Using deep learning techniques for solving AI planning problems specified through graph transformations

Deep learning (DL) is a branch of machine learning that uses deep neural networks (DNNs) to extract knowledge from raw data. DL techniques have been used successfully in many intelligence domains, such as general approximation, computer vision, pattern recognition, and many more. Planning problems with small search space can be solved by exhaustive exploration of the search space, whereas the big search space of some problems exposes the search space explosion due to computational limitations. This subject motivates us to propose an approach using DL techniques for solving such planning problems. The proposed approach tries to learn the knowledge about the application order of actions, before solving the given (main) planning problem. Actually, it reduces the size of the given planning problem such that it can be solved by exhaustive exploration of the search space. After solving the reduced problem successfully, a DNN is learned from the explored search space. The proposed approach then employs the learned DNN to solve the given planning problem. The proposed approach deals with the planning problems specified through graph transformations language because of its superiorities compared to planning domain definition languages. The main contribution of the proposed approach is that it uses DL techniques, for the first time, to solve planning problems specified through graph transformations. Based on experimental results, the proposed approach outperforms state-of-the-art techniques in terms of execution speed, accuracy, and generating short-length plans with the exploration of lower states.