Reproduce “LLM-Enabled Low-Altitude UAV Natural Language Navigation via Signal Temporal Logic Specification Translation and Repair” 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
Yuqi Ping, Huahao Ding, Tianhao Liang, Longyu Zhou, Guangyu Lei, Xinglin Chen, Junwei Wu, Jieyu Zhou and 1 more
This research introduces a unified framework enabling low-altitude UAV navigation from natural language instructions, achieving 98.0% exact-match accuracy in translating natural language to Signal Temporal Logic specifications using a reasoning-enhanced LLM. The system further incorporates an LLM-guided mechanism to repair planning infeasibilities by adaptively relaxing non-safety-critical task constraints, validated through simulations and real-world flight experiments.
Key insights
- Combining explicit reasoning (Chain-of-Thought) and structured reinforcement learning (GRPO) significantly enhances the accuracy and stability of LLM-based natural language to formal logic translation.
- Formal methods like MILP can precisely diagnose the sources of planning infeasibility by identifying minimal sets of conflicting constraints.
- Large Language Models can provide semantically aware guidance for repairing formal specifications, selectively relaxing non-safety-critical components while preserving core user intent and hard safety guarantees.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.