仿生CPG控制优化及在机器鱼上的实验验证

As an active research on robotic control, bio-inspired CPG-based control is one of the key technologies which are able to achieve multi-mode locomotion of multi-joint robots and high environmental adaptability. However, there are several open issues in bio-inspired CPG control, such as CPG feedback control, parameter tuning, CPG-mechanism-environment unified modeling, which need further research. This proposal is devoted to resolve the problems of fish-like robot CPG control modeling, feedback, parameter tuning, and optimal control through exploring bio-inspired CPG control methods and their applications to free-swimming robotic fish. The main research contents are as follows. Firstly, we will build a hierarchical CPG feedback control model by studying creature feedback mechanism and coupling higher control center with CPG network. Secondly, we will propose reinforcement learning based CPG model parameter tuning method and simplify the proposed parameter tuning algorithm to realize parameter online learning. Thirdly, the proposed methods will be validated on the biomimetic robotic fish. That is, we will build a CPG-coupled three dimensional dynamical model of robotic fish through Kane method so as to realize CPG-mechanism-environment unified modeling. Further, a population-based artificial intelligence technique will be utilized to optimize CPG feedback control. At last, all proposed theories and methods will be verified on the self-developed robotic fish platform. The outcome of this proposal will offer theoretical support to bio-inspired CPG-based control and optimization for robot applications.

仿生CPG控制作为机器人控制领域的研究热点之一，是实现多关节机器人多模态运动和高环境适应性最有希望的关键手段。但CPG反馈控制、参数整定、CPG与机器人机械本体一体化建模与优化控制等难点问题的研究还远远不足。本项目拟通过仿生CPG控制及在机器鱼上的应用研究，解决鱼游CPG控制的建模、反馈、参数整定、优化控制等问题。主要研究内容包括：首先探讨基于生物反馈机制，通过高层控制中枢与CPG网络的融合，建立分层CPG反馈控制模型；其次基于强化学习提出CPG模型参数自整定方法，并运用优化算法实现CPG模型参数的在线学习；再次将所建模型应用到机器鱼上，采用Kane方法建立耦合CPG的三维游动动力学模型，实现CPG-机械本体-环境一体化建模；接着利用群智能优化方法实现机器鱼CPG反馈控制优化；最后在仿生机器鱼平台上对所提理论和方法进行实验验证。本项目研究成果将为推动机器人仿生CPG控制与优化提供理论支撑。

本项目针对机器人控制领域的研究热点之一的仿生CPG控制进行研究，主要研究工作集中在CPG建模、参数优化、动力学建模、分层控制以及仿生机器鱼实验验证等方面。首先，结合机器鱼游动的特点，基于Hopf振荡器建立了改进型仿生CPG模型，由于模型中增加了相位调节因子和反馈因子，实现了仿生CPG输出信号间相位关系自由调整和环境反馈引入，将传感反馈机制引入到CPG中，丰富了CPG模型的输出信号，增加了CPG模型的模态处理能力；其次，基于上述建立的改进型仿生CPG模型，开展了针对多关节仿生机器鱼的转向闭环游动控制实验，结果验证了耦合传感反馈机制的CPG模型有效性，拓展了CPG模型的应用范围，为提高机器人本体的自适应能力提供了参考；再次，利用Kane方法建立耦合CPG的三维游动动力学模型，实现CPG-机械本体-环境一体化建模，接着以游动速度及推进效率为优化目标，采用粒子群优化方法（PSO）对CPG模型参数进行优化，得到最优参数集合，再次基础上进行了仿生机器鱼游动能耗的实验验证，结果表明仿生机器鱼能量消耗与CPG振幅和频率正相关，而CPG的相位滞后对能量消耗的影响较小；最后，构建了一种基于CPG的离散、连续一体化框架的分层CPG反馈控制模型，使运动单元能够根据具体接收的命令实现离散、连续运动的切换；并以一种基于嵌入式视觉的多控制面机器鱼为平台进行了验证性实验，结果表明了所提一体化控制框架的有效性。本项目研究成果将为推动机器人仿生CPG控制与优化提供理论支撑。