Creation of Developmental Object Recognition Mechanism and Motion Generation for Robot using Neural Network Model

在本研究中，我们基于机器人的目标操作任务，利用运动生成实现了目标识别特征和工具的自主提取。利用递归神经网络作为机器人的身体模型，学习机器人的运动和环境变化。在工具使用实验中，机器人能够根据T型和I型工具的经验识别未经训练的L形状的工具，以产生对目标对象的拉动运动。

In this research, we realized autonomous extraction of object recognition features and tool using motion generation based on object manipulation task of a robot. A recurrent neural network was utilized as the robot’s body model to learn the robot’s motions and the environmental change. During tool using experiments, the robot was capable of identifying the untrained L-shaped tool based on experiences with T- and I-shaped tools, to generate pulling motions for a target object.