Digital Twin–Guided AI Path Planning for Connectivity-Aware Mobility

Digital twins are increasingly used in wireless networks to provide geometry- and channel-aware replicas that support prediction and control, while large language models (LLMs) offer strong instruction-following capabilities but remain largely ungrounded in radio environments. This paper presents DTMAP, a digital twin-guided path planning framework that integrates a wireless digital twin with an LLM to enable connectivity-aware navigation in urban settings. Routing is formulated as a parameterized multi-objective optimization problem that trades off signal strength and travel distance through a continuous scalarization factor, and the digital twin generates objective-consistent trajectories used to fine-tune the LLM. The resulting model adapts to different operating points and natural-language instructions while mitigating shortest-path bias. Extensive evaluations across continuous trade-off values, diverse prompting conditions, and multiple baseline methods, including classical graph algorithms, reinforcement learning, and multiple foundation models, demonstrate high geometric validity, stable coverage–distance trade-offs, and reduced outage probability relative to learning-based baselines. Robustness experiments under perturbed radio maps and unseen configurations show consistent generalization without retraining, and measured sub-second end-to-end inference latency indicates practical feasibility. These results illustrate how digital twins can provide structured supervision that grounds generative models for connectivity-aware decision making in future wireless systems.