CART: A Context-Aware Decision-Tree Framework for Adaptive Personalized Route Planning with LLMs

Abstract

Drivers’ preferences for road attributes vary across driving contexts, requiring route planning systems to adapt to individual preferences in order to provide personalized and satisfying driving experiences. However, existing route planning systems often fail to adequately account for these contextual and individual differences. While the emergence of large language models (LLMs) makes it feasible to interpret complex driving contexts and generate personalized strategies, mechanisms that allow a system’s internal decision-making structure to progressively adapt to user preferences remain limited. Consequently, users are often forced to repeatedly specify their preferences as driving conditions change, increasing the interaction burden. To address this challenge, we propose CART (Context-AwaRe Decision-Tree), a framework that adapts to user preferences by retrieving and updating a decision tree using an LLM. Results from a user study with 13 participants demonstrate that CART effectively incorporates user preferences over successive trials. This work offers key insights into: (1) delivering adaptive route recommendations in complex road environments; and (2) the design of autonomous driving systems that continuously evolve user preferences through dynamic user–vehicle interaction.