A Joint Prediction Method of Multi-Agent to Reduce Collision Rate

TL;DR

This research builds upon the SIMPL baseline to explore methods for generating scene-consistent trajectories and significantly reduces the collision rate of joint trajectories within the scene.

Abstract

Predicting future motions of road participants is an important task for driving autonomously. Most existing models excel at predicting the marginal trajectory of a single agent, but predicting joint trajectories for multiple agents that are consistent within a scene remains a challenge. Previous research has often focused on marginal predictions, but the importance of joint predictions has become increasingly apparent. Joint prediction aims to generate trajectories that are consistent across the entire scene. Our research builds upon the SIMPL baseline to explore methods for generating scene-consistent trajectories. We tested our algorithm on the Argoverse 2 dataset, and experimental results demonstrate that our approach can generate scene-consistent trajectories. Compared to the SIMPL baseline, our method significantly reduces the collision rate of joint trajectories within the scene.

Figures (3)

• Figure 1: Visualization of Predicted Trajectories for Agents in Scenarios. Our method can generate joint predictions for all agents in the scene simultaneously. The ego vehicle is shown in red, while other vehicles are displayed in gray. The predicted trajectories are visualized using gradient colors.
• Figure 2: Overall structure Our model builds upon the SIMPL framework. Semantic instance features are encoded using simple encoders, relative positional embeddings, and a symmetric feature transformer. Subsequently, our proposed multimodal scene-consistent decoder and scene-consistent loss are used to train the model, generating scene-consistent results.
• Figure 3: Predicted Trajectories of the Two Methods The left images shows the scene-consistent joint prediction generated using our method, while the right images displays the joint prediction produced by the straight marginal method based on the SIMPL baseline. The red circles highlight collisions that occur in the trajectories generated by the straight marginal method. In contrast, our method avoids these collisions in both scenarios, achieving better scene consistency.