Learning-Based Resource Allocation in 5G NR Mode-2 Sidelink for Industrial AGV and AMR Communications

Abstract

As the manufacturing sector increasingly adopts Industry 4.0 technologies, the need for reliable communication among devices, such as autonomous mobile robots (AMRs), or automated guided vehicles (AGVs), becomes a fundamental necessity. To support direct device-to-device communication, the third-generation partnership project (3GPP) introduced 5G new radio sidelink communication mode 2 (NR-SL) in Release 16, and 17. NR-SL allows devices to select transmission resources based on local channel sensing. In NR-SL, however, autonomous resource selection can lead to collisions, particularly in dense industrial environments. In this paper, we propose a learningbased resource allocation scheme (LBRA) for 5G NR Mode-2 sidelink. LBRA is designed to support industrial AGV and AMR communications. It employs multi-agent reinforcement learning (MARL) to improve resource allocation in NR-SL and reduces the collision probability. The results show that LBRA decreases the collision probability by approximately 73% compared to NR-SL.