SwiftVLN for SatNav — Map Memory

Related Repositories

• SwiftVLN: training and evaluation code for these checkpoints.
• SatNav: satellite-image navigation environments, datasets, and evaluation tools.

This checkpoint is designed for SwiftVLN on SatNav, the continuous-state vision-and-language navigation benchmark over satellite imagery. It replaces historical RGB observations with rendered global and local explored-map memory.

Model

• Starting checkpoint: Qwen/Qwen2.5-VL-3B-Instruct
• Training data: SatNav-v0.1 offline expert trajectories
• Training: 1 epoch, full-parameter fine-tuning, learning rate 2e-5
• Context: 32 RGB frames plus global and local explored maps
• Prediction horizon: 4 actions
• Map memory: 1,000 m global view, 400 m local view, rendered at 448 px
• Map processing: 20-pixel mask dilation and average pooling with stride 2

Repository name

• swiftvln-satnav: SwiftVLN trained and evaluated on SatNav
• 3b: Qwen2.5-VL 3B backbone
• 1ep: trained for one epoch
• f32s4: uses a 32-frame window and predicts four actions
• overlap0: uses non-overlapping training windows
• map-g1000-l400-r448-d20-s2: encodes the map extent, rendering, mask, and pooling settings
• noembed: uses no embedding enhancement

Reference Results

Results reported in the SatNav paper, memory-design ablation, row Map memory.

SR, SPL, and OS are percentages. Changes in SR and OS are percentage points relative to the SwiftVLN reference model.

Usage

Use this checkpoint with the SwiftVLN evaluation guide. Keep the full repository name unchanged because SwiftVLN derives the evaluation configuration from it.

This checkpoint is a long-term-memory ablation in the SwiftVLN SatNav Ablation Model Zoo.