自治水下机器人避碰算法及其仿真研究

为了提高水下机器人(AUV)在复杂海洋环境下的自适应性,文章对传统的避碰方法进行了改进.对AUV的禁入区和潜在碰撞区进行了安全性分析.建立了前视声纳模型以及三维水下规划空间模型.提出了基于强化学习、BP神经网络及人工势场相结合的避碰算法.在AUV与环境的试错交互中,借助于来自成功和失败经验的奖励和惩罚值,不断改进水下机器人的自治能力.设计了AUV深度行为模糊协调器, 对潜深行为和距底高度行为进行协调融合.开发了AUV运动规划与仿真的半实物软件系统,仿真试验证明了算法的合理性与可行性.研究对于提高AUV的自主性和安全性具有重要的应用价值.

A Collision Avoidance Algorithm and Its Simulation for AUV

In order to promote the adaptability of AUV in complex ocean environment, conventional methods of collision avoidance are improved in this paper. The safety of forbidden areas and potential dangerous areas are analysed for AUV. The models of forward looking sonar and 3 dimensional underwater environment are built. The collision avoidance strategy is proposed integrated with Q - learning, BP neural net and artificial potential. The automation ability is enhanced by success and fail experiences through trial error with the environment. The AUV fuzzy coordinator for depth control behavior is designed, and the depth behavior and the height behavior are fused. The semi - physical software system for AUV motion planning and simulation is developed, and the feasibility and reliability are verified through simulation tests.