Evolutionary Creation of Adaptive Neurocontrollers Inspired by the Dynamically Rearranging Function in Biological Nervous Systems

近年来，进化机器人方法在机器人学和人工生命领域引起了极大的关注。与设计人员必须以自上而下的方式构建控制器的传统方法相比，ER方法具有显著的优势，因为它们可以通过实施(例如，机械动力学、机器人的物理尺寸和形状、传感器/电机属性和配置等)自主而有效地构建控制器。考虑到机器人与环境之间的相互作用动力学，在本研究项目中，研究了一种进化生成的自适应神经控制器。为此，引入了神经调节的概念，允许根据当前情况通过阻断和/或激活突触或神经元来进化神经网络，不仅可以调整突触的权重，还可以调整神经网络的结构。这一思想已被应用于本质不稳定的推桩机器人和腿部机器人的控制。已经进行了仿真，结果表明这种方法是非常有前途的。

Recently the Evolutionary Robotics (ER) approach has been attracting a lot of concern in the field of robotics and artificial life. In contrast to the conventional approaches where designers have to construct controllers in a top-down manner, the methods employed in the ER approach have significant advantages since they can autonomously and efficiently construct controllers by taking embodiment (e.g. mechanical dynamics, physical size and shape of robots, sensor/motor properties and disposition etc.) and the interaction dynamics between the robot and its environment into account.In light of these facts, in this research project, an evolutionary creation of adaptive neurocontrollers has been investigated. To this end, the concept of neuromodulation, allowing to evolve neural networks which can adjust not only the synaptic weights but also the structure of the neural network by blocking and/or activating synapses or neurons according to the current situation has been introduced. This idea has been applied to the control of a peg-pushing robot and a legged robot, which is intrinsically unstable. Simulations have been conducted and the results have showed that this approach is remarkably promising.