Digital Twin-Assisted Task Offloading and Resource Allocation in ISAC-Enabled Internet of Vehicles

TL;DR

The paper studies digital twin (DT)-assisted task offloading and resource allocation in ISAC-enabled IoV to minimize the long-term average system cost, defined as a weighted sum of delay and energy, while ensuring queue stability, expressed as $\min_{\{\mathbf{b}(t), \ell(t), F(t)\}} \lim_{T \to \infty} \frac{1}{T} \sum_{t} \mathbb{E}[\complement(t)]$. It proposes two transmission modes, DataT and InstrT, and leverages Lyapunov optimization to decompose the problem into per-slot subproblems, leading to the Ly-DTMPPO algorithm that operates within a CTDE framework with DT providing global state awareness and predictive guidance. A Markov decision process is formed for per-slot decisions, with a DT-enhanced state and a Lyapunov-based reward, and the DT-enhanced DRL algorithm employs MAPPO to coordinate multiple vehicles while respecting shared RSU resources. Empirical results show that Ly-DTMPPO achieves faster convergence, lower delay and energy consumption, and the lowest overall system cost compared with Lyapunov-driven baselines, and that the DT module enhances decision stability and computation offloading under dynamic mobility. This work demonstrates a scalable, cross-layer approach that combines Lyapunov stability with DT-enabled global visibility to optimize ISAC-enabled IoV in real time, with implications for robust, predictive resource management in future 6G-enabled intelligent transportation systems.

Abstract

The convergence of the Internet of vehicles (IoV) and 6G networks is driving the evolution of next-generation intelligent transportation systems. However, IoV networks face persistent challenges, including low spectral efficiency in vehicular communications, difficulty in achieving dynamic and adaptive resource optimization, and the need for long-term stability under highly dynamic environments. In this paper, we study the problem of digital twin (DT)-assisted task offloading and resource allocation in integrated sensing and communication (ISAC)-enabled IoV networks. The objective is to minimize the long-term average system cost, defined as a weighted combination of delay and energy consumption, while ensuring queue stability over time. To address this, we employ an ISAC-enabled design and introduce two transmission modes (i.e., raw data transmission (DataT) and instruction transmission (InstrT)). The InstrT mode enables instruction-level transmission, thereby reducing data volume and improving spectral efficiency. We then employ Lyapunov optimization to decompose the long-term stochastic problem into per-slot deterministic problems, ensuring long-term queue stability. Building upon this, we propose a Lyapunov-driven DT-enhanced multi-agent proximal policy optimization (Ly-DTMPPO) algorithm, which leverages DT for global state awareness and intelligent decision-making within a centralized training and decentralized execution (CTDE) architecture. Extensive simulations verify that Ly-DTMPPO achieves superior performance compared with existing benchmarks.

Figures (7)

• Figure 1: DT-Assisted ISAC vehicular network scenario. Part A illustrates the physical layer, where ISAC-enabled vehicles and RSUs cooperate in sensing, communication, and computation. Therein, ISAC communication is performed between the vehicle and the RSU. Part B presents the DT layer, which mirrors the physical environment through synchronized digital replicas, supporting predictive control and proactive resource optimization.
• Figure 2: Total task completion latency. Here, “Co-Trans” stands for coordinate transformation. Part A illustrates task generation at the CV, where data volume and computational attributes are defined. Part B shows the DT space that decides task-offloading ratios between local and RSU nodes to achieve low-latency cooperative computation.
• Figure 3: Framework of the proposed Ly-DTMPPO algorithm. Part A describes the DT-assisted vehicular environment where agents interact with the system. Part B stores and samples experiences for policy updates. Part C employs Lyapunov optimization to stabilize queue dynamics and balance delay–energy trade-offs. Part D adopts a CTDE structure with centralized critics and decentralized actors for cooperative policy learning.
• Figure 4: Average rewards under different algorithms.
• Figure 5: Task queue length comparison of different CVs and algorithms: (a) Task queue length of different CVs; (b) Task queue length of different algorithms.

Theorems & Definitions (1)

• Remark 1