Research On Motion Planning Of Mobile Robots Informed By Different Environment Models

In recent years,mobile robots have been widely applied in various scenarios,such as family service,logistics,epidemic prevention,and disaster relief.As a key technol-ogy for mobile robots to achieve autonomous operation,motion planning has always been a popular research topic in the field of robotics.The basis of motion planning is an environmental model presented in a certain map form.The currently widely used oc-cupancy grid map only records the information of environmental obstacles and is diffi-cult to efficiently describe various environmental structural characteristics.At the same time,the environment model based on a single grid map fails to provide the topological information of free space,which makes it difficult to fully serve the motion planning task and ensure the efficiency and safety of motion planning in different scenarios.How to select an efficient map form for environmental modeling according to the characteris-tics of environmental structure and motion planning requirements,and design a motion planning approach that takes into account both computational efficiency,motion effi-ciency,smoothness,and safety is still a challenging problem.In view of the above challenges,motion planning problems in different map forms are fully investigated in the dissertation.Firstly,appropriate map forms are selected for environment modeling according to the characteristics of environmental structure and motion planning requirements.And then,safe and efficient motion planning ap-proaches are carefully designed for different map forms.Finally,the proposed motion planning approaches are comprehensively and quantitatively evaluated.The work of the dissertation mainly includes the following four aspects:1)Environment modeling and motion planning based on line segment features.For indoor structured and semi-structured environments,a globally consistent and in-cremental line segment-based map building approach is proposed,which is employed to efficiently model the environment through a compact line segment map.In order to address the problem that the incremental line segment-based map building approach is difficult to update the global map after the previous poses are optimized,a tree-like data structure is designed to record the correspondence between the original line seg-ments and the corresponding global line segment.Furthermore,a new correlation-based accuracy evaluation metric is proposed to enrich the quality evaluation system of line segment maps.In order to realize the autonomous navigation of mobile robots based on line segment maps,a motion planning approach based on a line segment and pose map is proposed.With the help of pose indices,the area where the robot is located in the line segment map is quickly retrieved,and the collision detection process is accel-erated by the setting of adjacent regions.Comparative experiments on three public data sets and one self-recorded data set demonstrate the superior performance of the pro-posed line segment-based map building approach in terms of efficiency and accuracy.And autonomous navigation in a large-scale hallway environment with a perfect loop is realized to validate the effectiveness of the proposed motion planning approach.2)Motion planning with motion primitive pruning and spare-banded structure.To broaden the application scenarios of motion planning approaches,the dissertation fur-ther extends the map form to grid map and proposes an efficient three-layer motion planning framework named E~3Mo P,which consists of global path planning,local path optimization,and velocity planning.As for global planning,a new state lattice-based path planning approach is proposed,in which a heuristic guided motion primitive prun-ing strategy is designed to improve the computational efficiency of graph search.As for local path planning,a soft-constrained local path optimization approach is proposed,wherein the sparse-banded system structure of the underlying optimization problem is fully exploited,and the forward elimination and back substitution algorithm combined with the Levenberg–Marquardt algorithm is employed to efficiently solve the problem.As for velocity planning,a numerical integration-based trajectory planning algorithm is utilized to generate the optimal velocity profile along the given smooth path.Compara-tive simulations and experiments in complex indoor environments demonstrate the su-perior performance of E~3Mo P in terms of computational efficiency,motion efficiency,smoothness,and flexibility.3)Motion planning with grid-based generalized Voronoi diagrams.To fully em-ploy the topological structure information of free space to further improve the com-putational efficiency and safety of motion planning,the dissertation extends the map form from the single grid map to a grid-based generalized Voronoi diagram and pro-poses a safe and efficient motion planning named G~2Mo P.As for global planning,a novel Voronoi corridor-based state lattice planning approach is proposed,in which the search space is reduced from the whole map to the adjacent region of the shortest path in the Voronoi diagram to further improve the computational efficiency of path planning.As for local path planning,an efficient quadratic programming-based path smoothing approach is proposed,wherein the deviation from the original safe reference path is considered in the optimization objective to implicitly push the path from obstacles,and the problem that the optimized path obtained by the hard-constrained approach is usu-ally close to obstacles is addressed.Comparative simulation and experimental results in a complex maze simulation scenario and the large-scale outdoor environment demon-strate that the proposed motion planning approach has the advantages of high compu-tational efficiency and better safety.4)Mobile robot motion planning benchmark.Finally,a mobile robot motion plan-ning benchmark called MRPB is proposed to comprehensively evaluate the performance of mobile robot motion planning approaches.The benchmark consists of simulation scenarios and evaluation metrics.As for simulation scenarios,various challenging scenarios including large-scale shopping malls,narrow passages,dynamic crowds are designed based on the popular Gazebo simulation platform.As for evaluation met-rics,computational efficiency,motion efficiency,safety,and smoothness are employed to quantitatively evaluate the performance of motion planning approaches of mobile robots.Furthermore,the benchmark is employed to evaluate the proposed motion plan-ning approaches in the dissertation to validate the effectiveness and applicability.Fi-nally,the proposed benchmark is open-source to the robotics community to contribute a uniform and fair evaluation system for motion planning approaches.