Dual Mind World Model Inspired Network Digital Twin for Access Scheduling

Emerging networked systems such as industrial IoT and real-time cyber-physical infrastructures demand intelligent scheduling strategies capable of adapting to dynamic traffic, deadlines, and interference constraints. In this work, we present a novel Digital Twin-enabled scheduling framework inspired by Dual Mind World Model (DMWM) architecture, for learning-informed and imagination-driven network control. Unlike conventional rule-based or purely data-driven policies, the proposed DMWM combines short-horizon predictive planning with symbolic model-based rollout, enabling the scheduler to anticipate future network states and adjust transmission decisions accordingly. We implement the framework in a configurable simulation testbed and benchmark its performance against traditional heuristics and reinforcement learning baselines under varied traffic conditions. Our results show that DMWM achieves superior performance in bursty, interference-limited, and deadline-sensitive environments, while maintaining interpretability and sample efficiency. The proposed design bridges the gap between network-level reasoning and low-overhead learning, marking a step toward scalable and adaptive NDT-based network optimization.

💡 Research Summary

The paper introduces a novel scheduling framework for wireless IoT and cyber‑physical systems that merges a Dual‑Mind World Model (DMWM) with a Network Digital Twin (NDT). The DMWM architecture consists of two complementary decision‑making modules: a “Fast Mind” that executes a lightweight heuristic (essentially a Longest‑Queue‑First rule limited to K simultaneous transmitters) and a “Slow Mind” that performs symbolic, model‑based roll‑outs inside the digital twin to evaluate the long‑term consequences of candidate schedules.

The system model assumes a slotted wireless network with N sensor nodes, each maintaining a bounded FIFO queue. Packets arrive according to a time‑varying Poisson process, and each node may have a deadline D_i. Interference is captured by a conflict graph C, and at most K nodes can transmit per slot. The NDT mirrors the physical network, providing a fast, accurate simulation of traffic, queue dynamics, deadlines, and interference without affecting the real system.

The Slow Mind first filters all possible K‑node schedules through an Informed Constraint Navigation (ICN) module that checks (i) non‑empty queues, (ii) interference avoidance, and (iii) deadline feasibility. For each feasible schedule S, the symbolic world model simulates H steps (H=3 in the experiments) using a simple queue‑evolution rule: scheduled nodes decrement their queue by one packet, while others remain unchanged. The cumulative reward R(S) is the total number of packets transmitted over the horizon. The schedule with maximal R(S) is selected as S_slow. If the feasible set is empty, or if the roll‑out computation cannot be completed within the time budget, the Fast Mind fallback selects the K nodes with the largest weighted queue lengths (the weight incorporates deadline urgency). The final decision rule chooses S_slow when available, otherwise S_fast.

Complexity analysis shows the Slow Mind’s planning cost scales as O(N·K·H·N), but ICN dramatically prunes the search space in practice. The authors implement the framework in a configurable Python‑based simulator with 5 nodes, K=3, queue capacity 50, and four traffic scenarios: (1) default moderate interference and mixed deadlines, (2) bursty traffic without deadlines, (3) deadline‑sensitive traffic with short and long deadlines, and (4) interference‑constrained traffic without deadlines. Baselines include Random selection, Longest‑Queue‑First (LQF), a Deadline‑Priority heuristic, Max‑Weight Scheduling, and a Q‑learning reinforcement‑learning agent.

Results demonstrate that DMWM consistently outperforms all baselines. In bursty traffic, DMWM reduces average packet delay by 15‑30 % and maintains higher throughput than Max‑Weight. In deadline‑sensitive scenarios, deadline violations drop by 20‑45 % compared with both heuristic and RL methods. The symbolic planning provides foresight that mitigates queue buildup, while the Fast Mind ensures real‑time responsiveness under heavy load. Moreover, the learning‑free nature of the Slow Mind yields sample efficiency: the RL baseline requires thousands of training episodes to converge, whereas DMWM achieves its performance with fewer than ten roll‑out evaluations per decision epoch.

The paper’s contributions are: (i) a hybrid cognitive scheduler that blends reactive heuristics with deliberative symbolic planning, (ii) an efficient constraint‑checking module that guarantees interference and deadline compliance, (iii) a demonstration that digital‑twin‑based imagination can replace data‑intensive model‑free RL, and (iv) extensive simulation evidence of superior delay, throughput, and feasibility metrics across diverse IoT scenarios.

Limitations are acknowledged. The exhaustive enumeration of K‑node schedules becomes combinatorial as N grows, suggesting the need for additional pruning, sampling, or meta‑heuristic search for larger networks. The current design assumes a centralized scheduler; extending the approach to distributed settings would require synchronization mechanisms and consistent twin state updates.

Overall, the work showcases how symbolic world‑model roll‑outs inside a digital twin, coupled with a fast fallback, can deliver high‑performance, interpretable, and sample‑efficient scheduling for deadline‑critical, interference‑limited wireless networks, marking a significant step toward practical, AI‑augmented network control.