自主机器人在复杂环境如泥地滩涂等场合、尤其在跨介质行动的战场，视觉传感器作为核心传感部件，其光学窗口清晰度对机器人性能和安全影响至重。有效利用复杂多样的污损图像信息并仍能保持持续运动是一种挑战，目前尚无成熟和统一的理论和解决方法。本项目针对机器人视觉受污案例多样和机器人应急策略优化协调问题，采用自省CBR偶然性规划方法，得到具有自我评估、自我更新能力的自动规划器；针对CBR案例特征描述问题，采用隔帧和背景差分图像特征和污物物性提取及图形分形方法，解决CBR规划器案例表征压缩问题；针对受污视场残余信息利用和视觉功能恢复问题，采用仿生双目PTZ拼接优化模型、基于变刚度眼睑-泪器多元仿生优化模型，解决CBR规划器特征解分类问题；针对机器人保持持续运动功能问题，采用多模态感知补偿信息和代价函数优化，得到机器人应急行为策略。本项目研究成果将为自主作业机器人、移动作战平台提供应急自救的理论和技术基础。

For autonomous Robot in complex environment such as beach and muddy terrain, especially the trans-media unmanned vehicle in the battlefield, the clarity of the optical window in vision sensor of robot with the core sensing is vital to the performance and security. It is challenging for us to make use of contaminated image information with complexity and diversity and maintain continuous movement. There is no mature and unified theory and solution at present on the issue. To the problem of diversity of the contaminated and optimization of robot coordination emergency strategy, an automatic contingency planning with self-evaluation and self- renewing is investigated by an introspective CBR (Case based reasoning) method. To the problem of CBR case feature description, the frame difference and background image feature extraction and dirt property extraction and fractal method are employed to get the representation compression of CBR planner. To the problem of utilization of the residual contaminated image information and the visual function recovery, CBR planner classification decompression is obtained, by the bionic binocular PTZ image mosaic optimization model, the variable stiffness of eyelid- lacrimal multiple bionic optimization model. To the problem of robot holding the function of continuous movement, the robot emergency behavior strategy is rendered by using the compensation information from multi-modal perception and optimized cost function. This project will provide the theoretical and technical basis for autonomous robot and military mobile platform in terms of emergency self-rescue.