Hierarchical dual-timescale deep reinforcement learning for mechanically rotatable STAR-RIS assisted mobile edge computing networks

The integration of Mobile Edge Computing (MEC) with Reconfigurable Intelligent Surfaces (RIS) has emerged as a promising paradigm to address the challenges of latency-sensitive applications in dynamic wireless environments. This paper proposes a multi-rotational Simultaneously Transmitting and Reflecting RIS (STAR-RIS) assisted MEC framework that considers both mechanical reorientation and electronic control. Unlike conventional STAR-RIS systems with fixed orientations, the proposed architecture enables physical rotation of STAR-RIS panels to dynamically adjust spatial coverage and achieve improved geometric alignment with mobile users. A comprehensive system model is developed incorporating rotatable STAR-RIS panels, dual-region users, task offloading decisions, and wireless channel dynamics. To address the resulting mixed-timescale, non-convex stochastic optimization problem, a hierarchical dual-timescale Deep Deterministic Policy Gradient (DDPG) - based framework is proposed, where a slow-timescale agent controls mechanical rotation while fast-timescale agents jointly optimize electronic STAR-RIS parameters and MEC offloading decisions. A formal non-stationarity analysis demonstrates the superiority of dual-timescale DRL and a convergence rationale is established through two-timescale stochastic approximation theory. Simulation results demonstrate that the proposed approach achieves stable convergence and significantly outperforms benchmark schemes in terms of task completion rate, fairness, sum rate, and energy efficiency. The findings highlight that mechanical rotation introduces an additional spatial degree of freedom and enhances the MEC performance in dynamic mobile environments.