COTTA: Context-Aware Transfer Adaptation for Trajectory Prediction in Autonomous Driving

2026-03-31

AI Summary

This study addresses the performance degradation of trajectory prediction models—originally trained on Western road data—when deployed in new geographic regions such as Korea, where traffic patterns, infrastructure, and driving behaviors differ significantly. Focusing on the QCNet architecture, the work investigates cross-domain transfer strategies by comparing zero-shot transfer, training from scratch, full fine-tuning, and encoder-freezing approaches. The authors propose an efficient adaptation scheme that freezes the pretrained encoder while fine-tuning only the decoder. This strategy substantially improves prediction accuracy without sacrificing training efficiency, reducing trajectory prediction error by over 66% compared to training from scratch. The results validate the transferability of pretrained knowledge to novel geographic domains and establish a practical, efficient paradigm for cross-regional trajectory prediction.

Abstract

Developing robust models to accurately predict the trajectories of surrounding agents is fundamental to autonomous driving safety. However, most public datasets, such as the Waymo Open Motion Dataset and Argoverse, are collected in Western road environments and do not reflect the unique traffic patterns, infrastructure, and driving behaviors of other regions, including South Korea. This domain discrepancy leads to performance degradation when state-of-the-art models trained on Western data are deployed in different geographic contexts. In this work, we investigate the adaptability of Query-Centric Trajectory Prediction (QCNet) when transferred from U.S.-based data to Korean road environments. Using a Korean autonomous driving dataset, we compare four training strategies: zero-shot transfer, training from scratch, full fine-tuning, and encoder freezing. Experimental results demonstrate that leveraging pretrained knowledge significantly improves prediction performance. Specifically, selectively fine-tuning the decoder while freezing the encoder yields the best trade-off between accuracy and training efficiency, reducing prediction error by over 66% compared to training from scratch. This study provides practical insights into effective transfer learning strategies for deploying trajectory prediction models in new geographic domains.

Problem

• trajectory prediction
• domain discrepancy
• autonomous driving
• transfer adaptation
• geographic context

Innovation

• transfer learning
• trajectory prediction
• domain adaptation
• autonomous driving
• context-aware adaptation