基于元控制的微型无人机环境自适应功能智能自主导航定位研究

Context-adaptive precise autonomous navigation technology has been the key to enabling the autonomy and safety of Unmanned Aircraft Systems in ever more challenging environment. This proposal puts specific emphasis on the context-adaptive integration of the low cost multi-sensors in Micro Aerial Vehicles (MAV), which has been drawing much attention from academics but is still a challenge. To solve the problems, this proposal employs control science, computer science and cognitive neuroscience as foundational tools to attain the best results. A context-adaptive scheme based on meta-control theory is developed to derive reliable measurements. Functional system theory based intelligent information processing mechanism is proposed to ensure high-quality measurements as inputs of the reconfigurable integration system. Meanwhile, an innovative method based on coded filter is introduced to acquire the integrated dead reckoning information with higher accuracy as well as slower frequency. Furthermore, a functional system theory based multi-layer integrated position navigation strategy is developed to adapt to various context and behavior of autonomous MAVs based on factor graph formulation. The research achievements are valuable in terms of the context-adaptive functional intelligent integration architecture, the optimal pre-filter for uncertain noises of multi-sensors, the context determination approach for the object-oriented generic information, the functional system theory based multi-layer integrated navigation strategy with plug and play capability. It is also obvious that the results provide important strategic guidelines into both the theoretical and practical levels for our Unmanned Aircraft Systems and new generation aircraft to implement durably precise autonomous navigation and control tasks under challenging environment.

环境自适应精确自主导航定位技术是保证复杂环境下无人飞行器安全自主稳定飞行的关键。研究过程以微型无人机导航定位系统为对象，针对复杂环境中微型无人机强鲁棒、高精度自主导航定位亟待解决的难点问题，以控制科学、计算机科学、认知神经科学为工具，基于元控制机理研究环境自适应的可用信息获取方法；基于功能系统理论提出可用信息动态加工与整合智能方法；基于Coded滤波获得高精度低更新率的组合推算信息；在此基础上，结合因子图提出环境自适应多层次功能智能自主导航定位信息融合策略。研究成果将在微型无人机环境自适应功能智能自主导航定位架构、不确定非线性多样信息源最优预处理、可用信息获取及应用环境检测、多层次“即插即用”智能信息融合等方面提出新方法和新思路，为我国无人机及新一代战机精确自主导航定位的发展提供理论指导和技术支持。

环境自适应精确自主导航定位技术是保证微型无人机安全自主稳定飞行的关键。针对复杂环境下微型无人机自主导航定位过程中亟待解决的难点问题，本项目开展了基于元控制的微型无人机环境自适应功能智能自主导航定位研究。完成以下主要内容：. 信息源多样性和应用环境多变性增加了多源信息融合的复杂度，是环境自适应导航定位信息融合中限制信息源“即插即用”的根本原因。研究过程中，分析了应用环境转换导致的环境敏感信息源切换特征，提出了基于SceneNet和MobileNet-v2深度学习网络的双层场景辨识及可用信息评估方法；利用智能终端和四旋翼无人机开发了环境自适应功能智能导航平台，完成了开阔/约束/过渡空间内系列飞行试验，结果表明该方法在可用信息评估、场景识别、特征提取、环境分类中效果显著。. 信息源中的不确定性误差是影响环境自适应自主导航定位精度的关键因素。为了提高惯性信息精度，基于改进径向基函数神经网络、时间序列神经网络等数据驱动方法，建立了温度、振动等多物理场影响的惯性信息误差补偿模型；提出基于多时间尺度时空复合模型的不确定信息加工与整合方法，引入非线性互补滤波和自组织误差补偿理论，构建了微型无人机姿态估计主导推算信息模块；设计了基于元控制的改进Q-learning自主航路综合规划算法，并在开发的可变倾角双旋翼气动舵微型无人机、微型无人机航路规划虚拟仿真平台上进行了有效性验证。. 计算能力约束的信息融合实时性是微型无人机安全稳定飞行控制的重要内容。提出环境自适应多层次功能智能导航定位信息融合策略基础上，研究了基于因子图的“即插即用”异构多源信息融合方法，以及交互多模非线性模糊自适应滤波方法，通过索并联机器人等价物理实验平台验证了信息融合算法的正确性。开发了火星星表探测控制虚拟仿真实验平台（转化为虚拟仿真实验教学项目），解决了微型无人机复杂环境下的信息软故障适应与控制问题，提高了多层次功能智能自主导航的环境适应性和鲁棒性，保障微型无人机安全自主稳定飞行。