Video Submission for ICLR

Paper Title - NavTrust: Benchmarking Trustworthiness for Embodied Navigation

Video File - NavTrust_Video.mp4

Format - MP4
Resolution - 720p
Duration: 10 minutes 30 seconds
File Size: 69 MB

Video Structure:

1.Benchmark Introduction: NavTrust unifies VLN and OGN, stresses agents with RGB, depth, and instruction corruptions, and uses Performance Retention Score to measure robustness.

2. Experiment Analysis: We compare SR, SPL, and PRS across ETPNav, NaVid-7B, WMNav, L3MVN, PSL, and VLFM, finding double-digit drops under depth noise and gains from panoramic and late-fusion designs.

3. Simulation Visualizations: One MP3D/HM3D scenes show drift and stalls in defocus blur , multipath depth noise, and detours from professional instructions, each contrasted with a clean run.

4. Mitigation Techniques: 4 methods shown and the results of it 
 
The video complements the paper submission by visually demonstrating NavTrust test various state-of-the-art models in real-world corruption in simulation.

For more implementation and analysis details, refer to the full manuscript.
