ChatMPC: Natural Language based MPC Personalization

TL;DR

This paper introduces ChatMPC, a natural language-based personalization method for control systems that allows real-time updates of control specifications via chat, reducing user burden and enhancing adaptability.

Contribution

It presents a novel approach integrating chat-based natural language interaction into MPC for personalized control, enabling real-time specification updates.

Findings

Control specifications can be updated through natural language chat.

The approach reduces user burden compared to traditional data collection.

Simulations show effective behavior changes in autonomous robots.

Abstract

We address the personalization of control systems, which is an attempt to adjust inherent safety and other essential control performance based on each user's personal preferences. A typical approach to personalization requires a substantial amount of user feedback and data collection, which may result in a burden on users. Moreover, it might be challenging to collect data in real-time. To overcome this drawback, we propose a natural language-based personalization, which places a comparatively lighter burden on users and enables the personalization system to collect data in real-time. In particular, we consider model predictive control (MPC) and introduce an approach that updates the control specification using chat within the MPC framework, namely ChatMPC. In the numerical experiment, we simulated an autonomous robot equipped with ChatMPC. The result shows that the specification in robot control is updated by providing natural language-based chats, which generate different behaviors.