基于IHDR自主学习框架的无人机3维路径规划

UAV 3D Path Planning Based on IHDR Autonomous-Learning-Framework

提出一种基于自主学习框架的无人机3维路径规划方法.该自主学习框架由知识学习、知识检索和在线更新三部分组成.在该框架中,无人机在线路径规划时首先从过去的规划经验中提取控制量直接用于指导当前机器人的行动,另一方面,如果检索结果对于当前无人机的状态是无效的,可以在线启动常规3维路径规划算法,实时计算机器人的控制量,在控制机器人运动的同时将当前状态下的新决策量添加到知识库中从而对其进行更新.此外,分别采用增量分层判别回归算法(IHDR)和k-D树方法建立了路径规划知识库.其中,IHDR方法通过增量方式,可将以往的路径样本建立为一棵分层树.大量的仿真结果对比表明,在本文提出的框架下,基于IHDR的方法比传统的k-D树方法具有更好的实时性.

An autonomous learning framework for UAV （unmanned aerial vehicle） 3D path planning is proposed. This framework consists of three parts, i.e. knowledge learning, knowledge retrieving and updating online. In this framework, the control value will be retrieved firstly from the existed knowledge when UAV runs online, so as the current action of the robot can be guided by the results. If the decisions retrieved from the knowledge base are invalid for the current UAV states, the custom algorithm for UAV path planning will be launched online and it generates the decisions for UAV＇s movement in real time. In the meanwhile, the knowledge library is updated by adding the new decisions for the current states. Additionally, the knowledge library is constructed by the algorithm of incremental hierarchical discriminant regression （IHDR） and k-D tree, respectively. Among these methods, IHDR can construct a hierarchical tree by using the past path planning samples. By several simulations, IHDR method demonstrates better real time performance than the traditional k-D tree method under the proposed framework.