GPT-4 enhanced multimodal grounding for autonomous driving: Leveraging cross-modal attention with large language models | Synapse