EDA: Evolving and Distinct Anchors for Multimodal Motion Prediction

TL;DR

The paper introduces Evolving and Distinct Anchors (EDA), a novel method for multimodal motion prediction in autonomous driving that improves regression capacity and prediction accuracy by evolving and selecting distinct anchors for better matching.

Contribution

The paper proposes EDA, a new paradigm that enables anchors to evolve and be distinct, enhancing multimodal motion prediction performance over existing methods.

Findings

Achieves state-of-the-art results on Waymo dataset.

Reduces Miss Rate by 13.5% relative to baseline.

Improves all evaluation metrics compared to previous methods.

Abstract

Motion prediction is a crucial task in autonomous driving, and one of its major challenges lands in the multimodality of future behaviors. Many successful works have utilized mixture models which require identification of positive mixture components, and correspondingly fall into two main lines: prediction-based and anchor-based matching. The prediction clustering phenomenon in prediction-based matching makes it difficult to pick representative trajectories for downstream tasks, while the anchor-based matching suffers from a limited regression capability. In this paper, we introduce a novel paradigm, named Evolving and Distinct Anchors (EDA), to define the positive and negative components for multimodal motion prediction based on mixture models. We enable anchors to evolve and redistribute themselves under specific scenes for an enlarged regression capacity. Furthermore, we select distinct anchors before matching them with the ground truth, which results in impressive scoring performance. Our approach enhances all metrics compared to the baseline MTR, particularly with a notable relative reduction of 13.5% in Miss Rate, resulting in state-of-the-art performance on the Waymo Open Motion Dataset. Code is available at https://github.com/Longzhong-Lin/EDA.