This paper presents a robust global localization method using Artificial Neural Network (ANN) to learn sonar sensor patterns associated to points in a specified area. Given a set of unseen sonar sensor readings, the ANN is capable of predicting the corresponding point in the map accurately even with the presence of small random noises. This technique can also be extended into the dynamic environment by simply cascading two ANN and incorporating a suitable filtering algorithm (FA) for preprocessing data purposes. Thereafter, after filtering out the corrupted components based on the information disseminate from the FA module, a FeedForward Network (FFN) is used to make the prediction after training with sufficient filtered epochs.