Abstract:  According to the AUV software fault repair problem in partially observable environment, a method based on Partially Observable Markov Decision Process（POMDP） model and micro-reboot mechanism for solving failure repair of AUV is presented. AUV software self-repairing POMDP model is established based on the characteristics of AUV hierarchical structure while can design multi-level micro-reboot repair method. With the help of repair strategy which based on the algorithm of Point Based Value Iteration（PBVI）, the system can implement the repair at lower repair cost in the partially observable environment. Results of experiment prove the validity of algorithm and the applicability of the model.

Key words: Autonomos Underwater Vehicle（AUV）, micro-reboot, self-repair, Paritally Observable Markov Decision Process（POMDP）, Point Based Value Iteration（PBVI）

摘要： 针对AUV软件在部分可观环境中的故障修复问题，依据部分可观马尔科夫决策过程理论，提出基于POMDP模型和微重启技术的AUV软件故障修复方法。根据AUV 分层结构特点设计了多层次的微重启修复方法，构建了AUV软件自修复POMDP模型，同时采用基于点的值迭代算法求解生成修复策略使系统在部分可观环境下能够以较低的修复代价执行修复动作。仿真实验验证了算法有效性和模型适用性。

关键词: 智能水下机器人, 微重启技术, 自修复, 部分可观马尔科夫决策过程（POMDP）, 基于点的值迭代（PBVI）