Promobot a Dialogue system for service robots

Main articles:

• Golosovayabiometriya
• Technologies of biometric identification
• Speech technologies: on the way from recognition to understanding
• Neuronets (neural networks)
• Service robots

2020: Representation of a dialogue system for service robots

The Promobot company, the resident Klastera of information technologies of Skolkovo Foundation, developed for the service robots a dialogue system on the basis of neural networks. At the same time it does not need connection to the Internet. On November 12, 2020 the Skolkovo Foundation reported about it.

Before Promobot robots communicated with the help of linguistic base — i.e. a question answering system. Now they will be able freely to distinguish requests, to generalize them on the set subject and to generate answers using neural networks.

This dialogue system allows to compare a voice request of the interlocutor with location of the robot, request time and also with a context of the previous dialog. Promobot will be able to conduct long dialogues, and not just to respond to the single requests as it is done by the existing systems.

The robot, unlike voice assistants, in the answer to a question will be always guided by time and the place of a conversation. For service robots it is the major task as they render service in the specific place, say, at the airport, and always should 'realize' in what context they are, commented Andrey Nosov, the head of development of a dialogue system of Promobot company.

Developers of a dialogue system used 10 key subjects for business: for example, Transport (How to reach? How to purchase the ticket? Where is?) or Building (What address? How to receive the admission? In what office?)

To train one Intentas (need of the user) not less than two thousand relevant samples are necessary (specific examples of a request in a context). For a training of 10 subjects date engineers was necessary to analyze about 6.5 million lines of requests. On it 1000 working days were required.

For increase in accuracy dialog is separated into five modules.

The first relies on the simple communication which is not requiring a binding to any purposes. Its appointment — mitigation or maintenance of dialog.

The second module responds to the requests for business, processing them according to a necessary context.

The third is used as the searcher — for permanent updating of the database of the robot.

The fourth system module is directly connected with the third. It plans comparison of the facts taken from the searcher in cases when it is necessary to receive some comparison or the unloading important for decision making. For example, for obtaining answers to questions: "Where costs cheaper?", "Who is higher?", "It is far better to go?", etc. At the same time it is not obligatory for system to have Internet access at the moment of a request: bases are constantly updated while the robot in network, but the robot does not lose to them access if the Internet is disconnected.

The last, the fifth, the module is connected to linguistic base Promobot which creators of the robot brought together for five years.

All voice assistants who exist for November, 2020, for us in fact only one of system modules which is responsible only for chit-chat. The Internet is necessary for any voice assistant for a full-fledged work. Our solution can function and offline as is guided by specific objectives of business, told Oleg Kivokurtsev, the development director of Promobot company.

Implementation of a dialogue system on the acting Promobot robots will begin in December, 2020. In their our country for November, 2020 about four hundred work.

The Russian solutions in the field of service robotics are in steady demand in the world market. It occurs including because our companies in some aspects are legislators and founders new technology and business of trends in this area, noted Pavel Krivozubov,Robotics artificial intelligence the head "and" of the Cluster information technologies of Skolkovo Foundation.