基于动物运动神经系统的蛇形机器人控制方法研究

项目名称： 基于动物运动神经系统的蛇形机器人控制方法研究

项目编号： No.60875083

项目类型： 面上项目

立项/批准年度： 2009

项目学科： 金属学与金属工艺

项目作者： 马书根

作者单位： 中国科学院沈阳自动化研究所

项目金额： 32万元

中文摘要： 蛇形机器人缺乏实用性的一个原因是控制系统的不足限制了它的环境适应能力。根据对脊椎动物的运动神经机制的研究，提出一个层次化的控制系统，实现、改进它的各个组成部分。 基于振荡器建立的多模态CPG能生成多种步态。证明它有任意节数稳定性、给出了它的模型参数与输出之间的解析关系。基于神经元建立的反馈式循环抑制CPG有输出幅值统一、模型参数与输出有较好的调整关系等特点。根据动物的多层CPG机制建立的双层反馈式循环抑制CPG能以较小复杂度实现单节CPG输出多自由度信号。以此为基础提出一个三维步态控制方法。 在提出的基于CPG的头部引导运动模式和基于幅值调整的转弯控制方法的基础上，通过反馈神经网络将障碍物信息融合到CPG中实现自主避障。根据生物原理，分别提出一个基于地面摩擦、运动速度的环境适应策略，将二者分别融合到CPG中实现环境自适应。 根据经验学习机制，提出一个具有在线自主优化、学习能力的高层控制器。通过对已有优化结果的学习、记忆，得到环境与最优运动参数的近似关系，实现用预测的方法来实现环境自适应。 本研究为蛇形机器人控制系统的设计提供了新的方法，降低控制系统设计复杂度，提高环境适应能力。

中文关键词： 蛇形机器人；CPG控制器；自主避障；在线自主优化学习；环境自适应

英文摘要： An important reason for the snake-like robots lacking pactical applications is that the deficiency of their control system restricts their environmental adaptability. According to the researches on the vetebrate's neural mechanism of motion control, a hierarchical control system for the snake-like robots is proposed, and the different parts of the control system are implemented and improved based on the biology mechanism. The multi-phase CPG, built on the oscillators, can generate different gaits. Its stability for any number of segments is proved and the analytic relationship between its parameters and the output is derived. The closed-form cyclic-inhibition CPG, built on the neuron models, has features such as uniform ouputs and good relationship between its parameters and the output. According to the animal's hierarchical CPG mechanism, a double-layered closed-form cyclic-inhibition CPG is modeled, which can realise that each segment of this CPG can generate two degree of freedom outputs, with small complexitery. A three-dimensional gait contol method is proposed based on this CPG. Based on the proposed CPG-based head-navigated locomotion mode and the amplitude-modulated method based turning method,a signal feedback neural network is constructed to integrate the obstacle information is integrated into CPG to realise autonomous avoidance of obstacles. According to the biology mechanism, the environment adaptation methods are proposed based on the ground friction and the velocity, respectively. With these methods, the ground friction and the velocity are integrated into the CPG model to realise autonomous environment adaptation, respectively. According to the experiential learning mechanism, an high level controller with online autonomous optimization and learning ability is proposed. By leaning from existing optimization results and memorizing the learning results, the approximate relationship between the environment and its corresponding optimal locomotion parameters are derived and used to predict the optimal locomotion parameters to adapt to the environment effectively and precisely. This research provide a novel method for designing the snake-like robot' control system, reducing the complexity to design system and improve its environment adaptability.

英文关键词： snake-like robot; CPG controller; autonomous obstacle avoidance; online autonomous optimization and learning; environment adaptation