Fine-Tuning Large Language Models for Cooperative Tactical Deconfliction of Small Unmanned Aerial Systems

Iman Sharifi; Alex Zongo; Peng Wei

cs.ROMar 30, 2026

Fine-Tuning Large Language Models for Cooperative Tactical Deconfliction of Small Unmanned Aerial Systems

Authors: Iman Sharifi, Alex Zongo, Peng Wei

Organizations: George Washington University

Abstract

The growing deployment of small Unmanned Aerial Systems (sUASs) in low-altitude airspaces has increased the need for reliable tactical deconfliction under safety-critical constraints. Tactical deconfliction involves short-horizon decision-making in dense, partially observable, and heterogeneous multi-agent environments, where both cooperative separation assurance and operational efficiency must be maintained. While Large Language Models (LLMs) exhibit strong reasoning capabilities, their direct application to air traffic control remains limited by insufficient domain grounding and unpredictable output inconsistency. This paper investigates LLMs as decision-makers in cooperative multi-agent tactical deconfliction using fine-tuning strategies that align model outputs to human operator heuristics. We propose a simulation-to-language data generation pipeline based on the BlueSky air traffic simulator that produces rule-consistent deconfliction datasets reflecting established safety practices. A pretrained Qwen-Math-7B model is fine-tuned using two parameter-efficient strategies: supervised fine-tuning with Low-Rank Adaptation (LoRA) and preference-based fine-tuning combining LoRA with Group-Relative Policy Optimization (GRPO). Experimental results on validation datasets and closed-loop simulations demonstrate that supervised LoRA fine-tuning substantially improves decision accuracy, consistency, and separation performance compared to the pretrained LLM, with significant reductions in near mid-air collisions. GRPO provides additional coordination benefits but exhibits reduced robustness when interacting with heterogeneous agent policies.

Figures & tables

Figure 1: Architecture overview. The figure illustrates the end-to-end system architecture and the role of the proposed simulation-to-language dataset generation pipeline. Multi-agent traffic scenarios are generated in the BlueSky simulator, from which raw state data are extracted and converted into structured natural-language prompts using rule-based supervision. The resulting prompt–response pairs constitute the training dataset for LoRA-based fine-tuning. At deployment, the fine-tuned LLM generates tactical actions for multiple agents, which are executed in BlueSky, closing the simulation loop.

Parameter	Notation	Configuration
Parameter	Notation	X (strong)	Y (weak)
Speed Range (m/s)	$[v_{\text{min}},v_{\text{max}}]$	$[0,44.88]$	$[0,30.12]$
Acceleration (m/s 2 )	$\Delta v/\Delta t$	$\{-1.71,0,1.71\}$	$\{-1.02,0,1.02\}$
Sensing Range (m)	$\mathcal{R}$	$1000$	$750$

Table 1: Kinematic and sensing specifications for UAS configurations X (strong) and Y (weak).

Panel of Figure 2 — Figure 2: Training effectiveness of fine-tuning methods. (a) shows the supervised learning progress through loss reduction, hence accuracy increase, while (b) shows the GRPO reward evolution across training iterations.

Panel of Figure 3 — Figure 3: Traffic snapshots for the three scenarios (A, B, C) used in Table 3 . The LLM agents and the Rule-based agents are colored in pink and green, respectively. Each scenario has 5-6 routes, each of which hosts 5 agents with random spawning times. Throughout all scenarios, we considered 10 LLM agents, and the rest are Rule-based agents.

Model	Accuracy	Precision	Recall	F1-score
Base	27	75	20	31
SFT	88	75	66	69
GRPO	53	75	40	50

Table 2: Performance comparison on the evaluation dataset. All numbers are reported in percent (%).

Base

SFT

GRPO

Scen.

All

L–L

L–R

Time

All

L–L

L–R

Time

All

L–L

L–R

Time

3.5

\pm

1.1

2.7

\pm

0.5

0.8

\pm

1.0

0.12

\pm

0.38

3.7

\pm

4.1

1.0

\pm

0.8

0.7

\pm

0.7

0.3

\pm

0.5

0.77

\pm

0.29

5.7

\pm

0.6

1.7

\pm

0.5

1.1

\pm

0.7

0.6

\pm

0.5

0.57

\pm

0.31

5.2

\pm

0.1

3.4

\pm

1.1

1.8

\pm

1.4

1.6

\pm

0.9

0.20

\pm

0.56

3.3

\pm

7.0

1.9

\pm

1.2

0.9

\pm

0.7

1.0

\pm

0.8

0.62

\pm

0.36

8.1

\pm

0.9

3.0

\pm

0.7

1.3

\pm

0.4

1.7

\pm

0.5

0.27

\pm

0.22

6.9

\pm

0.3

4.0

\pm

0.9

2.5

\pm

1.2

1.5

\pm

1.2

0.05

\pm

0.58

1.6

\pm

5.0

1.9

\pm

0.7

0.8

\pm

0.6

1.1

\pm

0.9

0.52

\pm

0.37

7.5

\pm

0.7

2.3

\pm

0.8

0.6

\pm

0.5

1.7

\pm

0.7

0.42

\pm

0.29

6.6

\pm

0.1

Table 3: Safety and efficiency across configurations and LLM models (mean

\pm

std, for 10 episodes). Rates are NMACs/episode. Abbreviations: L–L = NMACs between two LLM agents; L–R = NMACs between an LLM agent and a Rule-based agent; All

=

L–L

+

L–R ; SR = success rate of LLM agents (fraction completing without NMACs); Time = average flight time of successful LLM agents. Bold indicates best values: lowest NMAC for (All, L–L, L–R), highest SR, and lowest Time among methods with SR

\geq 0.9\times

{}_{\text{best}}

for the scenario.

Appendix figures & tables3 assets

Supplementary material from the paper’s appendix.

Appendix

Figure 4: Decision rules for the rule-based policy, organized by ownship proximity to the next waypoint. The policy distinguishes between situations where the ownship is far from the waypoint (

d_{o}^{\mathrm{wp}}>d_{o}^{\mathrm{safe}}

) and near the waypoint (

d_{o}^{\mathrm{wp}}\leq d_{o}^{\mathrm{safe}}

), with speed constraint enforcement applied as a final override.

Figure 5: Example prompt for tactical deconfliction at a single time step. The system prompt establishes the model’s role and constraints, while the user prompt provides a structured description of the current traffic situation. Qualitative descriptors are derived from numerical thresholds to support natural-language reasoning.

Figure 6: Target response format corresponding to the prompt in Figure 5 . The constrained response format ensures consistent parsing during both training and closed-loop inference.

Language Model-Based Control

Language model-based control uses language models to interpret goals, choose actions, and steer AI systems or physical processes.

See all 180 papers

Separation Assurance between Heterogeneous Fleets of Small Unmanned Aerial Systems via Multi-Agent Reinforcement Learning
May 1, 2026Iman Sharifi, Hyeong Tae Kim, Maheed Hatem Ahmed +2Multi-Agent Coordination Multi-Agent Reinforcement Learning
Department of Mechanical and Aerospace Engineering, George Washington University, Washington, DC, USA · Department of Electrical and Computer Engineering, Purdue University, West Lafayette, IN, USA
LLM-Enabled UAV Dispatch: A System-Level Survey and Taxonomy
Oct 7, 2026Xiao Han, Aoyang Quan, Xiangyu Zhao +2LLM Agent Orchestration
School of Computer Science, Zhejiang University of Technology, China
MultiUAV-Plat: An LLM-Oriented Platform, Benchmark and Framework for Multi-UAV Collaborative Task Planning
Jun 30, 2026Sheng Zhang, Qinglin Li, Yuechao Zang +3Aerial Robotics Large Language Model-Based Robot Planning
National Key Laboratory of Information Systems Engineering, National University of Defense Technology, Deya Road, Changsha, 410073, Hunan, China

May 1, 2026J/K move · Enter open · S save

Separation Assurance between Heterogeneous Fleets of Small Unmanned Aerial Systems via Multi-Agent Reinforcement Learning

Iman Sharifi, Hyeong Tae Kim, Maheed Hatem Ahmed, Mahsa Ghasemi+1

Department of Mechanical and Aerospace Engineering, George Washington University, Washington, DC, USA · Department of Electrical and Computer Engineering, Purdue University, West Lafayette, IN, USA

Multi-Agent Coordination Multi-Agent Reinforcement Learning

Open paper AskPDF arXiv

In the envisioned future dense urban airspace, multiple companies will operate heterogeneous fleets of small unmanned aerial systems (sUASs), where each fleet includes several homogeneous aircraft with identical policies and configurations, e.g., equipage, sensing, and communication ranges, making tactical deconfliction highly complex for the aircraft. This paper aims to address two core questions: (1) Can tactical deconfliction policies converge or reach an equilibrium to ensure a conflict-free airspace when companies operate heterogeneous fleets of homogeneous aircraft? (2) If so, will the converged policies discriminate against companies operating sUASs with weaker configurations? We investigate a multi-agent reinforcement learning paradigm in which homogeneous aircraft within heterogeneous fleets operate concurrently to perform package delivery missions over Dallas, Texas, USA. An attention-enhanced Proximal Policy Optimization-based Advantage Actor-Critic (PPOA2C) framework is employed to resolve intra- and inter-fleet conflicts, with each fleet independently training its own policy while preserving privacy. Experimental results show that two fleets with distinct, shared PPOA2C policies can reach an equilibrium to maintain safe separation. While two PPOA2C policies outperform two strong rule-based baselines in terms of conflict resolution, a PPOA2C policy exhibits safer interaction with a rule-based policy, indicating adaptive capabilities of PPOA2C policies. Furthermore, we conducted extensive policy-configuration evaluations, which reveal that equilibria between similar policy types tend to favor fleets with stronger configurations. Even under similar configurations but different policy types, the equilibrium favors one of the heterogeneous policies, underscoring the need for fairness-aware conflict management in heterogeneous sUAS operations.

Abstract

Language Model-Based Control

Explore similar work

Separation Assurance between Heterogeneous Fleets of Small Unmanned Aerial Systems via Multi-Agent Reinforcement Learning

LLM-Enabled UAV Dispatch: A System-Level Survey and Taxonomy

MultiUAV-Plat: An LLM-Oriented Platform, Benchmark and Framework for Multi-UAV Collaborative Task Planning