Recent advances in pedestrian trajectory prediction aim to capture the complexity of human behavior in autonomous systems. Prior methods often rely solely on positional data, limiting their ability to model social and environmental interactions. We propose SMSCI, a novel model that integrates recurrent sequence modeling with generative adversarial networks (GANs) to generate realistic trajectory distributions. By incorporating both historical motion and environmental context, SMSCI effectively models social dynamics and scene constraints, achieving higher accuracy and improved collision avoidance. Experimental results confirm the model’s ability to produce socially and physically consistent trajectories, with strong implications for applications in self-driving vehicles and social robotics. The code is available at this link .