面向户外机器人开放导航的鲁棒逻辑传感与局部路径规划方法

During the navigation process of Unmanned Ground Vehicles (UGV), the sensors system prone to emergency malfunction, if the sensors system contain no assurance for robustness, the path planning mission failure may happen. Especially in the open environment, there are many anomalies in traffic circumstances, for instance, a ramp, a curve, it leads to be more difficult for local path planning, and it also severely effects on the active safe for the UGV navigation. In this proposal, we firstly propose a cognitive fault diagnosis method based on model space for inertial navigation subsystem, and localization subsystem is diagnosed by combining the self-learning particle filter and fuzzy logic to achieve unknown fault diagnosis in integrated navigation system of UGV. Secondly, we design a mission-driven optimal allocation model for robust logical sensing. Meanwhile, adaptive preference weight is adopted for building the fitness function. Then we devise a fast heterogeneous culture hybrid algorithm to solve the optimal allocation model which not only improves the robustness of sensors system, but also achieves collaborative management for the sensors system. Finally, the reference path generation method is designed for the anomalous traffic circumstances, meanwhile, a semi-supervised learning algorithm is utilized to identify the unascertainable characteristics of environment and obstacles, on this basis, we propose a local path planning method based on an unascertained support vector machine, and then the unexpected obstacles are regarded as a disturbance, which improve the performance of adaptation and instantaneity for local path planning. This proposal will form systematic theory and methods for robust logical sensing and local path planning. Moreover, the application of its new idea will promote the sustainable development of the intelligent navigation technology of UGV effectively.

在户外机器人导航过程中，传感系统易出现突发异常，若不具备容错能力，将导致路径规划失败。特别在开放环境下，存在匝道、弯道及突发障碍等异常路况，使得局部路径规划变得十分困难，严重影响了导航过程的主动安全。本项目拟研究基于模型空间的惯导子系统认知异常诊断方法，结合自学习粒子滤波与模糊逻辑诊断定位子系统，实现组合导航系统的未知异常诊断；其次，构建面向任务驱动的鲁棒逻辑传感最优分配模型，采用自适应偏好权值加权多目标，研究快速异质文化混合算法求解该模型，提升传感系统鲁棒性，并实现协同管理；最后，针对匝道、弯道等异常路况，利用几何规则生成参考路径，采用半监督学习算法，标识环境与障碍物特征点的不确定信息，研究未确知支持向量机的局部路径规划方法，将突发障碍物视为特征标识的扰动，提高户外机器人的路况适应性与规避实时性。形成较为系统的逻辑传感与路径规划理论与方法，其成果应用将促进户外机器人导航技术的可持续发展。

户外机器人导航过程中，传感系统的突发异常，直接影响其主动安全。特别在开放环境下，户外机器人路径规划与轨迹生成技术，应具备学习能力与强适应性。首先，本项目针对户外机器人不完备系统的航迹推算传感器的软故障检测与诊断问题，提出了粒子滤波器和模糊逻辑相融合的诊断框架。该框架能对所检测的未建模故障中软故障类型进行准确诊断。针对不完备空间混合系统故障诊断，提出了基于自学习采样粒子滤波器(SLSPF)的交互诊断方法。在高维状态空间下，SLSPF不但保证粒子滤波器的诊断精度，且提高了计算效率。其次，面向机器人导航任务的传感器配置方案并非唯一，研究了基于改进离散粒子群优化的传感器系统鲁棒最优配置方法，提升了户外机器人传感器系统的鲁棒性能，能使系统缓慢降级。针对户外导航过程中，车载摄像机获取交通标志牌图像，常产生旋转、比例缩放、平移或噪声问题，提出了filter-filter框架的仿生视觉不变属性特征提取方法，该方法在提升交通标志牌识别的稳健性与识别率方面，取得显著效果。而后，将驾驶行为决策要素抽象为一系列的行驶场景与行为动作状态机集合，探索了双层扩展有限状态机的导航行为决策模型，实现了驾驶行为决策估计，满足开放导航的主动安全性能要求。最后，针对陡坡的弯道匝道、高架隧道中积水路面、突发障碍物，提出了基于超限学习机的局部路径规划方法。使该方法具有学习能力与强适应性，能处理非特定场景中信息，保障规划的安全性与可靠性。在此基础上，研究了基于Quintic Bezier曲线的户外机器人轨迹生成方法，构建了路径长度与轨迹坐标间状态映射模型，降低了跟踪控制器设计难度。同时，针对户外机器人的机械惯性与横纵向参量耦合干扰，提出了基于自适应滚动预瞄窗的非时间参考轨迹跟踪控制方法。并利用了旋转映射技术，消除了奇异点的控制器品质影响，增强了轨迹跟踪控制的鲁棒性与稳定性。形成了系统的路径规划与决策控制的理论方法，其成果应用促进了户外机器人导航技术的可持续发展。

内容获取失败，请点击重试