Cloud-Connected Model Predictive Control for Autonomous Mobile Robots in the Presence of Network Latency and Message Losses

Abstract

Autonomous driving systems have become increasingly popular due to their potential to reduce traffic accidents compared to human-operated vehicles, as well as their significant potential to optimize traffic flow and reduce pollution.However, there remain cases in which autonomous driving system malfunctions lead to accidents. These failures often arise from the partial observability of the environment and the inherent limitations of onboard sensor systems. To address these shortcomings, Cloud-Connected Autonomous System (CCAS) has emerged as a promising alternative, leveraging cloud-based computation and multiple distributed sensors to build a comprehensive understanding of the environment (including the ego vehicle) and make collective decisions for all controlled agents in the cloud. This approach improves decision-making and reduces the need to install costly onboard hardware on every vehicle by centrally sharing sensors and computational resources. Despite its advantages, introducing cloud connectivity into autonomous systems presents significant challenges, particularly network latency and message loss. Such latencies can negatively impact vehicle control and safety, especially when they exceed the control sampling interval, leading to unstable maneuvers or potential collisions. This work introduces a cloud controller designed to compensate for the effects of latencies longer than a single control sampling interval. A Robot Operating System (ROS)-based simulation environment was developed for rapid algorithm prototyping and seamless integration with a real testbed. The proposed solution was validated through both simulations and real-world tests involving an autonomous hospital bed using Ackermann steering in an indoor scenario. The experimental outcomes highlight the method’s effectiveness and practical potential.