Abstract
The movable antenna (MA) technology has recently demonstrated significant potential in improving the communication performance by enabling local movement of the antennas to obtain better channel conditions. Simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS) has also emerged as a promising technology enabling full-space coverage by dividing incident signals into independently tunable transmitted and reflected components. In this paper, we propose a novel STAR-RIS-assisted multiuser communication system with MAs for simultaneous wireless information and power transfer (SWIPT). In order to achieve maximum sum-rate and energy harvesting efficiency, the proposed system optimizes beamforming, transmission/reflection coefficients (RCs), and MA positions simultaneously using a meta twin delay deep deterministic policy gradient (MTD3) reinforcement learning (RL) approach. The numerical results demonstrate the superiority of MTD3 over conventional baselines across various system settings, including different numbers of RIS elements, power budget, and minimum energy constraints, making it more robust and efficient for real-time resource allocation in SWIPT systems.