OpenFly for SatNav — Navigation-pretrained initialization
This checkpoint is designed for use with SatNav, the continuous-state vision-and-language navigation platform for satellite maps. The SatNav OpenFly integration uses it to predict the next action from a language instruction, three RGB observations, and recent action history.
Model
- Starting checkpoint:
IPEC-COMMUNITY/openfly-agent-7b - Training data: SatNav-v0.1 offline trajectories
- Training: 1 epoch, full-parameter fine-tuning, learning rate
2e-5 - Context: 3 RGB frames and up to 16 previous actions
- Action format:
stop,forward,left, andright
Repository name
openfly-satnav: OpenFly adapted to SatNavcontinue: navigation-pretrained initialization from the OpenFly Agent checkpoint1ep: trained for one epochactcompact: predicts compact textual action labelssample: the following fields describe trajectory-step samplinghk7: keeps the first seven action stepsfs3: keeps every third step in a consecutiveforwardrun; turning steps are keptstopx2: repeats eachstopsample twicestopw0: labels only the terminal step asstoptail5: keeps the final five steps beforestopstoph1: uses one history variant for eachstopsamplehist16: includes up to 16 previous actions in the promptlr2e-5: uses a learning rate of2e-5
Reference Results
Results reported in the SatNav paper, baseline comparison, row OpenFly (navigation-pretrained).
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) | 21.1 | 21.0 | 38.1 | 163.07 | 57.85 |
Test Unseen (val_unseen) | 17.1 | 16.9 | 34.9 | 196.90 | 65.56 |
Usage
Use this checkpoint with the SatNav OpenFly integration. It is also listed in the SatNav Baseline Model Zoo.
License
The checkpoint is released under MIT. OpenFly, OpenVLA, underlying model components, SatNav, and datasets remain subject to their respective terms.