Robotics has long developed separately from artificial intelligence (AI), but now automation has no prospects without it. Sberbank analysts reviewed the practice of using AI to create robots
The concept of artificial intelligence is inextricably linked with robotics. In fact, a robot is a machine that can perceive the surrounding reality, interpret it and act accordingly, that is, think.
The appearance of the terms “robot” and “artificial intelligence” is widely spaced in time (1921 and 1956, respectively), and over the past half century, the paths of development of the two areas have either converged or diverged. But now the progress of computing power, a solid amount of practical developments and the availability of information are forcing these disciplines to unite again, Sberbank analysts write in the annual review of the robotics market for 2019.
In the definition of the authors of the study, AI is the ability of programs and devices to interpret data, learn from them and use the knowledge gained to achieve goals, including independently. In turn, AI is divided into two types:
- strong AI – intelligence in a broad sense, capable of solving problems on a par with the human mind;
- weak AI is engaged in solving highly specialized tasks, achieves specific goals.
Today, Sberbank analysts write, strong AI does not yet exist in nature, and in general there are reasonable doubts about the possibility of its implementation. Therefore, artificial intelligence now and in the near future is a weak AI that deals with individual problems and tasks. A set of technologies, grouped by the authors of the report into five types, helps to solve them.
Artificial intelligence technologies
1. Computer vision
It is the processing of visual information to gain knowledge. The basic task within this technology is the detection of an object in images and videos, that is, the realization that a car is depicted in the corner in one photo, and a computer, keyboard and phone in the other. In robotics, the results of detecting objects give the robot an understanding of what to do and how to do it, and also contribute to its learning.
The logical continuation of detection is tracking, that is, at first the object is detected, then tracking of its movements begins. Robots need this to understand the visual scene and learn to predict the actions of other objects, which is indispensable, for example, for unmanned vehicles. Other computer vision tasks are image segmentation (understanding where the floor is, where the wall is, and where the door is) and depth estimation. The latter implies an understanding of the distance to an object and is solved by restoring three-dimensional geometry from a series of two-dimensional photographs.
2. Natural language processing
Communication with a person is impossible without understanding his language. AI specialists take apart individual morphemes, even the emotional coloring of words in the text, sewing it into the program. Robots need such technologies, for them it is like a dialogue box with a person, and it is not just about understanding, but also about responding and learning new concepts.
3. Speech analytics
If language processing is about textual information, then speech analytics is about sound. First of all, this is speech recognition, which by 2019 has already firmly entered the life of people. The next step is speech synthesis, improving the voice qualities of the robot itself and / or the program to the levels of human communication.
4. Decision making
In another way, this technology can be called the automation of processes when they take place without human intervention. Since, again, we are talking about weak AI, tailored to solve individual problems, decision-making technologies are perhaps the most understandable in their purpose. The authors of the review identify several areas of application of such technologies:
- navigation, for example avoiding obstacles, remembering and accounting for the path traveled, localizing oneself in space;
- learning by demonstration, when the robot learns visually or mechanically shown actions;
- emotional interaction, for which the machine needs to understand the mood of the person standing in front of you, superimpose it on its features of the “character” and give the result in the form of “facial expressions” or “emotions”;
- automation of machine learning, that is, a decrease in human participation in it, a partial transfer to self-learning.
Of course, such technologies should be used in conjunction with others: self-navigation along with computer vision, and emotions along with speech analytics.
5. Recommender systems
Remotely, this technology is similar to decision-making, but Sberbank analysts singled it out as a separate item. The reason is the potential for widespread use of recommender systems in service robotics. We are talking about the offer of goods and services, targeted advertising, a selection of movies and music. In relation to robots, the technology can lead to the spread of robots-waiters or sales assistants.
Present and future
Many of the above technologies are already used in robotics, not only in prototypes, but also in mass production. The greatest path has been traveled so far in the areas of computer vision and natural language processing – in other words, in the recognition of visual and textual information.
Already, there are robotic systems that successfully apply certain developments in the field of artificial intelligence. The most famous Sberbank analysts include three types of robots:
- self-driving cars. So far, these are self-driving, not unmanned vehicles. By law, the driver is still required, although it is the car that does the significant work on the perception and assessment of the surrounding reality;
- industrial robots. They have been used in manufacturing for a long time (for example, high-precision machine tools or manipulators for assembling machines), but AI technologies have begun to penetrate here recently, for example, machine learning of robots designed to correct the work of servomotors, or the use of computer vision to evaluate how best to package a product. ;
- kitchen robots. Computer vision helps them locate ingredients and utensils and plan how to cook a meal.
In the future, the development of robotics will occur primarily through a wider and deeper introduction of AI, rather than improving the material and technical base, the authors of the review are sure. They divide the prospects for the development of the market into short-term and long-term, however, they do not name specific dates.
1. Short term innovation:
- the capture of objects and their manipulation will be brought to the level of human actions;
- the mobility of robots, their overcoming obstacles will also be equal in capabilities to human skills;
- a conversation with a robot will be indistinguishable from a conversation with a person;
- the cost and time for programming robots will be reduced, making them cheaper, and the adoption of automation will be wider.
2. Long term innovation:
- by default, each robot will be able to solve any tasks inherent in a weak (highly specialized) AI;
- within the framework of solving their tasks, robots will become completely autonomous, while going beyond them will require human intervention;
- continuous exchange of information and some successful solutions between robots will speed up the process of self-learning;
- robots will not just communicate like people, they will be able to plan behavior taking into account the possible effect on others, in fact, they will develop social intelligence;
- thanks to AI technologies, robots will not only gain basic knowledge of a certain type of activity, but will also be considered highly qualified specialists, for example, as salespeople or nurses.