Deep Sensor Fusion
Deep Sensor Fusion with Constraint Safety Bounds for High Precision Localization
🛡️ IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2024
Overview
Deep learning-based sensor fusion achieves impressive localization accuracy — but can you trust it? For autonomous driving, a prediction is only useful if it comes with safety guarantees.
We propose a deep sensor fusion framework that embeds constraint safety bounds directly into the learning pipeline. This unique combination of deep learning expressiveness with principled constraint-based reasoning delivers both high-precision localization and formal safety guarantees.
Key Contributions
- 🎯 High precision — achieves accurate pose estimation through deep multi-sensor fusion
- 🛡️ Safety bounds — integrates formal constraint-based safety guarantees into the predictions
- 🔧 End-to-end — safety constraints are embedded directly in the learning pipeline, not bolted on as a post-processing step
- 🚗 Real-world ready — designed for deployment in safety-critical autonomous driving scenarios
Why It Matters
Accuracy without reliability is not enough for autonomous systems. This work shows that deep learning and formal safety guarantees are not mutually exclusive — paving the way for trustworthy autonomous localization.