ActiveFly-Bench: Aligning Embodied Question Answering with Vision-Language-Action for Aerial Embodied Perception

The paper introduces ActiveFly-Bench, a benchmark for active perception by UAVs that links cyberspace reasoning with physical-world interaction. It decomposes the problem into three hierarchical tasks: Aerial Embodied Question Answering (Air-EQA), Observation Behavior Planning (OBP), and Fine-grained Language-guided UAV Control (FLUC). The datasets contain real-world and simulated outdoor environments. The authors also present ActiveFly, a closed-loop UAV agent combining vision-language reasoning with fine-grained control, and deploy it on a physical UAV platform. According to the abstract, representative VLMs and VLA models still struggle with behavior planning, viewpoint adjustment, and robust task completion.