Work

A Hybrid Agent-based Navigation System with Semantic Reasoning

Year: 2026
Type: conference-paper
Abstract: Recent advances in Large Language Models (LLMs) enable natural language interaction in intelligent transportation systems, yet their direct application to navigation remains limited by reliability, spatial consistency, and computational constraints. This work introduces HAN, a hybrid agent-based navigation assistant that integrates classical routing algorithms with semantic reasoning provided by language models. Instead of generating routes directly, the proposed architecture separates route computation from semantic decision-making through a two-agent structure: an assistant agent responsible for interpreting user intent and extracting routing requirements, and a decision agent that selects among pre-computed candidate routes under contextual constraints. The system is evaluated as a complete pipeline, considering intent interpretation, routing decisions, and operational overhead. Experiments conducted in three cities with distinct urban characteristics, namely Nice (France), Barcelona (Spain), and Reykjavik (Iceland), demonstrate that the proposed approach achieves high precision and comprehensiveness in identifying the intended points of interest, while maintaining stable end-to-end execution times. Results show that constraining language model reasoning to a limited set of algorithmically generated alternatives improves decision reliability and prevents spatial inconsistencies commonly observed in fully generative approaches. The findings indicate that hybrid semantic–algorithmic architectures represent a practical and scalable direction for intent-aware navigation in connected and intelligent transportation environments.
Language: English
Open Access status: hybrid
OpenAlex ID: W7204489481
FWCI:
Cites: 22
Cited by:
Related to: 10