Font Size: a A A

Research On Key Technology Of Autonomous Landed Vehicle Based On Multi-Agent

Posted on:2011-03-13Degree:DoctorType:Dissertation
Country:ChinaCandidate:W X WangFull Text:PDF
GTID:1102360305457827Subject:Mechanical and electrical engineering
Abstract/Summary:PDF Full Text Request
Autonomous Landed Vehicle (ALV) is an integrated intelligent system with functions of environmental perception, planning decision and automatic driving, etc. It's a high-tech integrated system, and a typical representative of distributed intelligent system. With the developments of modern communications technology and intelligent information processing technology, ALV has been achieved or partially close to the practical application. The related technology of ALV have been successfully used in computer-aided vehicle driving system, space landing and detection system, and variety tactical missions of intelligent weapon equipment platform system on the ground, etc. In this paper, the supporting technology of multi-agent system theory of ALV and path planning technology and human-computer interaction technology have been deeply analyzed and researched.Firstly, multi-agent system is an important branch of distributed artificial intelligence. In recent years, the technology of multi-agent system has been developping and is introduced into the research field of autonomous ground vehicle. A new research method and means is developed for the modeling and structure study of autonomous ground vehicle. The autonomous ground vehicle architecture based on the multi-agent system is an autonomous ground vehicle system framework. The system is based on the technology of multi-agent. Each functional agent is organized by the coordination mechanism and the conscious thinking of their own, to finish the specific tasks. The conception and properties of the agent consciousness properties and the containing and relationship between them are analyzed and researched in the paper. The consciousness properties of the agent are described, with the tool of model logic. The awareness model is developed to maintain the rational balance of the agent consciousness in the multi-agent system of the autonomous ground vehicle. The rational balance between the model consciousness properties is maintained by this model in the dynamic and complex environment.Secondly, in the dynamic and complex environment, the global path planning and local path planning of of autonomous landed vehicles have been studied. In the paper some methods and ways for some deficiencies of current study have been described. A reinforcement learning method to solve the issues of difficultly obtaining environmental knowledge, establishing environmental models and poor adaptive capacities has been proposed. It analyzes the learning space, further reduces the amount of learning for global path planning and improves the learning convergence speeds and efficiencies through hierarchical reinforcement learning.Fuzzy control need not depend on the precise mathematical model of the controlled object, which can express and memorize the control experience through the control rules, and infer decisions through fuzzy logic; however, fuzzy control lacks the capability of knowledge acquisition and self-refinement of fuzzy rules. Neural network has a strong self-learning ability and nonlinear approximation capability. Fuzzy control and neural networks are combined to construct fuzzy neural network controller, which is combined with reinforcement learning to guide autonomous landed vehicles to make decision on the global path planning. In the learning process, the expertise can enhance self-adaptive abilities of autonomous landed vehicles in the environment of unknown road information.Thirdly, for the issues of local optimum and large computational problems in local path planning algorithm, this paper has designed a local path planning algorithm of grid potential field based on a rolling window by combining the potential field method and the grid method as well as introducing rolling window algorithm. It changes the dynamically changing driving environment into a relatively static drving environment at a time through the establishment of the grid model. It ensures the optimization on local path planning and meet the real-time requirements of autonomous landed vehicles on the local path planning by taking the grid with least value as the selected strategy of accessible path.Lastly, human-computer interaction technology is an important even the only channel of human-computer information interaction. Friendly human-computer interface can always provide better function and service according to the actual needs of the user, express self-state of intelligent system efficiently and real-timely, and respond to all reasonable requests from the outside. According to the function uncertaninty and the characteristic of singleness of interface definition in the human-computer interaction, through the introduction of configurable design idea of the configuration software, an agent of human-computer interaction was developed. The human-computer agent used multimedia interactive technology, with friend and function configurable interface and abundant function interface.
Keywords/Search Tags:Autonomous Landed Vehicle, Multi-agent System, Path Planning, Reinforcement learning
PDF Full Text Request
Related items