Human motion tracking and forecasting has posed multiple challenges due to its high dimensional interactions with the physical world. The uncertain nature of human robot interaction environment stresses on the need to develop prognostic methods for determining responses to ambivalent scenarios in the environment. However, recent growth in depth camera technologies has enhanced performance of such sequence prediction tasks. A growing requirement for co-robotic applications that place robots as partners with humans play a role in cooperative and tightly interactive tasks. Smart machines need to enhance social intelligence and the capacity to make stable, safe and consistent decisions to effectively operate in unconstrained crowded scenes. Autonomous robots should be modelled to circumvent obstacles and obey common sense rules by understanding the personal space of neighbors. Intelligent learning techniques are being exploited to build socially smart, safe and efficient systems to predict human motion in various situations. We propose three deep learning approaches namely-pooling technique, hierarchical RNN and attention mechanism to learn general human social movements, collision avoidance and predict their future paths. The results obtained from the different approaches are compared and analyzed.