Autonomous underwater vehicles (AUVs) are attractive tools for surveying earth science and for oceanography. However, many problems must first be solved. In order to realize useful and practical robots, underwater vehicles should take their action by judging the changing condition from their own sensors and actuators, and should make their behavior adaptable to the working environments. We have investigated the application of brain-inspired technologies, such as neural networks (NNs) and self-organizing map (SOM), into AUVs. In order to obtain the I/O relationships, such as dynamics and controller of AUVs, it is suitable to use NNs, which are trained by a supervised learning algorithm. An unsupervised learning algorithm that automatically recognizes the environment surrounding a robot is needed for a decision-making system. The objective of this research is to make a new self-organizing decision-making system for AUVs using the modular network SOM (mnSOM) proposed by Tokunaga et al. We have proposed a controller system for the AUV using mnSOM, and we have introduced the Forward Model Map, which expresses the relationship of control force and states variables of robots. In this paper, the adaptability of the proposed controller system is investigated through the simulations.