混合推进仿生机器鱼的优化设计与智能控制

Fishes over long-term natural evolution have been developing excellent morphologies and swimming styles to adapt to their survival environment, which can provide enlightening significance for the advancement of novel propulsive technology and design of the biomimetic underwater vehicles. This proposal mainly focuses on the hybrid biomimetic robotic fish which not only can realize a fast fishlike swim, but also can obtain high maneuverability and stability in slow motions. Specially, an integrated design methodology will be applied, and both bio-inspired mechanism design and intelligent locomotion control approaches will be taken into account so as to enhance the propulsive performance and the intelligence of the hybrid system. The main research contents are listed as follows. First, a hybrid bio-inspired mechanism with low resistance and modular structures will be designed through analyzing and simulating form drag and propulsive modes from some typical fishes. Second, in the framework of reinforcement learning, an Actor-Critic-based attitude controller will be proposed to modulate the locomotion attitude. Third, an intelligent decision center to smartly schedule multimodal locomotion will be constructed to cater to the mission demands. At last, with the multimodal intelligent transformation, the kernelized correlation filters (KCF) algorithm will be applied to accomplish underwater object identification and tracking in complex underwater environments. On this basis, a novel biomimetic propulsive system will be developed to verify the feasibility and applicability of the presented methods. The outcome of this proposal will not only help strengthen the propulsive capability of biomimetic robotic fish and further promote the practical course, but also provide important theoretical and technical support for designing and controlling novel high-performance biomimetic underwater vehicles.

鱼类经过漫长进化形成与生存环境相适应的不同形态结构和推进模式为新型水下推进技术和航行器设计提供了重要参考和生物范例。本项目拟从仿生机体设计和智能游动控制着手，采用一体化设计方法，针对兼具高速游动和低速机动及稳定性能的混合推进机器鱼开展深入研究，以提高混合推进系统的推进性能和智能水平。主要研究内容包括：通过综合多种鱼类的形态结构和推进模式，研究低阻力、模块化的混合仿生推进机构优化设计方法；基于Actor-Critic学习框架，构建多鳍面姿态控制器，实现快速、精准的姿态控制；构建多模态智能决策中心，根据任务特征智能切换运动模态；采用核相关滤波器跟踪算法，完成实时、准确的水下运动目标跟踪；在此基础上，建立新型仿生推进平台，对所提理论和方法的可行性和适用性进行验证。本项目的研究成果不仅有助于提升仿生机器鱼的游动性能，推动其实用化进程，而且为高性能仿生航行器的设计与控制提供重要的理论基础和技术支撑。

鱼类经过漫长进化形成与生存环境相适应的不同形态结构和推进模式，为新型水下自主航行器系统设计和智能控制提供了重要参考。本项目以鲔科鱼类为仿生对象，深入分析了其形态结构和游动方式，优化了其流线外形及仿生机构设计，同时构建了模块化的两侧长鳍机构，研制开发了仿生机器鱼及混合推进型仿生机器鱼2种具有完全自主知识产权的系统样机；基于多鳍面协同，提出了面向仿生机器鱼的航向姿态及深度控制方法，分别实现了仿生机器鱼的航向姿态及定深控制；采用对抗结构化控制思想，构建了面向仿生机器鱼运动优化方法，通过对全局基准控制和局部补偿控制进行多轮对抗性训练，实现了仿生机器鱼的运动优化控制；在此基础上，采用核相关滤波（Kernelized Correlation Filters，KCF）方法构建了仿生机器鱼的主动视觉跟踪系统，并提出了基于深度强化学习的水下目标跟踪控制方法，实现了仿生机器鱼对水下静态及运动目标的跟踪控制。本项目的研究成果不仅可以提升仿生机器鱼的游动性能，增强其水下感知能力及智能水平，而且为高性能水下航行器的设计与控制提供了重要的理论基础和技术支撑。本项目共发表学术论文22篇，其中SCI期刊论文10篇，包括IEEE Transactions论文7篇；研究成果3次获得IEEE CYBER、IEEE RCAR、IEEE ROBIO等国际会议最佳论文奖；申请发明专利4项，其中已授权发明专利1项；登记软件著作权2项。