Cooperative encirclement of multi-USVs under communication constraints via prediction-augmented multi-agent reinforcement learning

Wu, Yixian, Li, Hongbin, Tian, Shunyu, Ji, Ze ORCID: https://orcid.org/0000-0002-8968-9902 and Wei, Changyun 2026. Cooperative encirclement of multi-USVs under communication constraints via prediction-augmented multi-agent reinforcement learning. Ocean Engineering 364 (1) , 126872. 10.1016/j.oceaneng.2026.126872

Full text not available from this repository.

Abstract

Cooperative encirclement by multiple unmanned surface vehicles (USVs) is highly dependent on reliable inter-USV communication, which is often degraded in realistic maritime environments due to packet loss and transmission interruptions. Such communication constraints lead to incomplete observations and missing teammate information, thereby undermining coordination stability and task performance. To address these challenges, this paper proposes an enhanced multi-agent reinforcement learning framework termed ARPE-MATD3 (Augmented Representation via Prediction and Encoding based MATD3), which enhances state representation through trajectory prediction and temporal encoding. Specifically, a multi-step evader trajectory prediction module is introduced to capture the target’s future motion trends from historical observations, enabling proactive decision-making by incorporating predicted intention features into the observation space. To mitigate observation discontinuity caused by communication loss, a temporal encoding mechanism combined with a self-supervised reconstruction module is further developed, which leverages historical information to recover missing teammate states and learn robust latent representations. Extensive simulation results under varying communication conditions show that the proposed method achieves higher encirclement success rates, more stable formation behaviors, and better robustness than the compared baselines in the tested scenarios. These results highlight the effectiveness of integrating predictive priors and temporal representation learning for multi-USV cooperative tasks in communication-constrained maritime environments.

Item Type:	Article
Date Type:	Publication
Status:	Published
Schools:	Schools > Engineering
Publisher:	Elsevier
ISSN:	0029-8018
Date of Acceptance:	2 July 2026
Last Modified:	20 Jul 2026 10:45
URI:	https://orca.cardiff.ac.uk/id/eprint/188308

Actions (repository staff only)

Edit Item