For a mobile robot, navigation skills that are safe, efficient, and socially compliant in crowded, dynamic environments are essential. This is a particularly challenging problem as it requires the robot to accurately predict pedestrians' movements, analyse developing traffic situations, and plan its own path or trajectory accordingly. Previous approaches still exhibit low accuracy for pedestrian trajectory prediction, and they are prone to generate infeasible trajectories under complex crowded conditions. In this paper, we develop an improved socially conscious model to learn and predict a pedestrian's future trajectory. To generate more efficient and safer trajectories in a changing crowed space, an online path planning algorithm considering pedestrians' predicted movements and the feasibility of the candidate trajectories is proposed. Then, multiple traffic states are defined to guide the robot finding the optimal navigation strategies under changing traffic situations in a crowded area. We have demonstrated the performance of our approach outperforms state-of-the-art approaches with public datasets, in low-density and simulated medium-density crowded scenarios. and then predicted the position and velocity of humans for a finite horizon. Schulz et al. [6] combined pedestrian intention recognition with path prediction, through an interacting multiple model filter in combination with a latent-dynamic conditional random field model. However, most of these research works are limited by independent models that fail to capture the complex interactions between the humans in the crowd. An alternative and recent method utilizes learning techniques to model the joint distribution of future trajectories of interacting agents based on a spatially local interaction model. Trautman et al. [7] proposed an interactive Gaussian process approach, whose kernels were used to model human dynamics, to capture cooperative collision avoidance between humans and a robot. Alahi et al. [8] proposed long-short term memory (LSTM) networks with "social" pooling layers, which learned general human movements and predicted their future trajectories. Recently, we proposed a socially conscious model considering the added features that affect the pedestrian's future trajectory, such as the walking direction of other pedestrians [9]. However, these approaches do not carefully distinguish the effects of a pedestrian's own history trajectory, and that of others, to the pedestrian's future trajectory, and this may hinder the improvement of model accuracy.Developing efficient online path planning algorithms to generate robots' safe and smooth trajectory is a basic problem in robot navigation. Safe trajectory generation is a very active field in mobile robotics and there are many recent contributions. Ravankar [10] and David [11] reviewed the state-of-the-art in smooth trajectory generation with comparisons in terms of kinematic feasibility and safe path generation recently. Additionally, the approaches that are available with robot operating system ...