RNN-based default logic for route planning in urban environments.

Urban route planning is important for smart cities and intelligent transportation systems. Most of the exiting methods of urban route planning are either graph based search algorithms or optimization methods, however they show many weaknesses in real-world applications due to their high computational complexities and needs for global information of urban environments. To solve this problem, we propose a novel Recurrent Neural Network (RNN) based default logic for route planning. The proposed method is composed of three parts. The first part is the default theory of route planning, forcing no loops in the generated routes. The second part is the RNN based default reasoning to recommend default rules. The RNN outputs the probability distribution of the defaults used next, so the proposed method is a probabilistic method. The third part is the update algorithm of map models to improve the accuracy of default reasoning in dynamic environments. Training the proposed RNN is simple because no statistic computation is required for training. The time complexity of the proposed method during testing is just O(ρ2), where ρ is the length of the optimal route for two given nodes. To evaluate the proposed method, we build a new map model named BJ simulating the complex urban environments of the Beijing city, China. Extensive experiments on the BJ map model in both static and dynamic environments show the effectiveness and accuracy of the proposed method.