Transactions of Nanjing University of Aeronautics & Astronautics
UAV Confrontation Decision-Making Based on Fictitious Self-play Multi-agent Proximal Policy Optimization
- Article
- Figures
- Metrics
- Preview PDF
- Reference
- Related
- Cited by
- Materials
Abstract:
This paper addresses the confrontation decision-making problem of unmanned aerial vehicles（UAVs） based on fictitious self-play multi-agent proximal policy optimization. UAV confrontation relies on autonomous decision-making， enabling the UAV to generate action instructions based on environmental information. An innovative autonomous decision-making methodology for UAV confrontations is proposed within the context of red-blue air combat tasks. Initially， the current situation is evaluated by employing the relative angle between the missile attack area and the UAV. Following this， guided by the evaluated scenario， the design of state space， action space， and real-time reward feedback is implemented to streamline the training process. Subsequently， an advanced method is introduced for optimizing strategy through a virtual autonomous agent’s proximity， aiming to derive the advantage function and average strategy from the experience buffer of training data. Ultimately， the efficacy and superiority of the proposed method are validated through simulations of UAVs engaging in red-blue countermeasure tasks.
Keywords:
Project Supported:
This work was supported by the National Natural Science Foundation of China （No.62173242）.
Clc Number:
TP181
CHINA AERO-SCIENCES
Mobile website
