Towards On-Demand Metaverse Service Deployment in 6G Vehicular Networks Using Multimodal LLMs

The Sixth-Generation (6G) technology is envisioned to revolutionize vehicular networks by enabling dynamic, on-demand service deployment, particularly within the immersive domains of the metaverse. However, existing solutions are limited by generalized services for specific traffic events, the immobility of Roadside Units (RSU), and limited context awareness in unimodal data inputs. This paper proposes a vehicular Multimodal Large Language Model (MLLM)-driven framework for context-aware and heterogeneous service deployment in 6G vehicular networks. Specifically, we capitalize on multimodal data (e.g., text, images, and videos) with rich contextual meanings to enhance the vehicular MLLM's inference accuracy for on-demand service deployment recommendation. Considering the fixed locations of RSUs, we introduce OBU clusters as alternative node options for hosting recommended on-demand services. Then, we develop an optimal node selection algorithm that selects the most suitable RSU or OBU cluster node for an application-specific metaverse service deployment, considering their resource allocations, expected Quality of Service (QoS), and Resource Utilization (RU) offerings, as well as resource constraints. Comprehensive simulation results reveal that the proposed vehicular MLLM improves perception and service recommendation accuracy by about 12.5% and 8.9%, respectively, compared with GPT-4. Moreover, the node selection algorithm selects the optimal node with the highest expected utility for on-demand service deployment.