AMD: Adaptive Momentum and Decoupled Contrastive Learning Framework for Robust Long-Tail Trajectory Prediction

2025-07-02

AI Summary

Addressing the long-tailed distribution challenge in autonomous driving trajectory prediction—where tail-class trajectories are scarce, scenarios are complex and safety-critical, and existing methods rely solely on prediction error while neglecting trajectory pattern diversity and uncertainty—this paper proposes an Adaptive Momentum and Decoupled Contrastive Learning (AM-DCL) framework. Our key contributions are: (1) an enhanced momentum-based contrastive mechanism to strengthen representation learning for rare trajectories; (2) a decoupled contrastive module that explicitly separates semantic context from motion dynamics; and (3) a trajectory-aware data augmentation strategy comprising four complementary variants, coupled with an online iterative clustering-based pseudo-label updating scheme. Evaluated on nuScenes and ETH/UCY benchmarks, AM-DCL achieves state-of-the-art performance for long-tailed trajectory prediction, while simultaneously improving overall accuracy and robustness under distributional shift and occlusion.

Technology Category

• Search and Optimization: Learning to Search
• Machine Learning: Life-Long and Continual Learning
• Multiagent Systems: Adversarial Agents

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Accountability, Transparency, and Ethics for personalization

Abstract

Accurately predicting the future trajectories of traffic agents is essential in autonomous driving. However, due to the inherent imbalance in trajectory distributions, tail data in natural datasets often represents more complex and hazardous scenarios. Existing studies typically rely solely on a base model's prediction error, without considering the diversity and uncertainty of long-tail trajectory patterns. We propose an adaptive momentum and decoupled contrastive learning framework (AMD), which integrates unsupervised and supervised contrastive learning strategies. By leveraging an improved momentum contrast learning (MoCo-DT) and decoupled contrastive learning (DCL) module, our framework enhances the model's ability to recognize rare and complex trajectories. Additionally, we design four types of trajectory random augmentation methods and introduce an online iterative clustering strategy, allowing the model to dynamically update pseudo-labels and better adapt to the distributional shifts in long-tail data. We propose three different criteria to define long-tail trajectories and conduct extensive comparative experiments on the nuScenes and ETH$/$UCY datasets. The results show that AMD not only achieves optimal performance in long-tail trajectory prediction but also demonstrates outstanding overall prediction accuracy.

Problem

• Addressing imbalance in trajectory distributions for autonomous driving
• Enhancing recognition of rare and complex trajectory patterns
• Adapting to distributional shifts in long-tail trajectory data

Innovation

• Adaptive momentum contrast learning (MoCo-DT)
• Decoupled contrastive learning (DCL) module
• Trajectory random augmentation and clustering