EI、Scopus 收录
中文核心期刊

基于小脑脉冲神经网络的柔性机械臂运动控制

Motion Control of Flexible Robotic Arms Based on Cerebellar Spiking Neural Network

  • 摘要: 柔性机械臂由于自身材料的柔软特性,极容易受到环境中的不确定性干扰,从而发生意外形变,影响控制精度。针对该情况,本文借鉴人体中小脑对运动的调控和对环境的适应性,搭建了小脑脉冲神经网络模型,用于对柔性机械臂在环境干扰下的运动行为进行纠正控制。首先,基于分段常曲率方法建立了一个多自由度柔性机械臂模型,它具有移动关节和旋转关节,可以实现伸缩和弯曲的运动行为;并采用顺序二次规划算法近似计算得到机械臂的逆运动学模型,从而求解与期望轨迹对应的期望关节参数。然后,借鉴小脑皮层的神经系统结构与自适应功能,对颗粒层与浦肯野细胞层之间的突触可塑性进行建模,完整构建了小脑脉冲神经网络模型。最后,本文研究了环境干扰下柔性机械臂完成圆形轨迹和“8”字形轨迹的运动效果,发现与不使用小脑模型的直接开环控制运动结果相比,柔性机械臂末端执行器在小脑脉冲神经网络控制下的轨迹误差分别降低了95%和96%,验证了小脑脉冲神经网络模型对于控制柔性机械臂对抗不确定性干扰的有效性。相较于传统的控制方法,该方法更具有生物可解释性,为柔性机械臂在不确定性扰动下的控制提供了一种类脑智能方法。

     

    Abstract: Due to the soft material properties of flexible robotic arms, they are highly susceptible to environmental uncertainties, leading to unexpected deformations that affect control accuracy. To address this issue, this paper introduces a cerebellar spiking neural network model inspired by the human cerebellum's regulation of movement and adaptability to the environment, aiming to correct the motion behavior of flexible robotic arms under environmental interference. First, a multi-degree-of-freedom flexible robotic arm model is established using a segmented constant curvature method. This model features both movable and rotational joints, enabling stretching and bending motions. The inverse kinematics model of the robotic arm is then approximated using a sequential quadratic programming algorithm to compute the joint parameters corresponding to the desired trajectory. Next, drawing inspiration from the cerebellar cortex's neural system structure and adaptive function, the synaptic plasticity between the granule layer and Purkinje cell layer is modeled to fully construct the cerebellar spiking neural network model. Finally, the paper investigates the motion performance of the flexible robotic arm in completing circular and "figure-eight" trajectories under environmental interference. It is found that, compared to open-loop control without the cerebellar model, the trajectory error of the robotic arm's end-effector under the control of cerebellar spiking neural network is reduced by 95% and 96%, respectively. This demonstrates the effectiveness of the cerebellar spiking neural network model in enhancing the flexible robotic arm's resistance to uncertainty interference. Compared to traditional control methods, this approach offers greater biological interpretability and provides a brain-inspired intelligent method for controlling flexible robotic arms under uncertain disturbances.

     

/

返回文章
返回