Monte Carlo Localization for autonomous auxiliary transport vehicles used in coal mine

Autonomous vehicles for auxiliary transport become a technology trend of building a smart coal mine. To implement localization, which is one of the important technical components for autonomous vehicles in coal mine, Monte Carlo Localization(MCL) for the vehicle operating in the coal laneway is concentrated on. Firstly，scene analysis is finished. Considering both process characteristics and driving safety, it is analyzed that there are different requirements for pose estimations of the vehicle about the longitudinal and lateral directions of the laneway. In detail, the longitudinal position along the laneway axis should be acquired when necessary, but the lateral position relative to the center line of the laneway and the angular orientation should be calculated in real time. As a result, the limitation of feature-based localization applied to the vehicle in laneways is represented, and availability of the lateral laser range finder is discussed further. Through segmenting the globe map of the coal mine to a set of the typical scene based local maps, the strategy, that accurate localization of the vehicle is always maintained within one of the typical scene based local maps, is put forward. Secondly，the implementation process of MCL is represented. Through analyzing the optional steering mechanism of the auxiliary transport vehicles in coal mine, the control update is implemented with the velocity motion model of the vehicle. To eliminate the lack of smoothness for the measurement probability distribution in pose state, the likelihood fields for the laser range finder are chosen to implement the measurement update, in which the method of variable uncertainty parameter for the likelihood fields is put forward to avoid algorithm failure. Lastly，the simulation verification of strategy and algorithm of localization is achieved. The simulation of MCL for global localization in a typical scene is finished. Belief distributions are computed by extracting a density from particles through a Gaussian approximation, and the Gaussian means are regarded as the pose estimations. As a comparison, on the condition of the same real trajectory of the vehicle as MCL simulation, the simulation of feature-based Extended Kalman Filter(EKF) localization for position tracking is finished. Simulation results show that in MCL the estimations of both the lateral position and the angular orientation fit always the true states of the vehicle, and that the posterior probability distribution of the longitudinal position changes from a multimodal distribution to a unimodal one, so global localization can shift to position tracking. In addition, in EKF localization, which is not suitable for global localization, the accuracy of the lateral position estimation is closely related to the measurement error of the bearing of features. And considering applications in coal mine laneways, the performance of lateral positioning with MCL based on laser range finder is better than that with EKF localization based on a feature measurement. Both the global localization and the position tracking of the vehicle used in coal mine are resolved by MCL in a local map, and the characteristic real-time pose estimation in the laneway is achieved.