Trajectory generation using deep neural network

This paper introduces the state-of-art trajectory generation technique by the Deep Neural Network. This technique is one of the supervised learning methods. A certain trajectory generation technique provides the optimal solution to the neural network. The neural network studies how to generate a trajectory from training data. The first survey verifies the possibility for the Deep Neural Network to be able to create the trajectory, which the Artificial Potential Field method makes under obstacle presence field. Since the output of the neural network is heuristic, it is not optimal and may violate a constraint. Therefore, we inform the method that a trajectory generation method uses the result, which the neural network output, as an initial guess.