EchoNav: Evidential BEV Completion from Historical Observations for Visual Navigation
We present a framework that combines temporal aggregation and occupancy completion in an evidential bird's-eye-view (BEV) representation predicted from short RGB sequences.
ProofPaper ↗
Key points
- Visual navigation in cluttered environments requires robots to recover previously observed structure and infer obstacles and free space that remain occluded.
- A decoder adapts frozen VGGT-Omega features and combines historical residual proposals using cellwise weights and a learned no-update option.
- The evidential model treats observed free space separately and predicts occupancy with associated confidence across the remaining supported region.
- We train the predictor on simulated RGB sequences paired with complete and visibility-masked occupancy annotations.
Sources (1)
- [1]EchoNav: Evidential BEV Completion from Historical Observations for Visual NavigationarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 04:31 AM
We present a framework that combines temporal aggregation and occupancy completion in an evidential bird's-eye-view (BEV) representation predicted from short RGB sequences.
Visual navigation in cluttered environments requires robots to recover previously observed structure and infer obstacles and free space that remain occluded.
Extractive summary: sentences quoted from the sources.