SingularTrajectory is a universal trajectory prediction framework that uses diffusion models to address the performance gap across five trajectory prediction tasks: deterministic, stochastic, domain adaptation, momentary observation, and few-shot. The core of the framework is to unify various human dynamics representations across these tasks by projecting all motion patterns into a singular space. This space is constructed using singular value decomposition (SVD) to extract principal motion components from training data. An adaptive anchor is then introduced to correct prototype paths based on a traversability map, ensuring they are placed in appropriate locations. A diffusion-based predictor further refines these paths through a cascaded denoising process, enhancing the accuracy of trajectory predictions. The framework is tested on five public benchmarks, demonstrating significant improvements over existing models in predicting pedestrian motion dynamics across different scenarios. The model's ability to handle diverse input modalities and trajectory lengths makes it effective for various trajectory prediction tasks. The results show that SingularTrajectory achieves state-of-the-art performance, highlighting its effectiveness in estimating general human movement dynamics. The code is publicly available for further research and development.SingularTrajectory is a universal trajectory prediction framework that uses diffusion models to address the performance gap across five trajectory prediction tasks: deterministic, stochastic, domain adaptation, momentary observation, and few-shot. The core of the framework is to unify various human dynamics representations across these tasks by projecting all motion patterns into a singular space. This space is constructed using singular value decomposition (SVD) to extract principal motion components from training data. An adaptive anchor is then introduced to correct prototype paths based on a traversability map, ensuring they are placed in appropriate locations. A diffusion-based predictor further refines these paths through a cascaded denoising process, enhancing the accuracy of trajectory predictions. The framework is tested on five public benchmarks, demonstrating significant improvements over existing models in predicting pedestrian motion dynamics across different scenarios. The model's ability to handle diverse input modalities and trajectory lengths makes it effective for various trajectory prediction tasks. The results show that SingularTrajectory achieves state-of-the-art performance, highlighting its effectiveness in estimating general human movement dynamics. The code is publicly available for further research and development.