This work proposes a novel facial expression recognition approach to contribute to better human-machine interactions. To do that, edge features in facial expression images are combined with a recurrent neural network (RNN) to classify different facial expressions. Robust edge features are first obtained by using Local Directional Strength Pattern (LDSP) and applied with RNN. This LDSP-RNN approach achieves superior recognition performance than other conventional approaches on a randomly distributed training and testing datasets obtained from a public dataset. The proposed approach should be useful for various practical applications such as a robot analyzing and understanding different human emotions from facial expressions based on robotic vision.