Search papers, labs, and topics across Lattice.
This paper introduces UrbanGround, a novel sandbox designed to evaluate how multimodal large language models (MLLMs) translate local urban perceptions into reliable actions within a complex, real-scale replica of Hong Kong. Through a series of experiments, the authors assess the agents' ability to ground local scenes for spatial reasoning, navigate increasingly distant destinations, and adapt to dynamic changes in the environment. The findings reveal that while MLLMs excel in visual recognition and short-range reasoning, they struggle with sustained goal-directed behavior during extended exploration, leading to cumulative errors and unreliable navigation.
MLLMs can recognize urban scenes but fail to maintain reliable navigation and goal-directed behavior over extended exploration in complex environments.
Multimodal large language models (MLLMs) can interpret a street view, but urban agency depends on whether such local evidence remains useful after the agent starts to move. In this paper, we investigate how far current MLLM agents can turn local urban perception into reliable action in a complicated real-scale city. We propose UrbanGround, the first sandbox to make this question testable in a physically constrained replica of Hong Kong built from territory-wide 3D geospatial data. UrbanGround supports closed-loop interaction from a first-person view and provides an interactive map for navigation. Agents can directly enter the 3D city and explore from a first-person view. Our analysis follows the growth of the spatial problem through three research questions. We first test whether an agent can ground a local scene well enough to answer spatial questions after active observation. Then we ask whether that grounding supports navigation as destinations become farther away and less explicit. Finally, we examine whether the resulting behavior survives changes in route availability and pedestrian motion. Contemporary MLLM agents usually show useful atomic abilities in visual recognition and short-range spatial reasoning, while orientation and pedestrian-aware movement remain unreliable. Their central failure emerges over extended exploration, where local abilities do not compose into sustained goal-directed behavior and errors accumulate without effective correction. We hope UrbanGround will support broader study of how far current MLLM agents can explore reliably in complex, open-ended urban environments.
