当前位置: 首页  Research Areas
Research Areas

Research Areas

Autonomous system is a system which doesn’tneed to be completely controlled by the peoplebut can make decision independentlyand finish specific tasks in the system with the abilities of environmental awareness, mission planningand real-time control. Because of the existence of information distribution andmultiple objects, autonomous system is oftencharacterized by information coordination and multiple objects collaborationunder the network environment. A new theory framework of network control willbe formed by the method of multiple objects collaboration andfusion, coordination control theory of multipleobjectsas well as self-organization network theory and technology of autonomous systemoriented.

Thekey lab focuses the following research aspects:

(1)Theory and method of autonomous behavior

●Theoryand method of awareness and positioning

Thepremises to achieve autonomous control include learning, understanding andreconstructing of unknown environment, perceiving the unknown environment, buildinga small world of external environment in the system computer and locating thetarget and ontology in it. For information diversity and complexity obtained byheterogeneous multi-sensors, to analyze the object feature space structure constitutedby multi-source heterogeneous informationto make the unified representationof multi-source heterogeneous information; at the same time, to analyze theselective mechanism of human multi-source information fusion and to build multi-sourceinformation selection mechanism based on environment and task drivencombination, so that it can select adaptively the most effective information inparticular environment and the specific task to improve the reliability ofsensory information.

Thelab researches to carry out multi-source and multi-scale information of rapidconvergence and embedded implementation under the dynamic environment (multipleunmanned helicopter fast search and rescue, etc.); positioning and modeling (SLAM) externalenvironment at the same time is an important algorithm for unknown environmentperception, but in the actual system it’s usually difficult to meet therequirements of real-time and accuracy at the same time due to an excessiveamount of calculation. So it’s an important project to study the efficient andreal-time SLAM algorithm suitable for embedded implementation.

Gridmodel and probability model are both effective modeling methods. How to buildthe grid model to describe the determination and uncertain factors in theprocess of terrain modeling, positioning information and to provide the positioninformation quickly and efficiently for navigation planning, is the importantcontent of current perception positioning theory.

Activeenvironment perception is for the purpose of target environmentperception to realize switching sensors, location optimization and motionsensor node active perception algorithm based on target for research, and to optimizethe dynamic environment coordinate sensor location with the use of constraints suchas information entropy driven index function and network communication, to achieveactive target information acquisition and tracking.

●Autonomouscontrol theory and technology

Autonomousnavigation and planning: navigation is calculating reasonable system trajectoryaccording to the requirements of task in acquired environment model. The labwill merge the fast search algorithm based on grid and trajectory optimizationalgorithm based on virtual potential field, and combining the approximatedynamic programming (mainly for the Markov decision process) to find out the rapid effective path generationmethod. For dynamic and uncertain environment characteristics of autonomoussystem, integrate SLAM algorithm to build the external environment parameters real-time,construct the model predictive control based on mixed integer optimization, andonline plan trajectory of autonomous system real-time, tosolve the positioning and navigation at the same time.

Real-timemotion controlmultimodality robust switch control, priority control;

Forcomplex autonomous system, its typical control structure made up by layer andcoordination, the bottom motion control executes the instructions and planningof higher level; for multiple modals, relatively simple and reliable controlalgorithm can achieve better performance of the system is higher than thetraditional single controller through the proper switching mechanism. But forthe problem of designing the controlling rule and switching rule at the sametime, the major bottleneck is lack of thorough understanding of interactionbetween controlling rule and switching rule. Given that, firstly to use informationof appropriate state/ feedback transformation and system structure, to convert switchingsystem to canonical form or normalized form, then to discuss the common designof controlling rule and switching rule in view of canonical form.

●Diagnosisand the maintenance of unmanned system

Autonomoussystem is applied in more and more occasions (unmanned aerial vehicle (UAV),service robot, etc.), its reliability and fault tolerance has been concerned widespread.Complex autonomous system diagnosing, forecasting faults, and thereconstruction function are the performance of the system survivability and animportant part of autonomous system also.

Thelab research will put emphasis on the reconstruction strategy of integrateddiagnosis model from analyzing the system survivability in the event of failure, and build the integrated system health management based on uncertainty Bayesiananalysis method and the principle of predictive control .

(2)Coordinate control of network autonomous system

Information coordination and cooperation under thenetwork environment of the positioning

Mainlyresearch the multiple objects’ information sharing and integration under thenetwork environment, make cooperation perceptioncooperation orientationcooperation terrainconstruction by using multiple objects to obtain outside information. In thisway, it can upgrade the perception and navigation ability of the whole autonomoussystem, reduce the cost and increase the fault tolerance and reliability.

Thetheoretical basis of cooperation positioning method mainly include dispersedextended Kalman filtering/particle filter, Markov model, etc. The lab will focuson the combination of decentralized filtering method and the combination ofSLAM algorithm, using distributed SLAM to model and locate in unknownenvironment at the same time, and spread the results and realize to thedistributed fusion method of multi-source heterogeneous information.

Thesensor switch scheduling in limited area is an important research content ofdistributed information fusion research, which includes sensor schedulingdriven by information, according to the information gain and communication costdynamically select the sensor, based on the extended Kalman filteringprediction operation sensor area and sensor selection and time-sharing sensorscheduling based on entropy. Multiple objectives approximate dynamicoptimization method, according to energy consumption, tracking accuracy andreliability, will be the feasible research direction of heuristic subprimesensor scheduling method.

●Task coordination and cooperation plan

Groupcontrol and cooperative search is a kind of important coordination andplanning, using the method of extended Kalman to get the covariance matrix ofRiccati equation from network autonomous objects in order to optimize motion trailsreferring to eventual covariance, aiming to meet the target cooperative searchneeds.

Partiallyobservable Markov decision processes (POMDP) is an effective means to control planningin uncertain environment, and the distributed partially observable Markovdecision processes (Dec-POMDP) is an effective mathematical methods to cooperationplanning in uncertain environment. Countermeasure theory is a good method to multi-agentplanning in competition and cooperation, especially bringing approximateDec-POMDP in countermeasure theory to make further practical application on collaborativeplanning algorithm.

Comparedto cooperation search, collaboration of dynamic target tracking is a difficultyfor cooperative planning , it is suitable for field realization of the embeddedreal-time algorithm is one of the important direction of lab.

Taskallocation: Using the optimization and approximation optimization theory andmethod, analyzing and design task coordinated distribution of multiple objects,mainly doing research about mission objectives’ optimal quotas in eachsubsystem of the interaction among autonomous system and outside world of moreheterogeneous objects (with different kinetic and dynamic constraints, sensors,communications and awareness). Based on that the arbitration methods of bidding(bid), auction (auction) and the negotiations (negotiation) and countermeasuresof have been successfully applied in theeconomic system, using focus - distributed allocation algorithm based onmultiple objects’ task decomposition andcoordination in CPN (contract protocol network) to study the approximateoptimal allocation policy in the condition of incomplete information dynamicenvironment.

Human–ComputerInteraction (HCI)

Collaboration and interaction of humanand autonomous systemespecially complexnetwork , further expand the applicationof autonomous system: 1) Due to the distributed cooperative awareness of distributedcooperation awareness, autonomous system enhance the environment perception ofhuman so as to make the system has a higher cognition ; 2) Many people involvedin the autonomous system behavior becomes complex and control face newchallenges.

Autonomoussystem of co-ordination is general described as for optimal control model. Wemake in-depth research of people'scognition and behavior in the network model represents, using implicitexpression form of Markov model to study the quantification of human-computerinteraction in a framework of relativelycooperation awareness, cooperative learning and coordination planning.

(3)Self-organizationNetwork Environment And Network Control Of Autonomous System Oriented

●  MobileSelf-organized network topology structure and network protocol for autonomous system

Self-organizedtopological structure: The connection protocol and topology structure are thepreconditions of network autonomous system, the network bandwidth and distributionof the nodes data rate are important factors to affect the network quality. Thepurpose of the network dynamic topology control is:1) To know the connectionand the quality of network, dynamic topology construction algorithm of mobilenodes navigation control constraint;2) Dynamic changes in network and topologyadaptive fault-tolerant method of nodes frequent adjustment (to join or leave).Currently, existing algorithms on topology control are mostly single objective optimizing ofnetwork topology. Some representative algorithms including: COMPOWLINT/LILTLMN/LMACBTCLMSTRNGDRNG and DLSS. Approximationalgorithm based on adjacent graph to neighboring node information andcomputation is big, it is not applicable in computing capacity-constrained complexsensor network at present. In a hierarchical topology control aspects, TopDiscclusters algorithm have been proposed, which only considers on the premise ofnetwork coverage extent, to form the lowest cluster number, without consideringnodes energy and the robustness of the network. Based on the geographic gridclustering, GAF and its improved algorithm needs the accurate locationinformation of nodes; Establishing an effective complex sensor network topologystructure is a fundamental role to improve the efficiency of data routing andMAC protocoldata fusion, time synchronization and target positioningapplications, etc..

Researchpoints:1) The advanced control strategy and stability analysis of dynamictopology formation; 2) Dynamic topology adaptive fault-tolerant method research.Becauseof autonomous system node exit and rejoin in the second deployment, the dynamicnetwork topology changes frequently.We will use the statistical theory analysisof node failure characteristics, and give the node second deployment strategy,combining with statistical model , to design the distributed adaptive topologycontrol strategy and satisfy the optimal neighbor nodes constraints, so that tooptimize the complex independent network in network communication delaypower consumption andpacket loss rate, etc..

Fornetwork protocol analysis and design of autonomous system, our research will befocused on: 1)Study design the network protocol of mixing time characteristics,to establish the multi-agent information (routing) coordination and cooperationto control demand of communication. 2) On the basis of ad-hoc network protocol,to meet the demand of the task of temporal constraint and make protocoloptimization.

Inheterogeneous network, it appears multiple actuators, which are similar to thesink, and the fundamental changes take place in the communication architecture.Multiple sensor data acquisition and actuator mission cause the problems justlike information orderlysynchronization mechanismactuator redundancy, thereis not any literature concerned. Thecontrol system is very sensitive to end-to-end delay and loss rate. It has todesign system based on analyze data heterogeneous sensor network routingprotocol.

●Controlsystem design based on network

Comparedwith the traditional control system, the communications channel of the networkcontrol system has replaced the traditional signal lines. Autonomous system onthe structure and composition has more flexible, and the control system can getmore information and finish more complex tasks through the network. On theother hand, communication channel brings the incomplete feedback structure tothe feedback system. (Network signal is transport in data packet, and there istime intermittent characteristics on feedback signal) and informationincompleteness.(Sign is the limited word length binary signal which has randomcharacteristics just as error and replacement). The double incompleteness ofthese structures and information brings challenges to the feedback systemanalysis and design.

Thelab will research on the network information structure expression, the controlsystem performance limit analysis in network environment and the control systemperformance essential constraint caused by communication parameters (missinginformation). Make analysis the feedback system stability of the channelparameters and do optimal design of channel parameters, and to provide the completetheoretical framework on network control and control oriented network design.

(4)Application oriented network autonomous systems design and integration

Autonomous technology and application in safety, emergencyand extreme environment

1)Largearea of rapid nucleation detection reconnaissance system based on unmannedhelicopter

2)Three-dimensionalnetwork disaster early warning and environmental monitoring system

3)System in safety and rescue application

4)Remote high voltage line autonomous patrol system based on the unmannedhelicopter platform

●Keytechnology research and integration of intelligent transportation networkautonomous system

●Network autonomous systems design and integration in equipment manufacturing industry