EgoTraj-Bench: Towards Robust Trajectory Prediction under Ego-View Noisy Observations
Jiayi Liu, Jiaming Zhou, Ke Ye, Kun-Yu Lin, Allan Wang, Junwei Liang
AI summary
Problem
Existing trajectory prediction methods assume clean, bird’s-eye-view observations, failing to account for real-world ego-centric perception artifacts like occlusions, ID switches, and tracking drift, which severely limits deployment robustness.
Approach
We introduce EgoTraj-Bench, a real-world benchmark pairing noisy first-person historical trajectories with clean future ground truth, and propose BiFlow, a dual-stream flow matching model that jointly denoises corrupted histories and predicts future trajectories using an intent-aware EgoAnchor mechanism.
Key results
- EgoTraj-Bench: first real-world benchmark for trajectory prediction under ego-view noise
- BiFlow achieves state-of-the-art performance, reducing minADE and minFDE by 10–15% on average
- Demonstrates significant performance degradation of existing BEV-based models under ego-view noise
- EgoAnchor mechanism effectively stabilizes predictions via intent-aware feature modulation
Why it matters
Provides a critical foundation and robust modeling framework for deploying safe, socially compliant autonomous navigation in dense, human-centric environments.
Abstract
Reliable trajectory prediction from an ego-centric perspective is crucial for robotic navigation in human-centric environments. However, existing methods typically assume noiseless observation histories, failing to account for the percep- tual artifacts inherent in first-person vision, such as occlusions, ID switches, and tracking drift. This discrepancy between training assumptions and deployment reality severely limits model robustness. To bridge this gap, we introduce EgoTraj- Bench, built upon the TBD dataset, which is the first real- world benchmark that aligns noisy, first-person visual histories with clean, bird’s-eye-view future trajectories, enabling robust learning under realistic perceptual constraints. Building on this benchmark, we propose BiFlow, a dual-stream flow matching model that concurrently denoises historical observations and forecasts future motion. To better model agent intent, BiFlow incorporates our EgoAnchor mechanism, which conditions the prediction decoder on distilled historical features via feature modulation. Extensive experiments show that BiFlow achieves state-of-the-art performance, reducing minADE and minFDE by 10–15% on average and demonstrating superior robustness. We anticipate that our benchmark and model will provide a critical foundation for robust real-world ego-centric tra- jectory prediction. The benchmark library is available at: https://github.com/zoeyliu1999/EgoTraj-Bench.