Research On Autonomous Exploration And Reconstruction In Rough Terrains

Nowadays,mobile robots are widely used in different fields.As one of the most popular tasks in these applications,autonomous exploration and reconstruction has achieved extensive attention.Although in regular structured environments,it has already been widely studied.While it is still challenging to apply it in complex rough terrains.Thus,this work introduced some new approaches to address this problem.Three sub-problems in this task will be detailly studied,which are autonomous exploration strategy,self-localization technique and environment reconstruction algorithm.Firstly,a new exploration strategy,which directly decided the exploration goals based on the point cloud map,has been proposed.It improved the existing frontierbased exploration methods by defining the frontiers with a reachable map and a Kmeans clustering algorithm.For complex rough terrains,it ensured the safety in the exploration procedure,as well as reducing the computation complexity.What's more,a multi-sensor fusion based localization approach is presented.It complementary fused the measurement data from wheeled odometry,visual odometry and inertial sensors with a Kalman filter.To further improve the fusion result in complex working conditions,a fuzzy logic inference system was designed to inference the confidence coefficient of different sensors in real-time.This localization method adaptively adjusted the fusion parameters and reached high accuracy.Finally,we proposed a method to construct a point cloud map with laser-measured data as the robot explores around.A terrain-adaptive density mapping technique is used to balance the demands of small data size and high terrain accuracy by utilizing the local curvatures as the simplification criteria.The adaptive density mapping technique is further integrated within the map merging framework to improve the matching speed and accuracy.The proposed methods are validated on a set of real-world experiments.The experimental results indicate that our proposed autonomous exploration and reconstruction algorithms works well in various challenging conditions and an accurate environment map can be detailly constructed.