Sergey Popov

Robust transfer learning in multi-robot systems by using sparse autoencoder

Robust algorithms for transfer learning in multirobot systems based on elements of the deep learning are proposed in the paper. The algorithms are based on using the sparse autoencoder. The main ideas underlying the algorithms are to extend the set of set-valued observations by training examples having uncertain weights and to apply the robust minimax strategy in order to find an optimal autoencoder for dealing with set-valued observations. An interesting scheme for transfer learning is considered for which source learning set is reconstructed by means of the sparse autoencoder trained on the target learning set.

Network synchronization of vehicle multiprotocol unit system clock

Recent achievements of automotive telematics in the area of communication channels integration for providing a persistent bidirectional link between vehicle and cloud infrastructure have intensified research in the field of mobile multiprotocol networks of intelligent vehicles oriented on cloud and fog environmental services. Synchronization of multiprotocol unit system clock is crucial for intelligent vehicle networks in terms of security system functioning, navigation, driver and passenger services. In spite of the fact that clock accuracy requirements are comparable with those for stationary systems, limited lifetime route to server in the cloud or fog and substantial restrictions on wireless network traffic complicates the achievement of this objective. Method of mobile multiprotocol unit synchronization in dynamic wireless networks of different technologies with virtual cloud servers is presented drawing on Network Time Protocol (NTP). This method provides the required quality of multiprotocol unit system clock accuracy while minimizing network traffic. Synchronization path selection algorithm is described, as based on probabilistic approach and synchronization quality retrospectives in the chosen technology network. The way of local network traffic reduction while maintaining the required accuracy of multiprotocol unit system clock is shown. The method involved can be used for multiprotocol unit synchronization in intelligent transportation networks.

The rules selection algorithm for network traffic of robot groups in intelligent transportation systems

The article investigates the algorithm for distributing wireless network traffic in a group of robots. Within the framework of the research, an experiment was performed to compare the dynamic and static aggregation methods for wireless data transmission channels in the autonomous robots networks. The results of the conducted experiment showed that the static method of channel policy allows to increase the data transfer rate in networks with unstable connection, the dynamic method allows to efficiently distribute the load between channels when accessing communication data in conditions of spatially situational uncertainty. The results obtained allow us to justify the variant of aggregation of channels depending on the scenarios for data transmission in intelligent transport systems.

Implementation of the Algorithms to Retrieve the Data About Local and Global Wireless Networks from the Telematics Map.

This paper describes implementation and research of the algorithms to select the data about surrounding wireless networks from the moving vehicle. The data are retrieved from the telematics map, which is a cloud service containing the data about all the available wireless networks in the region. The paper contains the description of three scenarios of data extraction, relational queries to the telematics map which serve these scenarios, and the experiment to test the data extraction from the cloud service under real road environment. The experiment has shown that the time needed to extract the data about available local and global wireless networks does not exceed 0.2 sec, which is acceptable for the tasks of scheduling the wireless connections between the vehicle and the cloud services during the whole route. The results of this work may be used to retrieve the list of available wireless networks in the algorithms of intelligent scheduling of bidirectional data transmission for the connected vehicles.

Siamese neural network for intelligent information security control in multi-robot systems

Anomaly detection of the robot system behavior is one of the important components of the information security control. In order to control robots equipped with many sensors it is difficult to apply the well-known Mahalanobis distance which allows us to analyze the current state of the sensors. Therefore, the Siamese neural network is proposed to intellectually support the security control. The Siamese network simplifies the anomaly detection of the robot system and realizes a non-linear analogue of the Mahalanobis distance. This peculiarity allows us to take into account complex data structures received from the robot sensors.

The Technology of Management of Data About Wireless Networks for Vehicle’s Telematics Map

The article studies temporal characteristics of functioning of the system of registration and data updating about the wireless networks signal level by the vehicle telematics card. The article presents algorithms of placement and retrieval of data about the signal level of wireless local area network of a geographical region into the database of multi-protocol unit of the vehicle; conditions and results of experiments on the study of functioning time of the system of database management of telematics cards. The experiments showed that the technology for collecting and updating map data on 2 Hz request frequency can be applied. The results of experiments can be used as a basis for development of a specialized layer of GIS to provide information services of a moving vehicle to a driver and passengers.

Detection of anomalous behavior in a robot system based on deep learning elements

The preprocessing procedure for anomalous behavior of robot system elements is proposed in the paper. It uses a special kind of a neural network called an autoencoder to solve two problems. The first problem is to decrease the dimensionality of the training data using the autoencoder to calculate the Mahalanobis distance, which can be viewed as one of the best metrics to detect the anomalous behavior of robots or sensors in the robot systems. The second problem is to apply the autoencoder to transfer learning. The autoencoder is trained by means of the target data which corresponds to the extreme operational conditions of the robot system. The source data containing the normal and anomalous observations derived from the normal operation conditions is reconstructed to the target data using the trained autoencoder. The reconstructed source data is used to define a optimal threshold for making decision on the anomaly of the observation based on the Mahalanobis distance.