Dimensions overall score 7.0

GitHub Code Pulse

No public code linked for this paper yet.

• Supervised fine-tuning (SFT) with Low-Rank Adaptation (LoRA) substantially improves LLM decision accuracy, behavioral consistency, and separation safety compared to a pretrained baseline for tactical deconfliction.result 95%
• The pretrained Qwen-Math-7B model (Base model) achieves only 27% accuracy on the tactical deconfliction evaluation task without domain adaptation.result 95%
• Supervised fine-tuning with LoRA achieves 88% accuracy on the evaluation dataset for the deconfliction task.result 95%
• Preference-based fine-tuning using Group-Relative Policy Optimization (GRPO) provides additional coordination benefits but exhibits reduced robustness when interacting with heterogeneous agent policies.result 90%
• The paper proposes a simulation-to-language data generation pipeline based on the BlueSky air traffic simulator to produce rule-consistent deconfliction datasets.method 90%
• Direct application of Large Language Models (LLMs) to air traffic control is limited by insufficient domain grounding and unpredictable output inconsistency.limitation 85%
• The GRPO fine-tuned model achieves 53% accuracy on the evaluation dataset, which is lower than the SFT model's 88%.result 90%
• Major companies deploying sUAS fleets face safety- and privacy-related constraints that limit the availability of high-fidelity operational data for training.market 80%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card