Curve-Localizability-Based Active Localization Research For Mobile Robots In Outdoor Environments

Applications of autonomous mobile robots basically focused on industry,medicine and exploration fields in recent years,and working areas of robots have been expanding to large-scale outdoor environments from initial structured indoor environments.Increase of environment scale,change of sensor observation data amount and diversity of working tasks bring larger challenges and higher demands to robot localization and navigation,meanwhile,adaptable active localization strategy is required because of localization requirements and robot autonomy needs in different working scenes.In terms of the existing localization demand for mobile robots in complex outdoor environments,a curve-localizability-based active localization method was proposed in this paper.Mobile robots can obtain better localization paths based on the observation model of sensor and 3D map model of the environment,and the efficiency of global localization algorithm was improved by extracting and analyzing environment features.The main research contents contains the following three aspects:1.Considering that the outdoor environment may have features like large scale,complex structure and uneven ground,a curve-localizability-index based on Octo Map and 3D Lidar observation model was proposed and proved to be efficient and rational by theoretical analysis in this paper.Besides,in order to obtain map features for requirements of curve-localizability-index and the following calculations,this paper introduced an improved-multiple-grid-map algorithm to realize the storage and calling of different kinds of feature information.2.Mobile robots need to autonomously select the path that assists particle convergence for active localization,this paper proposed a curve-localizability-based path planning strategy.Obtained by setting the constraint space and objective function of the planning algorithm,where curve-localizability is the main constraint,the path helps improve the convergence speed and stability in different application environments of the active localization algorithm.3.Another challenge that needs to be considered when using 3D lidar sensors for global localization is that the increasing data will influence the real-time performance of the algorithm.In order to reduce the amount of calculation while retaining the advantage of rich observation information as much as possible,this paper introduced a particle filtering algorithm based on the terrain of the environment.Height and normal information of the robot motion curve is obtained by the terrain analysis of the 3D pointcloud map,and is then used to constrain the state of particle initialization,for the purpose of improving the environmental adaptability of localization algorithm and reducing the amount of redundant calculation.Meanwhile,to avoid that the particles converge to a wrong pose in highly similar area,this paper used a pointcloud-matching-based strategy to check the robot pose obtained by global localization,and started relocation if the matching index is too low.In conclusion,this thesis proposed a curve-localizability-based active localization solution for automobile robots in outdoor environment,relevant algorithms was tested in different environments on both simulation and actual platforms.Results of experiments showed that the method has higher global localization efficiency and more stable active localization effect on the same conditions.