Mind the Noise: Sensitivity of Transformer-based Interaction-Aware Trajectory Prediction Models to Noisy Data

This paper analyzes the sensitivity of Transformer-based interaction-aware trajectory prediction models to noisy data, highlighting the degradation in prediction accuracy due to real-world perception uncertainties and localization errors. The study reveals that accuracy can decrease by a factor of 1.3 under small noise levels and up to 3.9 under high noise conditions, emphasizing the necessity for more realistic training datasets and effective noise mitigation strategies. This is crucial for practitioners in autonomous vehicle development, as it underscores the importance of robust data handling in enhancing model reliability.