ELSS - Landing Site Filtering (Potsdam): leaderboard
ELSS (Hua et al., 2026) tests multimodal models as the visual safety filter of a UAV emergency-landing pipeline: shown a remote-sensing image patch of a candidate landing site, the model answers Safe or Unsafe. F1 in percent (higher is better) against human suitability labels for 100 candidate sites proposed by semantic segmentation on 0.05 m airborne imagery of Potsdam.
Metric: F1 (%; the model judges each candidate landing site in a remote-sensing image patch Safe or Unsafe, scored against human suitability labels of 100 semantic-segmentation recommendations; Potsdam airborne imagery, 57 suitable sites). Source: arxiv.org. Saturation forecast: Around February 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Seed 2.0 Pro | 80 | #209 |
| 2 | Gemini 3.1 Pro (Preview) | 74 | #91 |
| 3 | Grok 4.20 | 72.7 | #296 |
| 4 | Claude Opus 4.6 | 66.7 | #100 |
| 5 | GPT-5.4 | 65.3 | #121 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=elss-landing-site-filtering-potsdam · How It Works · Data refreshed daily, snapshot 2026-11-12.