Llm and nlp-based drone control: A distributed edge computing architecture for Vietnamese

This study proposes an edge-based Vietnamese voice-control system for unmanned aerial vehicles. The system integrates a large language model, natural language processing modules, and real drone actuation to improve the reliability of naturallanguagebased drone control. The main contributionisan embedded mechatronic control framework experimentally validated on a commercial drone platform. This framework combines a multi-layered prompt structure with a preprocessing mechanism to reduce repetitive-word, invalid command generation, and redundant textual output. The proposed system uses a distributed edge architecture. An NVIDIA Jetson Orin NX performs Llama 3.1:8b inference, while a Raspberryhandles PhoWhisper-based speech recognition and direct DJI Tello control. Under noise-controlled conditions, the system achieved 100% accuracy for single, compound, and inferential command categories. The average response time was 2-3 s for single commands and 8.45 s for a representative complex compound command.