Density-Adaptive Model Based on Motif Matrix for Multi-Agent Trajectory Prediction

Multi-agent trajectory prediction is essential in autonomous driving, risk avoidance, and traffic flow control. However, the heterogeneous traffic density on interactions, which caused by physical laws, social norms and so on, is often overlooked in existing methods. When the density varies, the number of agents involved in interactions and the corresponding interaction probability change dynami-cally. To tackle this issue, we propose a new method, called Density-Adaptive Model based on Motif Matrix for Multi-Agent Trajectory Prediction (DAMM), to gain insights into multi-agent systems. Here we leverage the motif matrix to represent dynamic connectivity in a higher-order pattern, and distill the interaction information from the perspectives of the spatial and the temporal dimensions. Specifically, in spatial dimension, we utilize multi-scale feature fusion to adaptively select the optimal range of neighbors participating in interactions for each time slot. In temporal dimension, we extract the temporal interaction features and adapt a pyramidal pooling layer to generate the interaction probability for each agent. Experimental results demonstrate that our approach surpasses state-of-the-art methods on autonomous driving dataset.

1. Introduction

Multi-agent trajectory prediction is becoming increasingly attractive not only in academia but also in industry [18], particularly for automatic system [57], safety planning [63] and traffic flow control [28]. Existing methods [12], [19], [24], [34], [43], [57] often neglect the influence of varying traffic density on interactions between a target agent and its neighbors. They often treat all vehicles (agents) in the same scene equally, leading to the lack of adaptability. More exactly, it lacks of multi-scale neighbor selection and dynamic interaction analysis. Specifically, they do not adaptively choose participating neighbors and assume a constant interaction probability between the target agent and its neighbors. Figure 1.

The impact of heterogeneous traffic density in High-Density (1a) and Low-Density (1b) scenes. Here the star represents the target agent, the circles depict the selection ranges of neighbors and the numbers in the flags indicate agent ID. This highlights the significant influence of agent density on their interaction behaviors. Importantly, the range and interaction probability vary as the density changes across different time steps and among distinct agents.