RIDE: Resilient Intent Dynamics Extrapolation for Asynchronous Multi-UAV Coordination
Year: 2026
Type: conference-paper
Abstract: Multi-agent reinforcement learning (MARL) has been widely applied to optimize unmanned aerial vehicle (UAV)assisted communication networks. However, most existing MARL algorithms rely on an idealized assumption that agents are connected by continuous, synchronized, and flawless communication links. In realistic and complex communication environments, severe interference and network congestion often cause frequent burst packet losses and time-varying delays, leading to sudden breakdowns in multi-agent coordination. To address this challenge, this paper departs from the constraints of traditional discrete-time communication models and proposes a continuoustime asynchronous communication framework based on Resilient Intent Dynamics Extrapolation (RIDE). RIDE unifies packet loss and delay as irregular asynchronous event streams and introduces a time-aware Flow-Jump process to continuously extrapolate teammate latent states during communication interruptions and instantly correct the extrapolated results when delayed packets arrive, thereby mitigating the information vacuum caused by packet loss and the temporal misalignment caused by delay. Extensive experiments demonstrate that RIDE maintains highly effective coordination under high packet loss rates and severe dynamic delays, while its extremely small computational overhead makes it well suited for edge-computing deployment.
Authors: Minhan Gao, Hao Gong, Zhongwei Huang
Institution: Guangdong Polytechnic Normal University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7204171133