Answers are grounded in this paper and cite their sources.
In this paper, to satisfy the diverse task demands of Internet of Remote Things (IoRT) devices, we propose a multi-satellite collaborative model deployment and satellite–terrestrial inference framework for IoRT devices. Moreover, we formulate a joint model deployment, task scheduling, and resource allocation (MTR) problem for IoRT devices, aiming to minimize the long-term average cost measured by weighted latency and energy consumption under constraints. Considering the different timescales of these subproblems, we decompose the MTR problem into a model deployment subproblem and a task scheduling–resource allocation subproblem. We define the model deployment subproblem as a large-timescale process and the task scheduling–resource allocation subproblem as a small-timescale process. For the model deployment subproblem, we propose a large-timescale surrogate-assisted model deployment (LT-SAMD) algorithm. For the task scheduling–resource allocation subproblem, we model it with a constrained Markov decision process (CMDP), and propose asmall-timescale hybrid proximal policy optimization and convex optimization (ST-HPCO) algorithm to solve it. In addition, we propose a global two-timescale decouple execution (TT-DE) algorithm that integrates ST-HPCO and LT-SAMD algorithms to solve the MTR problem.Simulation results demonstrate that, compared with the PPO-only baseline and the AOS-PPO algorithm, our proposed algorithm achieves cost reductions of up to 60% and 28%, respectively.
No takes yet. Share an insight, caveat, or question.
Liu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: