混沌神经动力学行为在多自由度机器人上的应用研究

Study on Application of Chaotic Dynamic Behavior in Multi Degree of Freedom Robot

摘要:

把高维的混沌神经动力学行为应用到多但有限自由度的机器人以实现适应性控制是一个困难的问题;借鉴大量神经元控制少数肌肉的生物事实,一种简单的神经编码方法被用来使高维的神经网络模式转化成了低维的运动参数;虽然只在神经网络中嵌入了3种简单的姿势动作,但是在混沌神经动力学行为出现时,机器人手臂呈现出复杂的组合运动;利用这一点,提出了一个简单的控制算法用来解决病态问题(不一定有解或者确定的解无法保证的问题);实装实验进一步表明,尽管只有粗略甚至不确定的光源信息,利用提出的控制算法,机器人手臂可以成功的寻找到光源。

Abstract:

It is quite difficult that chaotic neural dynamics in higher dimension is applied to robots with multiple (but finite) degree of freedom (DOF). Inspired by the fact that fewer muscles are controlled by a larger number of neurons, a simple coding method enables higher dimensional neural pattern to be translated into lower dimensional motion parameters. Although only three typical gestures are embedded into the neural network, the robot shows complex motions when chaotic neural dynamics emerges. By means of this point, a simple control algorithm was proposed to solve an ill-posed problem, which means that the existence of solution cannot be guaranteed or certain solution cannot be guaranteed. Furthermore, experiments on hardware implementation indicate that the arm robot can find light source with rough and uncertain information successfully using the algorithm.