基于灵长目动物视觉认知原理的旋翼无人机自主规划方法研究

The rotary-wing unmanned aerial vehicle (UAV), which exhibits excellent maneuverability in three-dimensional space, has attracted extensive attention in recent years. However, its potential application has been limited by the existing path planning technologies for which usually need accurate environment modeling and vehicle modeling or local approximation of three-dimensional environment.Therefore, the efficiency and reliability of the existing path planning methods can not satisfy the requirement. The proposal mainly deals with this problem so as to break the above limitations and primarily studied how to extract, store and retrieve the mapping relationship from the planning context to the corresponding decision. Based on the hierarchical projected mechanism and empirical intelligence of the cognitive principles of the primate vision, a methodology for solving the UAV path planning problem called Hierarchical Projected Regression is derived. First, the linear planning model is adopted in the proposed approach to get initial samples. Then, the hierarchical projected mechanism is simulated for describing the complex mapping relationship between the high dimensional environment information and the decisions. Finally, the autonomous planning framework of the rotary-wing UAV is established based on the blocks of incremental-learning, decision-retrieving and online-updating, referring to the empirical intelligence. Consequently, the efficiency and reliability for the UAV planning in real three-dimensional environment is improved. This proposal break a new path for three-dimensional path planning of rotary-wing UAV, lay a foundation for UAV flying fully autonomously, and has great theoretical significance as well as application value.

近年来,具有良好空间机动性能的旋翼无人机得到了广泛关注，然而它的应用一直受到现有路径规划技术的制约。主要因为现有方法通常需要建立精确的环境模型和无人机模型，或者需要对三维运动环境局部逼近，存在实时性不好、可靠性差等不足。本课题跳出现有方法的制约，将研究重点放在由路径规划环境到决策量之间映射关系的抽取、存储与检索上。基于灵长目动物视觉认知的分层投影机制和经验智能，提出一种基于分层投影回归的三维路径规划方法。该方法首先采用线性规划模型创建初始样本，然后通过模拟分层投影机制来描述高维环境信息与决策量之间的复杂映射关系，最后借鉴经验智能，基于路径规划知识库的增量学习、在线检索与在线更新构建旋翼无人机自主规划框架，从而提高无人机在真实三维环境中路径规划的实时性与可靠性。该课题为三维自主规划理论提供了新视角，为无人机实现全自主飞行奠定了基础，具有较高的理论意义和应用价值。

自主规划能力指的是机器人以优化的形式自主计算或选择合理路径的能力，反映了机器人顶层智能的水平。本课题以旋翼无人机为研究对象，主要研究如何从以往的规划经验中提取有益的知识，并以合理的结构进行表示和存储，为以后的规划提供帮助，使无人机具有一定的自主规划能力。按照项目计划，课题组开展了以下几个方面的研究：.（1）研究了路径规划的分层投影回归机制。本项目借鉴灵长目动物视觉认知原理，使用分层投影机制模拟旋翼无人机由高维路径规划环境信息向低维决策映射过程。本课题提出的算法将需要解决的映射关系用一种类似于视觉皮层形态特征的有序层状数据结构进行描述，从而使抽取映射关系的过程由粗到精地逐步进行。由于分层数据结构建立在低维特征空间回归结果的基础上，具有较小的计算量，在确保路径规划结果正确可靠的前提下提高了规划的实时性。.通过模拟视觉皮层形态学分层特征，本课题提出的相对状态树方法将不同属性按样本值域空间进行非均匀离散化，得到可以代表整个值域空间的子状态列表。该方法使用有限的属性离散化的子状态，精简了子状态列表的数目，提高了计算效率。.（2）研究了路径规划知识库的构建方法。为了避免机器人在线重复计算相同或相似的规划问题，本课题提出了自主规划框架，包含了增量学习、在线检索与在线更新三个部分。本课题分别采用增量分层判别回归算法和k-D树算法构建了自主规划框架，并进行了对比。与单一的路径规划算法相比，本课题建立的自主规划框架有两大优势。一是建立了由状态到行为的映射关系，更加接近人在规划时的智能，避免了对机器人本体和外在环境的建模，提高了规划效率。二是该方法具有在线增量学习机制，知识库会随着学习过程的进行变得越来越丰富和准确。.（3）开展自主规划实验研究。为了验证本课题提出的自主规划方法，设计了半物理仿真平台，开发了旋翼无人机自主规划仿真软件，验证了本课题所提方法的可行性。