Title: Autonomous navigation and task scheduling for agricultural robots using a transformer and deep Q-network

Authors: Yuequan Qiu; Jiusong Chen; Dejun Miao

Addresses: Yangzhou Vocational University, Yangzhou 225000, China ' Yangzhou Vocational University, Yangzhou 225000, China ' Yangzhou Vocational University, Yangzhou 225000, China

Abstract: Achieving precise autonomous navigation and efficient task scheduling for agricultural robots is a vital topic. Addressing the existing issues of navigation bias and poor scheduling efficiency, this paper first utilises a variational graph autoencoder to learn the latent features of the robot's state, then inputs the latent features into a Transformer to obtain global spatial interaction and simultaneously uses a deep Q-network (DQN) to find the optimal navigation strategy. Furthermore, the reward function is improved through the robot's task attention state. Priority experience replay is proposed to enhance training accuracy and is constructed in conjunction with the data in the experience pool for task scheduling. The simulation outcome indicates that the suggested method achieves a navigation success rate of 92.38%. When the number of tasks is 500, the task scheduling time is only 22.9 seconds, verifying the superiority of the suggested method.

Keywords: agricultural robots; autonomous navigation; task scheduling; transformer; deep Q-network; DQN.

DOI: 10.1504/IJSN.2025.150134

International Journal of Security and Networks, 2025 Vol.20 No.4, pp.243 - 254

Received: 22 Jun 2025
Accepted: 17 Jul 2025
Published online: 01 Dec 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article