Control Adjoining Cell Mapping and Reinforcement Learning (CACM-RL) is a promising technique used to implement controllers. However, it needs many resources so that it can be only applied to simple problems. The contribution of this work is to describe MultiScale approach in order to be used together with CACM-RL technique to overcome its limitations. The main challenge is to verify and validate its efficiency in real-time and in resource-limited systems. MultiScale approach is truly useful when different levels of resolution are needed in the state space, regardless of the number of dimensions. In this way, a set of different regions inside the state space where each region has a specific optimal policy (also different resolutions) is defined. The results described in this article show the feasibility to run MultiScale in real time and find the minimum number of policies to solve the optimal control problem in an automatic way. In the considered test cases, a significant reduction in the total number of cells used is achieved when using MultiScale.