Original title: Large Language Model-Driven Autonomous UAV Systems: Technical Evolution, Core Architectures, and Critical Challenges
Meijie Zhang, Hao Wang · September 2026
Drones, also known as unmanned aerial vehicles, need to follow complex missions safely. Researchers are now adding large language models, or smart AI text programs, to these flying robots. The researchers surveyed how these systems developed and how their structures work. In these designs, the AI handles understanding natural human speech, processing sensory data, making decisions, planning, and coordinating multiple drones across edge devices and cloud computing. The paper notes that testing these systems requires tracking thinking delays, control timing, power use, hardware, and safety separately. The authors also found major challenges. These include differences between simulations and reality, patchy wireless connections, cyberattacks, privacy leaks, and AI mistakes that cause dangerous actions. Future work focuses on lighter reasoning and formal safety checks.
You are giving spoken directions to a friend driving a car. You tell them 'turn left at the next street.' The friend might mishear and turn into a one-way street. Here, the drone is like that friend, but it also has a language model that can make up entire false directions. The paper proposes a safety system that, like a second person in the car checking the GPS, intercepts those made-up commands before the drone moves.
Drones are being used for deliveries, inspections, and search-and-rescue. If they rely on language models, a misunderstood command could cause a crash or worse. This survey highlights those dangers—hallucinations, cyberattacks, privacy leaks—so engineers and regulators know what to fix before these systems are widely deployed.
As a survey, this paper does not present new experiments or validate any proposed solutions. It only reviews existing work and identifies challenges. The analysis is high-level across many dimensions, so it does not deeply investigate any single approach. The concept of semantically adaptive safety certificates is defined but not tested.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Large Language Model-Driven Autonomous UAV Systems: Technical Evolution, Core Architectures, and Critical Challenges — Meijie Zhang, Hao Wang — Intelligence & Control — doi:10.53941/ic.2026.100011
Abstract licence: CC BY
Explanation generated by Synthesis from the abstract — it is not the original paper.