With rapid developing E-commerce, the demand for low-altitude logistic which integrates drones and trucks for delivery service has surged. An efficient simulation platform is essential for evaluating schemes and viability economic about large-scale network and complex operation elements. Combining the advantages of Large Language Model's (LLM) reasoning and Agent's action capabilities, this paper proposes an automatic modelling and simulation framework based on Unity3D. We firstly design Model Context Protocol (MCP) Servers and register corresponding tools for the agent, enabling map data acquisition, also modelling and simulation instructions generation. These instructions are transmitted to the Unity3D. Secondly, we design scripts within Unity3D to interpret instructions and execute tasks automatically for static scene modelling or delivery simulation, so that users can simply enter generation requirements in a dialog box with the agent, automatically drive the engine. Furthermore, a road network recognition and data alignment method based on the LLM's multimodal capabilities is studied, which can generate structured road network data to support dynamic simulation element modelling (e.g. road network, traffic facilities), and then support low-altitude logistics simulation. Finally, experiment results demonstrate that the effectiveness of this framework on automatic scenes modelling (even using low-parameter LLMs) for supporting low-altitude logistics simulation.
Chen et al. (Wed,) studied this question.