Adaptive PHD-SLAM Based on Strong Tracking and UKF
The traditional probability hypothesis density-simultaneous localization and mapping(PHD-SLAM)method lacked online adaptive adjustment capabilities and was easily affected by uncertain noise,the choice of initial system pa-rameters,and linearization approximation errors,which led to particle degradation problems and subsequently affected the estimation accuracy of the pose and map features of the robot.This study addressed this issue by proposing a PHD-SLAM algorithm based on strong tracking and an unscented Kalman filter(UKF),which integrated the latest observa-tion data to generate importance density(strong tracking UKF PHD-SLAM,SUPHD-SLAM).In the proposed algo-rithm,during the importance sampling stage,the pose and map features of the robot at the previous moment were aug-mented into a joint vector.To avoid the linearization errors introduced by the extended Kalman filter(EKF)in the tradi-tional PHD-SLAM,the UKF was used to predict the particles,and the fading factor in strong tracking filtering was in-troduced to correct the inaccurate pose state covariance after the UKF prediction,maintaining the orthogonality of the measurement innovation and thereby suppressing the influence of uncertain noise and inaccurate initial system parameter settings on state estimation.Subsequently,each pose particle was updated through the UKF,guiding the particles to-wards the high likelihood region to obtain a more accurate pose importance density,thus avoiding particle degradation.New pose particles were sampled from the importance density,and the PHD filter based on UKF was used to calculate the map features for each pose particle.The single cluster(SC)strategy was used to update the weights of each pose par-ticle.Finally,the pose particle with the highest weight and its corresponding map were extracted as the state estimation.Simulation experiments showed that,compared to PHD-SLAM 1.0 and PHD-SLAM 2.0,SUPHD-SLAM effectively improved the estimation accuracy of the pose and map features of the robot while ensuring computational efficiency.
NETL
NSTL
万方数据
