Evolving Plastic Neural Controllers stabilized by Homeostatic Mechanisms for Adaptation to a Perturbation
Evolving Plastic Neural Controllers stabilized by Homeostatic Mechanisms for Adaptation to a Perturbation
复制标题
通过稳态机制稳定进化塑性神经控制器以适应扰动
DOI：
10.7551/mitpress/1429.003.0015
复制
发表时间：
2004
期刊：
影响因子：
--
通讯作者：
Takashi Ikegami
中科院分区：
文献类型：
--
作者：
Jordan Pollack;M. Bedau;P. Husbands;Richard A. Watson;Takashi Ikegami
本文介绍了我们正在进行的工作，包括不断发展的生物启发的塑料自主机器人的神经控制器提交各种内部和外部的扰动：传输中断，滑动，腿损失等，我们提出了一个经典的神经元模型，使用自适应突触和扩展两个生物启发的稳态机制。我们进行了比较研究的两个自我平衡机制的进化的神经网络控制的单腿机器人，在轨道上滑动，这是面对外部扰动的影响。机器人必须达到操作员给定的所需速度目标。进化的神经控制器进行长期模拟测试，统计分析其稳定性和自适应扰动。最后，我们进行行为测试，以验证我们的结果与正弦输入控制的机器人，而发生扰动。结果表明，稳态机制提高了这些控制器的进化性，稳定性和自适应性。
This paper introduces our ongoing work consisting of evolving bio-inspired plastic neural controllers for autonomous robots submitted to various internal and external perturbations: transmission breaking, slippage, leg loss, etc. We propose a classical neuronal model using adaptive synapses and extended with two bio-inspired homeostatic mechanisms. We perform a comparative study of the impact of the two homeostatic mechanisms on the evolvability of a neural network controlling a single-legged robot that slides on a rail and that is confronted to an external perturbation. The robot has to achieve a required speed goal given by an operator. Evolved neural controllers are tested on long-term simulations to statistically analyse their stability and adaptivity to the perturbation. Finally, we perform behavioral tests to verify our results on the robot controlled with a sinusoidal input while a perturbation occurs. Results show that homeostatic mechanisms increase evolvability, stability and adaptivity of those controllers.