基于約束型TD3的動(dòng)態(tài)探索噪聲改進(jìn)算法

打開(kāi)文本圖片集

中圖分類號(hào)：TP181；TP301.6；TP242 文獻(xiàn)標(biāo)識(shí)碼：A 文章編號(hào)：2096-4706（2025）07-0103-06

Abstract： Aiming atthe problem that unconstrained exploration maycause damage to the mobile car，thisstudy proposes a ReinforcementLearning methodthatcombinesadaptive noiseexplorationandLagrangemultiplierconstraints，aiming tooptimize thetrajectoryplaningofthecarreachingthe targetpoint.Thismethodimprovestheexplorationefciencybydynamically adjusting the noise，uses the TD3algorithmtodeal with thecontinuousaction space，and uses the Lagrange multiplier method to deal withtheconstraints，whichis diferentfromthe wayofaddingthepenaltyofunexpectedbehaviordirectlyintheMarkov Decision Process（MDP）.Simulation experiments show that this methodcan effectively guidethecar to avoid obstacles，educe theviolationofconstraints，andensurethesafetyandreliabilityofthetask，showinggoodtrainingconvergencecharacteristics.

Keywords： SafetyReinforcementLearning; ConstrainedMarkovDecision Proces;trajectoryplanning;TD3algorithm

0 引言

隨著自動(dòng)化技術(shù)的飛速發(fā)展，機(jī)器人技術(shù)已在工業(yè)制造、服務(wù)業(yè)等眾多領(lǐng)域得以廣泛應(yīng)用[1]，成為提升作業(yè)效率與操作精確度的關(guān)鍵要素。（剩余8693字）