Pedestrian behavior exhibits inherent multi-modality, necessitating predictions that balance accuracy and diversity to adapt effectively to various complex scenarios. However, conventional noise addition in diffusion models is often aimless and unguided, leading to redundant noise reduction steps and the generation of uncontrollable samples. To address these issues, we propose a Prior Condition-Guided Diffusion Model (CGD-TraP) for multi-modal pedestrian trajectory prediction. Instead of directly adding Gaussian noise to trajectories at each timestep during the forward process, our approach leverages internal intention and external interaction to guide noise estimation. Specifically, we design two specialized modules to extract and aggregate intention and interaction features. These features are then adaptively fused through a spatial-temporal fusion based on selective state space, which estimates a controllable noisy trajectory distribution. By optimizing the noise addition process in a more controlled and efficient manner, our method ensures that the denoising process is effectively guided, resulting in predictions that are both accurate and diverse. Extensive experiments on the ETH-UCY, SDD, and NBA datasets demonstrate that CGD-TraP surpasses state-of-the-art diffusion-based and other generative methods, achieving superior efficiency, accuracy, and diversity.

A reliable prediction of traffic participants’ trajectories is challenging for automated driving. We attempt to integrate human trajectory’s temporal and spatial reciprocal consistency into the prediction network structure. Moreover, humans are intention-driven agents, and modelling human intentions will continuously provide more accurate and detailed information for future trajectory prediction. In this paper, we propose a Reciprocal Consistency Prediction Network (RCPNet), a bi-directional prediction network consisting of the forward and the backward with a similar structure. The prior work of human trajectory prediction is mainly a unidirectional process, while RCPNet fully uses the past and future trajectory information at reverse temporal scales. By training the backward part to optimize the whole network parameters based on the reciprocal consistency, we can improve the prediction accuracy of the forward network. In particular, the framework incorporates a destination variational auto-encoder (DVAE) to estimate the target intention of humans and generate multi-modal trajectory prediction. In order to adapt to the offline prediction situation, we refer to the calculation method of correlation between two variables to present “Predictive Confidence” to clarify the reliability of offline predicted trajectories. Experimental results show that the RCPNet improves state-of-the-art performance on the SDD and ETH/UCY benchmarks.