Abstract: Atpresent, theresearch on the kinematics- and dynamics-based longitudinal decision-making system of electric unmanned tracked vehicles are confronted with problems such as poor adaptability and difficulty to obtain accurate model parameters. Aiming at the driving scenarios of the unmanned tracked vehicle straight-linedriving and approaching obstacles, this study introduces the longitudinal decision-making mechanism for driversby analyzing the driving data and constructsa model usingthe combination of Gaussian Mixture Model (GMM) and Hidden Semi-Markov Model (HSMM) to simulate the longitudinal decision-making process of experienced drivers. In the GMM-HSMM system, the GMM is utilized to identify the driving intention as well as cluster and quantifythe driving behavior duringtheobstacle-approachingprocess;the HSMM is applied to model the decision transfer process and the duration of the same decision. This system is verified by a real platform under different road conditions. The experimental results indicate that the proposed driver model canwellsimulate the longitudinal decision-making mechanismfor drivers,where the acceleration is limited to 3.5 m/s², the deceleration is larger than -4.5 m/s², andthe average value of absolute acceleration at the decision boundary approaches 0.8 m/s². Meanwhile, the GMM-HSMM-basedsystem is shown to be able to adapt to different road conditions withoutrelying on accurate road parameters by retraining the decision durationdistribution.

Key words: electrictrackedvehicle, unmanneddriving, drivermodel, longitudinaldecision-makingsystem