基于神经网络的水下机器人力控策略研究

Force Control Strategy and Test Research of Underwater Dexterous Hand

水下机器人抓取物体时,物体与指尖存在力控制问题,但是由于动力学模型、被抓取物体位置和刚度的不确定性,采用传统阻抗控制方法不具有鲁棒性,因此对基于位置的神经网络阻抗控制方法进行了研究,构建了基于位置的神经网络阻抗控制器,采用三层前向反馈神经网络构建补偿器结构,基于BP算法和Delta学习规则,得到了反向传播的更新规则;该神经网络控制系统具有很强的自适应性,可以很好地完成机器手抓取物体的任务;在水下机器人单手指上分别对软性材料(泡沫)和硬性材料(木块)施加5N恒定力进行反复的实验,结果表明,该方法具有较好的补偿和控制效果,为水下机器人的准确抓握和合理操作奠定基础。

Underwater robot grasping object, the object and fingertip force control problems, but due to the dynamic model, grab objects location and stiffness of uncertainty, the traditional impedance control method is not robust, so the impedance control based on the location of the neural network method are studied, impedance controller is constructed based on the location of the neural network with three layers for- ward feedback neural network to build the compensator structure, based on the BP algorithm and the Delta learning rule, get the update rules of back propagation. The neural network control system has strong adaptability, can very good finish machine hand grasping object task. Un- derwater robot single finger respectively for soft material （foam） and hard materials （wood） on 5N repeated experiments, the constant power of results show that the method has a good compensation and control effect, for underwater robot accurately grasp and reasonable operation lay the foundation.