Abstract:Accurately predicting the trajectories of surrounding vehicles is a crucial and complex task in autonomous driving due to the inherent uncertainty in driving behavior. Multimodal trajectory prediction methods have emerged as promising approaches to reduce uncertainty. However, these methods frequently face the “mode collapse” issue, where the generated trajectories are limited to one or a few modes, or the majority of the generated trajectories do not comply with road constraints. To address this problem, we propose a novel multimodal trajectory prediction model named Intention Inference with Lane Graph representation (IILG). This model divides the problem into three subtasks: encoding of traffic agents (i.e., road users) and scenes with interaction considerations, predicting goal set through learning and optimization, and decoding multimodal trajectory using multi-head attention. The Agent-node attention method is implemented to capture complex interactions among the target vehicle, surrounding agents, and scenes. To encompass all potential reasonable intentions, we creatively incorporate the maximum entropy principle into the optimization function in multi-goal selection. This approach obviates the need for complex manual anchor settings and overcomes the limitations of noise sampling that lacks semantic information. Additionally, we introduce an improved attention map suitable for graph-structured inputs to enhance model interpretability. The IILG model demonstrates state-of-the-art performance on the nuScenes dataset, achieving a reduction in the final displacement error by over 8.4% when the number of output trajectories is set to 5. Furthermore, the generated trajectories exhibit enhanced diversity and adhere more closely to road constraints. Ablation analyses further validate the effectiveness of both the Agent-node attention and goal set prediction modules. Our model presents a novel perspective for guiding multimodal trajectory generation and inferring reasonable potential driver intentions, thus enhancing fault tolerance in predictions and aiding in the analysis of potential vehicle conflicts and risks in complex scenarios.

Supplementary Material for GRIN: Generative Relation and Intention Network for Multi-agent Trajectory Prediction

GRIN: Generative Relation and Intention Network for Multi-agent Trajectory Prediction.

Multi-Agent Trajectory Prediction Based on Graph Neural Network

Goal-Guided and Interaction-Aware State Refinement Graph Attention Network for Multi-Agent Trajectory Prediction

Spatio-Temporal-Interaction Graph Neural Networks for Multi-Agent Trajectory Prediction

Conditional Goal-Oriented Trajectory Prediction for Interacting Vehicles

Dynamic-group-aware networks for multi-agent trajectory prediction with relational reasoning

Learning Group Interactions and Semantic Intentions for Multi-Object Trajectory Prediction

A Spatial-Temporal Attentive Network with Spatial Continuity for Trajectory Prediction

GA-STT: Human Trajectory Prediction with Group Aware Spatial-Temporal Transformer.

Multi-granularity scenarios understanding network for trajectory prediction

Intention-Driven Trajectory Prediction for Autonomous Driving

Relation Learning and Aggregate-attention for Multi-person Motion Prediction

Multi-Modal Vehicle Trajectory Prediction by Collaborative Learning of Lane Orientation, Vehicle Interaction, and Intention

Multimodal Vehicle Trajectory Prediction Based on Intention Inference with Lane Graph Representation

INT2: Interactive Trajectory Prediction at Intersections

Q‐EANet: Implicit Social Modeling for Trajectory Prediction Via Experience‐anchored Queries

ProIn: Learning to Predict Trajectory Based on Progressive Interactions for Autonomous Driving

GroupNet: Multiscale Hypergraph Neural Networks for Trajectory Prediction with Relational Reasoning

Intention-Interaction Graph Based Hierarchical Reasoning Networks for Human Trajectory Prediction