基于新的图优化方法的拓扑地图创建

Topological Map Creation Based on New Graph Optimization Method

在机器人同时定位与地图构建(simultaneous localization and mapping,SLAM)过程中,图优化方法如捆绑约束(Bundle Adjustment,BA)通过对位姿误差的分布进行迭代优化同步提升机器人位姿和地图的准确性,但优化的过程需要较大的计算量。与度量地图相比,拓扑地图因其简洁的环境表示更加适用于大范围环境的自主导航。论文针对拓扑地图在线创建过程中的闭环检测问题提出一种新的图优化方法,该方法只采用里程计航迹推算信息,引入机械臂逆向运动学模型减小位姿跟踪和地图创建误差,并使用模拟退火算法实现变步长快速迭代。实验结果表明该方法具有如下优点:采用间断性运作模式,只是在拓扑节点位置进行图优化操作,适用于计算能力较弱的机器人建图应用;只利用节点间的相互位置关系完成闭环检测,如在此基础上增加场景匹配信息将极大提高地图创建的准确性。

In the process of simultaneous localization and mapping(SLAM)of the robot,graph optimization methods such as bundle adjustment(BA)optimize the distribution of the pose errors and simultaneously improve the accuracy of the robot’s pose and map,but the optimization process requires a large amount of calculation. Compared with metric maps,topological maps are more suitable for autonomous navigation in large-scale environments because of their concise environmental representation. This paper proposes a new map optimization method for the closed-loop detection problem in the online creation of topological maps. This method only uses the odometer track estimation information,and introduces the inverse kinematics model of the robotic arm to reduce the error of pose tracking and map creation,and uses simulated annealing algorithm to realize fast iteration with variable step size. Experimental results show that this method has the following advantages,it adopts a discontinuous operation mode,and only performs graph optimization operations at the topological node positions,which is suitable for robot mapping applications with weak computing power,only uses the mutual positional relationship between nodes to complete closed-loop detection,on this basis,adding scene matching information will greatly improve the accuracy of map creation.