火星漫游巡视态势主动感知方法研究

In the current Mars exploration mission, the analysis of key situation, such as Mars landform and environment is usually carried out by ground staff, such manual operation mode poses severe restriction to the flexibility of rover exploration mission and the reachability of scientific objectives, mostly due to the communication delay between Mars and earth. Therefore, future Mars rover is favored to have the ability to perceive on-site situations of itself and its ambient environment. To this end, this project considers the rover roaming patrol mission as the application background, and studies the active perception methods of Mars landform, dynamic environment and rover pose information. The concrete research contents are listed as follows: first we study the method of on-line Mars landform description using modern set theory, image local feature and global saliency evaluation algorithm; after that, the landform description method will be employed to develop the active perception scheme of Mars static landform and dynamic environment under reinforcement learning structure; finally, we combine the Simultaneous Localization and Mapping (SLAM) technique with active landform recognition and re-visit scheme to deal with the problem of inadequate localization accuracy and flexibility in rover self-localization. This project aims at building a complete active situational perception method for Mars roaming patrol mission under the predictable software/hardware development environment for a Mars rover, achievements from which are expected to provide schematic references and theoretical supports for our country’s future Mars exploration mission.

在目前的火星探索任务中，对火星的地貌环境等关键态势的理解是通过地面工作人员来完成的，这种人工态势分析方式因较长的火地通讯延迟严重制约了巡视器的灵活性与获取科学目标的能力，这就要求在未来的火星漫游巡视任务中，巡视器必须具备主动态势感知技术来自主的分析自身及周边态势。本项目以火星巡视器漫游巡视任务为应用背景，针对火星地貌、动态环境及巡视器位姿状态这三类态势的主动感知方法展开研究。具体内容包括：利用现代集合理论、图像局部特征描述算子与图像全局显著性方法实现火星地貌的在线描述；在此基础上，利用增强学习框架设计地貌与动态环境的主动感知方法；并建立基于主动观测三维地貌特征的（SLAM）方法，解决星表漫游巡视过程中因路标特征纹理稀疏造成的定位精度差、任务灵活程度低的问题。本项目旨在未来可期的巡视器软硬件环境下，建立一套完整的火星漫游巡视任务态势主动感知方法，为我国火星表面探索任务提供方案借鉴和理论支撑。

本项目以火星巡视器漫游巡视任务为研究背景，研究了火星地貌、动态环境及巡视器位姿状态这三类态势信息的主动感知方法。具体内容包括：利用现代集合理论、图像局部特征描述算子与图像全局显著性方法实现行星地貌的实时描述；在此基础上，利用增强学习框架设计了地貌与动态环境的自主感知方法；建立可基于主动观测三维地貌特征的自主定位与构图（SLAM）方法，旨在解决星表漫游巡视过程中由视觉特征纹理稀疏造成的匹配精度差、任务灵活度低的问题。项目在上海宇航系统工程研究所的配合下，完成了静态地貌主动感知方法、巡视器主动SLAM方法的算法性能测试，测试结果表明，本研究所提出的算法能够基本适应火星表面的稀疏地表环境，地貌感知的有效率达到了90%；主动SLAM自主定位精度达到了2%R。本项目形成了一套完整的火星漫游巡视任务态势自主感知方法，相关成果可为我国未来的火星表面探索任务提供方案借鉴和理论支撑。