Global Control Lyapunov Function design based on decision by Q-learning

In nonlinear control theory, multilayer minimum projection method is proposed for Control Lyapunov Function (CLF) design. The method generates a global control Lyapunov function from local control Lyapunov functions. However, the automatic generation method from local functions is not developed. In this paper, we consider the control problem including learning from exploring space. The learning from exploring is defined in discrete space, however the control problem is defined in continuous space. Thus, we need to relate the discrete space to the continuous space. This paper focuses on the CLF design including learning from exploring by Q-learning. Our goal is to develop a CLF design method.