Conceptual voicebot in the context of a passenger information system in an automated bus

Haas, Benedikt; Sax, Eric · 2024 · Crossref

DOI: 10.54941/ahfe1004553

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the operational gap created by the removal of human drivers in automated public transportation, specifically focusing on automated buses. In traditional bus services, drivers perform critical non-driving tasks, such as answering passenger inquiries regarding routes, stops, and alternative paths. This function is particularly vital in bus systems, which operate on flexible public roads subject to dynamic changes like traffic jams or roadworks, unlike fixed-route rail systems. The authors propose conceptualizing a voicebot-based passenger information system to replace the driver’s communicative role, ensuring passengers receive personalized, understandable, and timely answers to reduce uncertainty. The study conceptualizes and compares three distinct technical approaches for the system’s processing unit: a rule-based system, a pure Large Language Model (LLM) system, and a hybrid system combining both. The proposed architecture includes speech-to-text conversion, noise filtering, and a secondary filter for inappropriate language. The rule-based approach uses predefined rules or Hidden Markov Models to extract information and generate answers from prepared sentences. The AI-based approach relies entirely on an LLM to process questions and generate natural language responses. The hybrid approach uses rule-based logic to infer intent and retrieve data, then employs an LLM to formulate the final answer. The authors evaluate these concepts against requirements such as multilingual support, privacy protection, handling of emergencies, and robustness against background noise and offensive language. The findings indicate that each approach has distinct trade-offs. The rule-based system offers high correctness and relevance but lacks flexibility and natural phrasing. The pure AI-based system provides natural, personalized answers but risks generating irrelevant, incorrect, or biased responses, potentially endangering passengers if it extrapolates beyond its operational design domain. The hybrid system is rated as the most suitable solution. It combines the deterministic control of rule-based systems—ensuring only relevant, correct, and up-to-date information is retrieved—with the natural language generation capabilities of LLMs. This configuration allows for the integration of real-time backend data while maintaining a conversational tone. The significance of this work lies in providing a structured framework for implementing human-machine interaction in driverless public transport. By identifying the hybrid model as optimal, the paper suggests a path toward systems that are both safe and user-friendly. The authors also highlight critical challenges for future implementation, including the need for extensive testing of inappropriate language handling, strategies for privacy preservation without constant listening, and protocols for forwarding emergency calls to management systems. The study concludes that while the hybrid approach is superior, further extensions, such as using a secondary LLM for intent extraction, could improve question understanding at the cost of increased computational resources.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• generative ai voice assistants