Researches on System Stabilization based on Learning Process of Multiple Agents.

为了克服多智能体同时学习困境带来的不稳定局面的困难，我们提出了几种控制多智能体对环境变化和信息流不敏感的方法。RASP是一种根据环境变化程度自动控制强化学习步长参数的方法。我们还研究了一种控制探索/利用比的方法来稳定学习过程。

In order to overcome difficulties of unstable situation caused by dilemma of simul-taneous learning of multiple agents, we proposed several methods to control insensitiveness to environmental changes and information flow for the multiple agents. RASP is a method to control stepsize parameter of reinforcement learning auto-matically according to the degree of changes of environment. We also investigate a method to control Exploration/Exploitation ratio to stabilize learning processes.