Particle Filter Estimation Method of Parameters Time-varying Discrete Dynamic Bayesian Network with application to UGV Decision-making

摘要

Unmanned ground vehicles(UGV) autonomous control technology contains lots of methods, which autonomous decision-making is the most important one. The reliability of its strategy determines the result of UGV missions. To improve the environmental adaptability of traditional methods such as Bayesian network and dynamic Bayesian network, this paper proposes a parameter estimation method of particle filter based on the parameter time-varying discrete dynamic Bayesian network (PTVDDBN) for the situation where the parameters change smoothly with time, and applies it to UGV decision-making tasks. First, formally describe the general model of the discrete dynamic Bayesian network with time-varying parameters under the condition of invariable structure. Secondly, a novel method of parameter estimation and inference decision framework of PTVDDBN is put forward. Thirdly, a parameter estimation method of PTVDDBN based on particle filter is proposed. Finally, with UGV cooperative decision-making in battlefield as the application background, time-varying parameter estimation and PTVDDBN model inference and decision-making experiments were carried out. By comparing static Bayesian network and dynamic Bayesian network models, the analysis showed that the PTVDDBN model algorithm can estimate the time-varying parameters more accurately, and the decision-making inference is more reliable.