A Generalized ChatGPT-Based Collaborative Multi-Objective Decision-Making Framework for Robust Vehicle Platoon Collision Avoidance

Abstract

In this work, we propose a ChatGPT-based framework that collaborates with human expert instructions for safe and efficient vehicle platoon control. The collaborative framework integrates multi-objective decision-making to address challenges such as robust collision avoidance in complex traffic environments. To incorporate human-like reasoning, ChatGPT is firstly used to interpret human expert driving instructions, which often involve linguistic variables. It generates behavioral decisions and computes feasible trajectories for each vehicle to ensure driving safety. A distributed model predictive controller is then utilized for accurate, real-time path tracking. To efficiently handle ChatGPT response failures, a receding instruction utilization strategy is proposed. For critical response failures, hindsight experience replay is used to supplement ChatGPT's knowledge database and improve response accuracy in complex scenarios. Simulation tests and experimental validations demonstrate the framework's effectiveness and generalizability across various traffic conditions, highlighting its potential for broader applications in platoon control.