Abstract: Current research on embodied task planning of UAV (unmanned aerial vehicle) doesn't fully use the reasoning abilities of language models, and lacks the deep logical reasoning and complex context understanding capabilities while generating the detailed substep plans, resulting in poor interpretability and demonstrating inadequacy in complex urban environments. For this problem, an aerospace embodied chain-of-causality enhanced UAV embodied vision-language model is proposed to improve UAV task planning capabilities in complex urban environments by simulating the logical chains of causality. Additionally, a method for generating embodied chain-of-causality data is proposed to balance data quality and cost, efficiently producing high-quality UAV aerospace embodied task planning chain-of-causality data at low cost. Extensive experiments are conducted on models with different parameter sizes on the AeroVerse UAV embodied task planning dataset SkyAgent-Plan3k. The results demonstrate that the aerospace embodied chain-of-causality significantly enhances the performance of vision-language models in UAV embodied task planning. On the BLUE metric, AeroVerse-SkyPlan outperforms GPT-4o by 35% 47%, demonstrating that the aerospace embodied chain-of-causality simulating human cognitive reasoning can enhance the task planning capability a UAV in complex environments, while the automated data generation pipeline provides data support for model training.