机器学习方法在无人机室内导航中的应用与实现

Application and Implement of Machine Learning Methods in Indoor Navigation for Drones

移动节点的准确定位对于无人机的环境感知、路径规划及控制系统的开发至关重要。对决策树、SVM和贝叶斯等三种机器学习方法在使用锚节点的室内空间进行定位准确率对比研究,最终选择了具有高泛化准确率的贝叶斯方法作为定位方法。结合改进后的A*算法在室内空间对无人机进行自动路径规划,并借助激光传感器感知环境中未知障碍,不断更新室内地图并指导无人机进行避障,无人机导航实验验证了机器学习方法用于室内空间定位的可行性。

Precise localization of mobile node is crucial to the development of environ perception,path planning,and control systems of drones.Three machine learning methods,namely decision tree,support vector machine(SVM)and Bayesian methods,are comparatively studied on their localization accuracy in a space where anchor nodes was employed.Then the Bayesian method was chosen to locate the drone with its high generalization accuracy and the modified A*algorithm was used to navigate the drone to self path planning and the laser sensor was employed to sense the unknown obstacles.The map was then updated so that the drone could avoid the obstacles.The feasibility of the Bayesian method for indoor localization was verified by the experiments of indoor navigation for drones.