The recent advancements in trajectory forecasting for autonomous systems have significantly enhanced the ability to predict pedestrian and vehicle movements, addressing critical challenges in real-world scenarios. A notable trend is the integration of socially-informed models that account for dynamic interactions within the environment, leading to more stable and accurate trajectory predictions. Additionally, the development of memory-enhanced frameworks for urban flow prediction has shown robustness against distribution shifts, ensuring reliable forecasting even in unpredictable spatial-temporal contexts. Pre-training strategies using pseudo-labeled data have also emerged as effective methods for motion forecasting, improving performance across diverse datasets and enhancing cross-dataset generalization. Furthermore, instantaneous trajectory prediction models, designed to handle scenarios with limited observed data, have demonstrated the ability to quickly and accurately predict future trajectories, enhancing the safety and reliability of autonomous vehicles. These innovations collectively push the boundaries of what is possible in trajectory and urban flow prediction, paving the way for more sophisticated and adaptable autonomous systems.