Reproduce “ScoutVLA: UAV-Centric Active Perception via a Dual-Expert VLA Model for Open-World Embodied Question Answering” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Wenhao Lu, Zhengqiu Zhu, Xiaofeng Wang, Xiaoran Zhang, Yatai Ji, Yong Zhao, Yue Hu, Yingzhen Nie and 2 more
ScoutVLA enables Unmanned Aerial Vehicles to conduct question-conditioned fine-grained active perception for embodied question answering in open-world settings, demonstrating significant improvements with a 10.48x higher average strict success rate and 7.72x higher average QA correctness compared to existing approaches. The National University of Defense Technology and GigaAI developed this system.
Key insights
- Fine-grained active perception, beyond macroscopic navigation, is essential for acquiring complete visual evidence required for accurate answers in open-world EQA scenarios.
- A dual-expert VLA model, inspired by natural biological systems, can effectively translate complex linguistic queries into precise, continuous 3D spatial adjustments for optimal viewpoint alignment.
- Careful decoupling of continuous action control and discrete language reasoning during training, via mechanisms like knowledge insulation, is crucial to maintain both robust flight control and multimodal question answering abilities.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.