Bus-Conditioned Zero-Shot Trajectory Generation via Task Arithmetic

TL;DR

This work tackles the challenge of generating mobility trajectories for a target city with no target-city mobility data by leveraging source-city mobility data and publicly available bus timetables. It introduces MobTA, which embeds task arithmetic into trajectory generation by modeling and transferring the parameter shift from bus-timetable-based to mobility trajectory generation across cities, using a unified trajectory representation. The paper provides a theoretical stability analysis showing that task vectors derived from base and instruction-tuned LLMs project stably onto a low-dimensional, task-relevant subspace, and demonstrates through extensive experiments on Shanghai, Wuxi, and Singapore that MobTA outperforms baselines and approaches performance of models finetuned on target-city trajectories. This approach offers a practical, data-efficient pathway for cross-city mobility synthesis, with implications for urban planning and privacy-preserving data analysis in data-inaccessible settings.

Abstract

Mobility trajectory data provide essential support for smart city applications. However, such data are often difficult to obtain. Meanwhile, most existing trajectory generation methods implicitly assume that at least a subset of real mobility data from target city is available, which limits their applicability in data-inaccessible scenarios. In this work, we propose a new problem setting, called bus-conditioned zero-shot trajectory generation, where no mobility trajectories from a target city are accessible. The generation process relies solely on source city mobility data and publicly available bus timetables from both cities. Under this setting, we propose MobTA, the first approach to introduce task arithmetic into trajectory generation. MobTA models the parameter shift from bus-timetable-based trajectory generation to mobility trajectory generation in source city, and applies this shift to target city through arithmetic operations on task vectors. This enables trajectory generation that reflects target-city mobility patterns without requiring any real mobility data from it. Furthermore, we theoretically analyze MobTA's stability across base and instruction-tuned LLMs. Extensive experiments show that MobTA significantly outperforms existing methods, and achieves performance close to models finetuned using target city mobility trajectories.

Figures (14)

• Figure 1: Conceptual overview of MobTA. It uses task arithmetic to model the parameter shift from bus-timetable-based trajectory generation to mobility trajectory generation in the source city, and applies this shift to the target city, where only bus timetables are accessible, to enable mobility trajectory generation for target city.
• Figure 2: Overall framework of MobTA. MobTA constructs three task vectors corresponding to mobility trajectory generation in the source city, as well as bus-timetable-based trajectory generation in both the source and target cities. It then models the parameter shift from bus-timetable-based trajectory generation to mobility trajectory generation in the source city and applies this shift to the target city’s bus-timetable-based trajectory generation task vector, enabling mobility trajectory generation for the target city.
• Figure 3: Screenshot of bus timetable in Google Map APP.
• Figure 4: Screenshot of bus timetable in Naver Map APP.
• Figure 5: Hourly temporal distribution of bus timetables and mobility trajectories in Shanghai.