StreamVLN for SatNav — Base initialization
This checkpoint is designed for use with SatNav, the continuous-state vision-and-language navigation platform for satellite maps. The SatNav StreamVLN integration uses it to predict actions from a language instruction and a streaming history of RGB observations.
Model
- Starting checkpoint:
lmms-lab/LLaVA-Video-7B-Qwen2 - Vision tower:
google/siglip-so400m-patch14-384 - Training data: SatNav-v0.1 offline trajectories
- Training: 1 epoch, full-parameter fine-tuning, learning rate
2e-5 - Context: 32 frames, 8 history samples, 4 future actions
Repository name
streamvln-satnav: StreamVLN adapted to SatNavscratch: base initialization from the LLaVA-Video checkpoint1ep: trained for one epochf32h8s4: uses 32 frames, eight history samples, and four future actionslr2e-5: uses a learning rate of2e-5
Reference Results
Results reported in the SatNav paper, baseline comparison, row StreamVLN.
SR, SPL, and OS are percentages. NE is measured in meters. Steps is the average number of executed environment actions, including turns and stop.
| Split | SR (%) | SPL (%) | OS (%) | NE (m) | Steps |
|---|---|---|---|---|---|
Test Seen (val_seen) | 64.3 | 63.7 | 71.7 | 53.24 | 51.58 |
Test Unseen (val_unseen) | 52.2 | 51.8 | 61.0 | 84.99 | 57.80 |
Usage
Use this checkpoint with the SatNav StreamVLN integration. It is also listed in the SatNav Baseline Model Zoo.
License
The checkpoint is released under CC BY-NC-SA 4.0. StreamVLN, SatNav, model components, and datasets remain subject to their respective terms.