[5I1-OS-3] OS-3

9:00 AM - 9:15 AM JST(12:00 AM - 12:15 AM UTC)

[5I1-OS-3-01] ChatMPC for Driver AssistanceRethinking the Human Role in Autonomous Control Systems

〇Yuya Miyaoka¹, Masaki Inoue¹ (1. Keio University)

Keywords:

Control Engineering,Autonomous Driving

This study presents an application of ChatMPC, a framework that enables humans to adjust control-system specifications using natural language. Although autonomous driving technology has made significant progress in recent years, achieving safe operation in complex environments requires addressing a wide variety of edge cases. Thus, incorporating human judgment as a supplementary information remains important. In this work, we assume a scenario in which humans act as “patrons” who actively contribute to the safe operation of autonomous vehicles, and we propose a natural language interface and the supporting control architecture to realize the human-machine interaction.This presentation focuses on situations where an autonomous vehicle fails to perceive external information, such as obstacles, and introduces a method that leverages user-provided information to overcome such edge cases. Specifically, when a user provides verbal information about an obstacle, a language model interprets the user’s intention and estimates the location of the obstacle. Based on this inferred information, the control system generates an avoidance operation.IT services where AI systems make autonomous decisions supported by human input will be trending. This presentation aims to extend this emerging human-AI relationship to physical systems.