Research On Autonomous Run Control Of Intelligent Vehicle Based On Multi-Agent System

Intelligent vehicle(IV),which is a comprehensive intelligent system which integrates the function of environment perception,planning and decision-making and driving assistance,etc, is a product of that contains multi-technical subjects such as computer vision,artificial intelligence,auto-control theory and electronic technology. IV can successfully reach its destination or move along a desired path in an environment characterized by terrain. Therefore, it has a highly expansive application prospect.This paper take a four-wheels IV as the control object. A high precision line scan CCD camera,high speed DSP and ultrasonic sensors are used to build the vision guildance system,the motion control system and the obstacles avoidance system of IV. The paper center is attached to discuss image processing,intelligent control,path planning and multi-motors controlling etc.The major work in this paper can be summarized as the following:1)Mult imagine processing methods are studied to enhance the adaptability and flexibility of vision system and solve different image flaws,such as fuzzy,distortion and noise.These imagine processing methods are as follows,Firstly,smoothing-filtering approaches to the road image introduced in my dissertation contain neighborhood methods and median methods etc.Secondly,edge enhancing approaches adopted in my dissertation include Robert arithmetic operator,LOG arithmetic operator and Prewitt arithmetic operator etc. Lastly, thresholding segmentation approache to the road image contain simple iterating method of optimum thresholding segmentation. A lot of contrastive experiments are performed to bring forward the optimum thresholding segmentation image processing methods for navigation of IV in view of practical processing effects and real-time performance .This method highlights edge line between the marking line and background, and provides the basis for the calculation of lateral range and angular deviation between vehicle middle line and the marking line.2)IV must own an real-time and active autonomous ability in decision in non-structure or structure environment , therefore, the paper brings forward and builds the Multi Agent System(MAS) composed of coordinated level and control level. on the one hand, different control agent with different algorithm control IV in term of lateral range and angular deviation between marking line and vehicle center line ,on the other hand, coordinated level can give IV more macroscopical control such as giving sub goals,supervising motion of IV and giving optima paths.The MAS proposed in this paper has an active robust ability in local behaviors and long-range optimal planning.3)A intelligent Variable Structure Control based on GA and LS-SVM is proposed. Through offline optimization of parametersδandεusing typical sample data,δandεSVM are established for providing real timeδandεto the Variable Structure Control. This method not only owns the globaloptimization characteristic of GA, but also exhibits good learning generalization and small sample size properties of LS-SVM. Meanwhile, this method is able to eliminate shattering problem of Variable Structure Control due to exponentially reaching law.4) To IV steering system which are of characteristics of complex, nonlinearity and time-variation, a self-adaptive single neuron PID control method is proposed based on RBF Neural Network On-line Identification, which identify the Jacobian matrix of steering system by means of RBF neural network identifier and acquired on-line tuning information of PID parameters, and the self-tuning of controller parameters was implemented by the single neuron controller, and the intelligence control of system was achieved. The result indicates that the control system, compared to conventional PID control method, possesses the advantages of high precision ,great adaptability and robustness.5) A complete vehicle hardware control platform with double chip DSP (DSP2812+DSP643)made by TI as nucleus was built. The system software include two parts: (1)imagine processing software based on BIOS as embedded operation system , (2) measurement and control system. The test and experiments result indicate the MAS has the characteristics of reasonable design,stable performance and realiable operation.