Research map / Vision-language and multimodal / Robot action and navigation
AIR-VLA: Vision-Language-Action Systems for Aerial Manipulation
Jianli Sun, Bin Tian, Qiyao Zhang, Chengxiang Li, Song Zihan, Zhiyong Cui, Yisheng Lv, Yonglin Tian
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use large-scale pre-trained VLA (pi0.5, pi0) instead of ACT or Diffusion Policy | aerial manipulation VLA · AIR-VLA overall | weighted total score: 13.9 -> 42.0 (pi0.5) | Improved |
| use pi0-FAST (distilled, fewer steps) instead of pi0 (ablation) | pi0 · AIR-VLA overall | weighted total score: 34.5 -> 18.8 | Hurt |
| use pi0.5 instead of pi0 | pi0 · AIR-VLA overall | weighted total score: 34.5 -> 42.0 | Improved |
| inject Gaussian noise into UAV state (disturbance) (ablation) | pi0.5 · AIR-VLA | weighted total score: 42.0 -> 41.0 | No clear effect |
| remove fixed third-person view camera (ablation) | pi0.5 · AIR-VLA | weighted total score: 42.0 -> 34.5 | Hurt |
| remove fixed third-person view camera (ablation) | pi0.5 · AIR-VLA | task progression: 10.6 -> 5.1 | Hurt |
| add dense visual distractors and complex semantic instructions to tasks | pi0 and pi0.5 · AIR-VLA Object & Spatial and Semantic suites | weighted total score: pi0 38.1 (Base) -> 32.4 (Object & Spatial) | Hurt |
| use VLM (Qwen3-VL) for high-level planning instead of VLA execution on long-horizon tasks | VLMs · AIR-VLA Long-Horizon | planning performance | Improved |
| use implicit instead of explicit instructions (ablation) | VLM planners · AIR-VLA VLM planning | total score | No clear effect |
| add visual distractor interference to scenes (ablation) | VLM planners · AIR-VLA VLM planning | object grounding | Hurt |
Closest papers by meaning
- Towards Realistic UAV Vision-Language Navigation: Platform, Benchmark, and Methodology (ICLR 2025 · similarity 0.88)
- UAV-Flow Colosseo: A Real-World Benchmark for Flying-on-a-Word UAV Imitation Learning (NeurIPS 2025 · similarity 0.88)
- UP-VLA: A Unified Understanding and Prediction Model for Embodied Agent (ICML 2025 · similarity 0.88)
- CoT-VLA: Visual Chain-of-Thought Reasoning for Vision-Language-Action Models (CVPR 2025 · similarity 0.88)
- VLA-REPLICA: A Low-Cost, Reproducible Benchmark for Real-World Evaluation of Vision-Language-Action Models (NeurIPS 2026 · similarity 0.88)
- MoManipVLA: Transferring Vision-language-action Models for General Mobile Manipulation (CVPR 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
