Risk-Aware Imitation Learning with Environmental Context for AIS
Imitation-learning anomaly detection for AIS augmented with environmental data so the model separates risky behavior from weather-driven detours.
2025.08 –
·Motivation
- Most imitation-learning anomaly detectors rely only on vessel trajectories.
- Without environmental context, hazard-avoidance maneuvers are often misclassified as anomalies.
- Adding environmental context (wind, waves) helps distinguish unsafe actions from safe, weather-driven detours.
·Methodology
- Redefine the imitation-learning state space to include ERA5 environmental variables.
- Train policies on normal trajectories under environmental context.
- Evaluate with hazard-injection scenarios to test decision robustness.
·Contribution (Expected)
- A hazard-injection evaluation pipeline for imitation learning under weather impact.
- A framework showing how enriched state representations improve risk-sensitive anomaly detection.
