Research On Accurate Localization Methods For Mobile Manipulator In Complex Industrial Environments

Due to the increasing labor costs and the multi-varieties of product requirements,it has become more and more popular for manufacturing companies to use robots to replace people.At the same time,robotics is getting more attention and support from the Chinese government.Compared with robot arm,mobile manipulator^[1-2]is a new type of robot,which combines the advantages of AGV and robot arm.Also,it effectively avoids the disadvantages that AGV requires guidance facilities and robot arm is not suitable for working in a large scope.Therefore,mobile manipulator is considered as core equipment for loading and unloading tasks and shipping material into or out a warehouse.However,in China,mobile manipulator is still far from mature industrial applications.One of the key reasons is that its localization system cannot meet the requirement of reliability and accuracy.Many effective localization algorithms have been proposed to solve the robot localization problem.However,those algorithms are rarely designed and optimized for industrial applications.Specifically,global localization algorithms are time-consuming and not reliable in a large-scale environment,pose tracking algorithms are not accurate and reliable enough in complex industrial environments.To deal with the challenges posed by practical industrial applications,this paper conducts the following researches and proposes corresponding localization algorithms.Aiming at the time-consuming problem of global localization in large-scale environments,this paper proposes a fast and reliable global localization algorithm called convex hull sampling based branch and bound global localization（CS-Bn BGL）.The convex hull sampling（CS）is used to sparse dense Lidar data and the multi-resolution likelihood fields（MLFs）based branch and bound search（Bn B）is used to speed up global search.Through thousands of global localization in two real industrial environments,it is verified that in an environment that is nearly 3,000 m²,it only takes about 100ms to accomplish a global localization by our algorithm and its localization success rate is more than 97%.Aiming at the difficulties for accuracy pose tracking in harsh manufacturing workshop,this paper proposes a pose tracking algorithm called Bn B aided adaptive Monte Carlo localization（Bn B-AMCL）.Hybrid maps are developed to enhance the localization performance of our algorithm.Moreover,Bn B-AMCL seamlessly combines Bn B search and AMCL and performs two times of iterate closest point algorithm（ICP）for pose refinement,which can handle the large odometry error caused by rough and oily ground and the drastic fluctuations of Lidar data caused by the convex plate of a CNC.Besides,a distance filter improved by the unscented transform is integrated into Bn B-AMCL to mitigate the influence of dynamic obstacles.Our algorithm is evaluated through different experiments including 463times of loading and unloading tasks in a real manufacturing scenario,resulting in an average positioning error of 5 mm/0.111 deg.Aiming at the accumulated localization error in environments that contain symmetrical or featureless areas,this paper develops a new grid map called ambiguity grid map（AGM）to explicitly account for the environmental ambiguity.Besides,a Dynamic Bayes network（DBN）to model the localization problem in ambiguous environments is designed.Based on this DBN,this paper proposes the AGM based AMCL（AGM-AMCL）for reliable localization.AGM-AMCL employs a new motion model referred as to portal motion model which significantly improves the chance of localization failure recovery after a robot moving through an ambiguous area.The effectiveness of AGM and the localization performance of AGM-AMCL is evaluated through simulation and real-word experiments in 3 typical ambiguous environments and the results demonstrate that the localization success rate of AGM-AMCL is2 times higher than that of AMCL.At last,to further avoid the negative impact of complex industry environments on mobile manipulator localization,this paper proposes localization algorithms by using reflector landmarks.By employing the information of reflectors,those localization algorithms contains a global localization algorithm,a pose tracking algorithm and an accuracy localization algorithm.The global localization algorithm and the accuracy localization algorithm is an extension of CS-Bn BGL and ICP.For pose tracking,we integrate the reflector detection process into the observation model to mitigate the influence of false-positive of reflector detection.Through experiments in real-world scenarios,it is demonstrated that those reflector aided localization algorithms has better accuracy and localization success rate than localization algorithms without reflectors.In short,localization methods proposed in this paper allow accurate and reliable robot localization in complex industrial environments.Those localization methods have been successfully applied in different industrial scenarios and created value for manufacturing enterprises.