Available online at ScienceDirect. Procedia Computer Science 103 (2017 )

Transcription

1 Available online at ScienceDirect Procedia Computer Science 103 (2017 ) XIIth International Symposium «Intelligent Systems», INTELS 16, 5-7 October 2016, Moscow, Russia Network-centric supervisory control system for mobile robotic groups Vladimir Muliukha a, *, Alexander Ilyashenko b, Leonid Laboshin a a Peter the Great St.Petersburg Polytechnic University, 29 Polytechnicheskaya st., Saint Petersburg , Russia b Russian State Scientific Center for Robotics and Technical Cybernetics, Tikhoretsky Prospect 21, Saint Petersburg , Russia Abstract The paper analyzes network-centric control methods for mobile robotic groups. We consider the task of controlling a group of mobile robots, performing reconnaissance for space and terrestrial applications. In the paper each robot is a cyber-physical object consisting of mechatronic part and computer module, which analyzes data from sensors and generates commands. The use of network-centric approach allows separating these two parts, transferring resource-intensive computing tasks in a highperformance hybrid cloud environment. A cloud environment as a single point for processing of all sensory data allows us to use the advantages of centralized control method and to manage the resources of mobile robots more effectively. In this case, each robot keeps the autonomy if there is no communication with the control center. So a key element of such control systems is reliable communication links. In our work we use redundancy and duplication of various communication channels to provide a continuous connection between a mobile robot and control center The Authors. Published by by Elsevier B.V. B.V. This is an open access article under the CC BY-NC-ND license ( Peer-review under responsibility of the scientific committee of the XIIth International Symposium «Intelligent Systems». Peer-review under responsibility of the scientific committee of the XIIth International Symposium Intelligent Systems Keywords: network-centric control system; cyber-physical approach; mobile robots; high-performance cloud system. 1. Main text Modern researches in the field of control systems are focused on the dynamic distributed systems. Complex engineering tasks concerning control for groups of mobile robots are developed poorly. Recently lots of new * Corresponding author. address: vladimir@mail.neva.ru The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license ( Peer-review under responsibility of the scientific committee of the XIIth International Symposium Intelligent Systems doi: /j.procs

2 506 Vladimir Muliukha et al. / Procedia Computer Science 103 ( 2017 ) methods appeared like cyber-physical approach to the description of robots 1,2 and network-centric method for the description of information exchange 2,3. The paper analyzes network-centric methods for control of mobile cyber-physical objects like robots. The research is done in the framework of the agreement between Peter the Great St.Petersburg Polytechnic University and the Ministry of Education of the Russian Federation. In this work at the current stage a group of mobile robots should realize reconnaissance missions, later also expected to add some manipulative operations and assembly work. The final application of the developed technologies can be the use of robotic groups to explore Mars and to build a base on its surface. The whole project is implemented by a consortium composed of Peter the Great St.Petersburg Polytechnic University, Russian State Scientific Center for Robotics and Technical Cybernetics, Center for Laser Technologies and Science and Technology Center ROCAD. In the framework of this project Peter the Great St.Petersburg Polytechnic University develops and researches network-centric supervisory control system for mobile robots. In our work mobile robots interact with each other by transmitting information via computer networks. Each robot is considered as a cyber-physical object consisting of mechatronic part and computer module, which analyzes data from sensors and generates commands. The use of network-centric approach allows separating these two parts, transferring resource-intensive computing tasks in a high-performance hybrid cloud environment. Data processing in the cloud is conducted is based on the actor model of calculations. A cloud environment as a single point for processing of all sensory data allows us to use the advantages of centralized control method and to manage the resources of mobile robots more effectively. A data exchange between the actors of mobile robots in cloud environment can significantly accelerate the processes of information interaction of mobile agents. Actors describe physical and informational properties of controlled objects and control devices and also exchange information on behalf of the agent. The selection of cyber-physical systems as a special class of controlled objects is due to the necessity of integrating into classical robot various components responsible for computing, communications and control processes. Cyber-physical approach in robotics allows taking into account the characteristics of information processes that are generated during the operation of the physical objects (like robots) and are available for monitoring, processing and transmission via computer network. 2. Network-centric control system architecture In the paper, we propose architecture of a model for network-centric control system, which includes one stationary unit and two mobile ones. Stationary unit consists of heterogeneous high-performance cloud computing and the routing unit, the mobile node also includes a routing unit and a mobile computer. Routing unit monitors the status of the available communication channels and selects a free channel with the highest transmission rate. For the information exchange between the routed units, there are three types of channels: laser network, which is the primary and has the highest bandwidth (1 Gb/s) and range (up to 2 km); Wi-Fi network that can transmit data stream at a rate up to 300 Mb/s for a distance of 600 meters; MESH Wi-Fi network that can transmit data at the lowest rate consistently from one mobile robot to another. Laser network also has the possibility to organize a hierarchical communication system from remote mobile robots to a stationary unit. Each mobile robot has 4 on board two-way laser communication antennas, three of them collect data from remote robots and one uplink antenna transmits the data to the stationary unit for processing. There is a possibility of implementing multiple levels of such hierarchy. Thus, a stationary unit receives a stream of data from a large number of mobile units and should have computing resources that allow to process obtained Big Data 4. So the basis of the stationary unit is a high-performance hybrid cloud environment.

3 Vladimir Muliukha et al. / Procedia Computer Science 103 ( 2017 ) Figure 1. Functional hierarchical scheme of cloud computing environment. The base idea for the implementation of cloud technologies in the robotics is based on the paradigm of digital physics 5 ( it from bit doctrine ), which implies that for robots the whole universe is measurable and computable. It means that if an object of the real physical world cannot be converted to the information from a sensor, for the robot and its control system such an object does not exist. So the key function of the robot s control system is to process data received from sensors into the commands for actuators of the robot. For a number of situations, when such processing should not be performed in real time, it can be performed remotely in the cloud. The application of cloud for the solution for modern robotics tasks is called cloud robotics 6. The use of this approach reduces the processing load on each robot in the group and increases the efficiency of all cyber-physical objects by increasing the duration of battery life and reducing the redundancy of robots characteristics for its tasks. Researches in the field of cloud technologies in robotics are relevant and have been particularly active in recent years. For example, Rapyuta platform should be noted (the RoboEarth Cloud Engine). It allows robots to share knowledge, through a centralized knowledge base, avoiding duplication of information. Rapyuta is a cloud platform, in which the robots can create their own computing environments to perform intensive data processing. These created environments may be used with one or more robots simultaneously. This work is realized using the cloud based on modified OpenStack platform that is installed on heterogeneous high-performance computers containing SIMD and MIMD processors. In our work, we offer the following solutions: a central server located on a secure heterogeneous cloud receives data from sensors of all available cyber-physical objects (mobile robots), processes the data and commands from the operator, and sends control commands to cyber-physical objects.

4 508 Vladimir Muliukha et al. / Procedia Computer Science 103 ( 2017 ) Figure 2. Appearance of controlled mobile robot. Figure 3. Scheme of sensory information flows. The developed cloud software performs a decomposition of the complex target, which is set by operator, into the simple operation, transmits them to mobile robots, and checks their implementation, as well as provides synchronization of commands execution when it is needed. The use of the heterogeneous computing environment allows effectively realize a wide range of tasks, including processing of streaming data from cyber-physical objects (e.g., video or lidar data). An actor approach resolves the problem of dividing a whole system into several parallel streams allowing the horizontal and vertical scaling of the developed system. A functional hierarchical scheme of cloud computing environment is shown on Fig. 1.

5 Vladimir Muliukha et al. / Procedia Computer Science 103 ( 2017 ) As part of this work a following concept was developed: each robot or cyber-physical object consists of mobile hardware and software part, so called agent and virtual avatar represented by a set of computational processes that are implemented in heterogeneous cloud computing environment, and are interacting with the agent using wireless high-speed networks. Such operating model of cyber-physical objects, in which onboard computing resources are supplemented by cloud computing environment s services, can solve a variety of control tasks, and provide a number of important benefits, including: advanced class of algorithms for control and operations planning because of integrating all available information resources from cyber-physical objects; ability to store and to structure large volume of sensory data; common data space and situational awareness of all agents in the system; possibility of rapid change in avatar algorithms (reboot avatar of cyber-physical object in real time); interaction between avatars through the high-speed data network in the cloud environment. High Performance cloud environment shown in Figure 1 is a mobile robot control center. The appearance of a controlled object is shown in Figure 2. The mobile robot is a six-wheel platform with the following installed sensors: forward looking LIDAR, circular scanning LIDAR, two circular view cameras and two different navigation systems, as well as the platform equipped with a hardware for laser and radio communication and several mobile computing devices. As noted above, the implementation of network-centric supervisory control system is based on the networkcentric approach. So the key elements of the control system are the information flows between mobile robot and centralized high-performance cloud system and communication channels, providing data with a requested quality of service. Sensory information flows scheme for the control system is shown in figure 3. Mathematical model of network-centric supervisory control system has been developed to describe the effect of the network characteristics on the sensory information flows. 3. Model of network-centric supervisory control system Model data streams generated by large amount of devices: LIDARs, video cameras, navigation systems. The total data flow rate is greater than even the possibility of laser communication system. As data flow intensity only from optical navigation system can reach 1Gb per sec. One more high-intensity flow is getting formed by another robot in system in MESH mode. To avoid clogging network, divide these data streams into two types: high-priority and low-priority. For the operation of the system first need to be sure that highest priority has data from surround view LIDAR, the navigation system and retransmission of the same data from another robot in the absence of its connection to the control unit directly. As low-priority data used data from other sensors like optical navigation system, video cameras and front view LIDAR. Systems with priorities allow to reduce the probability of network packet loss 7,8,9. It is proposed to process data transmitted in two stages. At first, we need to filter low-priority data in order to ensure the transmission of high-priority data, and then determine which of the communication channels to use for transmission. As a model of the filtering device, consider queuing system with two incoming flows, one service channel, specify priority for one of flows on service and taking place in storage. Such systems belong to the class 10 / /1/ / i M2 M k f. j Such systems are discussed in detail in 7,8 and 9. As results can be used expressions for probabilities of packet loss and an effective way to manage them by using randomized push-out mechanism in which the packet from system buffer is not pushed-out with a given probability instead of 1. Also, have been identified the effects of linear behavior and "locking" for loss probabilities in case or randomized push-out mechanism. For these effects were calculated areas in which they take place, and using their boundaries can help to control the behavior of the system without additional calculations in real time and affect its performance. In considered model to find an average service time can be used model of type M / M3 /3/ k. In this model each service channel has its own intensity and can be used for modeling three different types of network channels.

6 510 Vladimir Muliukha et al. / Procedia Computer Science 103 ( 2017 ) Getting average time that packets spend in such model can be used for getting intensity of service in previous model of type / /1/ / i M. 2 M k f 4. Conclusion j The paper analyzes network-centric control methods for mobile robotic groups. Each robot is a cyber-physical object consisting of mechatronic part and computer module, which analyzes data from sensors and generates commands. The network-centric approach allows separating these two parts, transferring computing tasks in a highperformance hybrid cloud environment. In the paper we proposed the functional hierarchical scheme of cloud computing environment. A key element of proposed control systems is reliable communication links. In our work we use redundancy and duplication of various communication channels to provide a continuous connection between a mobile robot and control center. The resulting system was modeled using queuing theory. A two level model was considered. On first level used priority queuing model with two incoming flows and randomized push-out mechanism for filtering data and on second level was used basic queuing model with one incoming flow and three service channels and limited buffer size for modeling transfer node with three types of channels: laser, Wi-Fi and mesh. Acknowledgements. This research was supported by the Ministry of Education of the Russian Federation an agreement from , research id: RFMEFI57815X0143. References 1. Zaborovsky V, Muliukha V, Ilyashenko A. Cyber-Physical Approach in a Series of Space Experiments Kontur. Lecture Notes in Computer Science 2015;9247: Zaborovsky V, Guk M, Muliukha V, Ilyashenko A.Cyber-physical Approach to the Network-centric Robot Control Problems. Lecture Notes in Computer Science 2014;8638: Lukashin A, Zaborovsky V, Muliukha V. Cloud-Centric PaaS Framework for Robots Operation. Proceedings of the 17th Conference of Open Innovations Association FRUCT 2015; Zaborovsky V, Muliukha V, Popov S, Lukashin A. Heterogeneous Virtual Intelligent Transport Systems and Services in Cloud Environments. Proceedings of The Thirteenth International Conference on Networks 2014; Wheeler JA. Information, physics, quantum: the search for links. Proceedings III International Symposium on Foundations of Quantum Mechanics (Tokyo) 1989; Kehoe B, Patil S, Abbeel P, Goldberg K. A survey of research on cloud robotics and automation. IEEE Transactions on Automation Science and Engineering 2015;12(2): Ilyashenko A, Zayats O, Muliukha V, Laboshin L. Further Investigations of the Priority Queuing System with Preemptive Priority and Randomized Push-Out Mechanism. Lecture Notes in Computer Science 2014;8638: Muliukha V, Ilyashenko A, Zayats O, Zaborovsky V. Preemptive Queuing System with Randomized Push-Out Mechanism. Communications in Nonlinear Science and Numerical Simulation 2015;21(1-3): Ilyashenko A, Zayats O, Muliukha V, Lukashin A. Alternating priorities queueing system with randomized push-out mechanism. Lecture notes in computer science 2015;9247: Gnedenko BV, Danielyan EA, Dimitrov BN, Klimov GP, Matveev VF. Priority queueing systems (in Russian). Moscow: Moscow State University; 1973.