The Research Of AUV Decision-making Control System Based On ELM

Underwater Robot is also named Autonomous Underwater Vehicle. It is a kind ofrobots working underwater in extreme dangerous environment. The underwaterenvironment is full of danger. The depth human beings can dive is limited. Under thiscircumstance, AUV is becoming an important tool which human beings can use todevelop the ocean. Whether or not the AUV can finish its task mostly depends on theperformance of the Control and Decision System of AUV. In order to improve theperformance of the Control and Decision System of AUV, this article aims atdesigning a new set of Control and Decision System based on Extreme LearningMachine Theory.This article describes the AUV decision-making control system based on theExtreme Learning Machine which has many advantages. To build a more rationalstructure of decision-making of the AUV control system, the AUV’s movementbehaviors system is divided into two kinds: rational behavior and perceptual behavior.The rational behavior is the behavior which contains an advanced algorithm. Theperceptual behavior is the behavior which contains a simple algorithm behavior.Emotional behavior is good at real-time performance. Rational behavior is good atdecision-making.In order to ensure the AUV successfully complete its task, thesystem the article describes must be composed of both behaviors.For the decision-making and control system described above, the final output ofthe complex behavior of the decision-making and control system is the weightedvector sum of simple behaviors. Simple control behaviors are already set in thedecision-making system such as obstacle avoiding behavior and target trendingbehavior. In other words, the final act of AUV is determined only by the weightingfactor matrix of simple behaviors. In this paper, the Extreme Learning Machine isused to calculate the AUV’s weighting factor matrix to achieve the goal of precisecontrol of AUV’s behaviors. The output of ELM is the weighting factor matrix that can directly determine the act of AUV. The input of ELM is the characteristic valuematrix of the environment. The characteristic value matrix of the environment is theobtained through AUV’s real-time calculation based on the data sensors. Therefore,the weighting factor matrix changes as the environment changes. As a result, AUV’sact will change. This makes the AUV have a good ability of adapting to newenvironment.In the last part of the paper， two most representative behaviors ofdecision-making control system based on ELM are selected to be simulated in thesimulation platform based on MATLAB. The simulation results are AUV’s navigationtrajectory maps in two-dimensional space and the AUV’s navigation speed simulationdiagrams. Through analysis and research on these simulation results, we can see thatin this paper, the design of decision-making control system based on the ExtremeLearning Machine is able to make AUV successfully avoid obstacles and reach thetarget location even in a relatively complex environment.