and in noisy environments.

Face and Emotion Detection Technology

In order to organize a qualityThe interaction between man and robot, it is necessary to understand who is in front of the machine and what emotions a person is experiencing. Such data will allow him to choose the most effective communication strategy, to make relevant offers. For example, to offer a discount on your favorite variety of ice cream so that you do not be sad, or to report on a promotion in the cosmetics department, if the robot sees a girl in front of you.

The Russian company VisionLabs offers the mostdifferent branches of application of this technology. LUNA face recognition platform will allow the owner to open the car without a key and pay for purchases with a selfie. Facial recognition from VisionLabs is implemented in the Sberbank school access system, used to verify a student during exams at the Moscow Institute of Psychoanalysis.

If we talk about the recognition of emotions, itclaimed by the industries where the service is expected. For example, Alfa-Bank is testing a client's emotion recognition system. The algorithm analyzes the visitor's facial expression, after the service gives an estimate. So the bank receives feedback without resorting to surveys and interviews.

Promobot and Neurodata Lab launched a pilota project of an empath robot capable of recognizing up to 20 emotional states of a person. In accordance with the recognized emotion, the robot will build communication - to encourage or reassure the interlocutor, will begin to joke and bold if he sees a positive reaction. While the project is at the testing stage, however, the robot has already been submitted to CES-2019.

Navigation system

Depending on the tasks, there are outdoor- andindoor technology Outdoor navigation is needed for unmanned vehicles and aircraft, indoor navigation is for security and service robots in buildings.

Today there are two types of navigation: global and local. Global suppose navigation on satellite systems, they are in demand in outdoor systems, but unsuitable for indoor. There is not always a connection and low accuracy of position display. Local includes navigation through ultrasound, optical and infrared systems. Existing systems are expensive, so the main challenge for 2019 will be to reduce their prices.

For example, the sensor system for car Teslacost several hundred thousand dollars. But due to the large spread of cars with autopilot, the cost of the lidar decreased due to the transition from a niche and expensive product to a widespread area. And also with the advent of affordable and cheap microwave solutions in robotics, millimeter-wave radars are beginning to be used, which was previously only available as an expensive premium car option.

For example, startup Marvelmind has created a highly accurate$ 349 indoor navigation system. However, its operation requires four stationary beacons and one mobile beacon, which makes it difficult to use it in large areas and outdoors.

Navigation devices from Marvelmind

If we talk about robots in the usual sense, thenIn order for the robot “Promobot” to move independently, without collisions, the developers use almost all types of measurements: ultrasound, infrared short-range sensors, lidars. This provides the maximum level of security for the movement of the robot.

Information Security

Security is the most important direction of robotics. After the decline of euphoria about robots, people began to think about the security of themselves and their data.

Information Security Threat Trends inRobotics, in general, do not strongly depart from the common ones in the information environment. The development of the Internet of Things has influenced the active distribution of botnet networks, which, unfortunately, is also relevant for the overwhelming number of robotic devices.

Manufacturers often neglect serious protection against cyber threats or even neglect it, which leads to the use of robots for the purpose of espionage, phishing or data theft.

Recently, Positive Technologies researchershave noticed that vacuum cleaning robots are eavesdropping on their owners and transmitting this information via the Internet - and they can even mine cryptocurrency. Using vulnerabilities in the security system, an attacker can intercept confidential data through network traffic: it is not only your photos, but even the bank account data.

At the beginning of the year a report was publishedvulnerabilities robot Pepper. The experts were able to transfer third-party files to the device without authentication, and even log in to the superuser account. They were also able to intercept payment information, data from video cameras and microphones.

In terms of severity, todayThe issue is most acute in the field of industrial robotics. At the end of 2018, the number of attacks on information networks of Russian automated process control systems is higher than on banks or individuals, and resonant situations with encryption viruses confirm that even nuclear power plants can become victims of cyber attacks.

One solution to this problem may be the use of AI for security management, which is already being gradually implemented by leading manufacturers of anti-virus systems today.

However, in the next few years the numberthe robots we encounter every day will increase significantly. Human security depends on this criterion - a key area, world robotics will pay attention to this area as early as possible.

The number of automated processes is growing, andit means that more and more robots penetrate our everyday life. Requirements for the quality of work of robots are increasing as they reach the level of the infrastructure unit, and not wonders and quirks. With the development of the most demanded branches of robotics, it is necessary to improve the quality of voice assistants, including speech recognition, the quality of processing requests and answering them. For the distribution of UAVs and service robots, navigation systems should be cheaper. Mainly, to ensure the safety of humans and their data when interacting with robots, it is necessary to eliminate all security vulnerabilities. These are the main challenges for 2019.