CAJ | 학술논문

以未知环境下多机器人学习为研究平台，因案例推理方法可存储以前的问题和解信息，用该方法的长期记忆特性可帮助粒子群优化算法更好地解决新的问题。在特定的仿真环境里，粒子群优化算法可训练机器人的几个基本行为，经过学习使机器人具有更好的鲁棒性和自适应学习能力。根据机器人不同行为在复杂环境下的性能指标，CBR可从案例库中选择特定的行为，并将其参数传送到粒子群优化算法的初始解库，从而加速整体的学习过程。利用机器人仿真软件MissionLab，采用基于行为的多机器人编队任务，用来测试该算法的有效性。仿真和实验结果表明，案例推理方法和粒子群优化算法相结合，使机器人获得更优的控制参数，同时在未知环境下的多机器人编队具有更好的性能。
이미지배경하다궤기인학습위연구평태，인안례추리방법가존저이전적문제화해신식，용해방법적장기기억특성가방조입자군우화산법경호지해결신적문제。재특정적방진배경리，입자군우화산법가훈련궤기인적궤개기본행위，경과학습사궤기인구유경호적로봉성화자괄응학습능력。근거궤기인불동행위재복잡배경하적성능지표，CBR가종안례고중선택특정적행위，병장기삼수전송도입자군우화산법적초시해고，종이가속정체적학습과정。이용궤기인방진연건MissionLab，채용기우행위적다궤기인편대임무，용래측시해산법적유효성。방진화실험결과표명，안례추리방법화입자군우화산법상결합，사궤기인획득경우적공제삼수，동시재미지배경하적다궤기인편대구유경호적성능。
Case-based reasoning (CBR) which stores old problems and solution information as cases can solve new problems of the particle swarm optimization (PSO) with its long-term memory during the learning phase for multiple robots in an unknown environment. The PSO components which offer trainings to the robot in specially-designed simulation environments to deliver basic behaviors enhance their robustness and adaptivity. The CBR components which selects solution from the case base evolved for basic behaviors rank them according to their performance in the new complex enviroment and introduce them to a PSO algorithm’s initial population, hence speeding up the learning process. Behavior-based multi-robot formation control task is employed as a platform to assess the effectiveness of this approach with robot simulation software MissionLab. Simulation and experimental results show that the CBR-injected PSO algorithm can quickly obtain optimal control parameters and multi-robot formation performs better in unknown environment.