Human-Centric Control Design for Safe & Connected Vehicles
Time: Thu 2025-03-27 09.00
Location: Kollegiesalen, Brinellvägen 8, Stockholm
Language: English
Subject area: Electrical Engineering
Doctoral student: Frank J. Jiang , Reglerteknik
Opponent: Professor Pierluigi Nuzzo, University of California, Berkeley, Department of Electrical Engineering and Computer Sciences
Supervisor: Professor Karl H. Johansson, Reglerteknik; Professor Jonas Mårtensson, Reglerteknik
Abstract
The road transportation network is one of the leading causes of injury and death in the world. Compared to aviation or rail, road transportation is significantly more dangerous due to its continued reliance on human drivers and the frequent occurrence of unsafe, complex traffic scenarios. Over the last decade, there has been a significant push to introduce vehicle automation into road transportation to address these challenges. By replacing the human driver, vehicle automation has the potential to revolutionize both the safety and efficiency of the road transportation network. However, in the most recent years, we have seen slower progress in this transformation. We attribute this slow down to the continued struggle for vehicle automation to handle a long tail of unexpected traffic issues, often stemming from occlusions, sensor uncertainty, or even system faults. One approach for addressing unexpected traffic issues is the integration of remote human operators who monitor, assist, and, when needed, control the vehicles. Although a key goal of vehicle automation has been to take humans out-of-the-loop, these remote human operators form a layer of resiliency that help fill in automation gaps and mitigate failures throughout a vehicle's operation. However, by integrating remote human operators, we risk introducing new human errors into the road transportation network.
In this thesis, we seek to address this challenge by designing a new control framework that explicitly and safely integrates remote human operators into the engineering and automation of connected vehicles. Our core design approach is to closely inspect the roles that remote human operators play when supervising connected vehicles and adapt traditional control principles to these roles. For this adaptation, we detail a new methodology that combines formal methods and reachability analysis to enable online verification. We show that we can verify an operator-designed specification by constructing a computational structure called temporal logic trees using either hybrid zonotope-based or Hamilton-Jacobi reachability analysis. Through their modularity, temporal logic trees ensure that when a connected vehicle's specification is changed, the verification result can be updated in real-time. Moreover, we show that when the temporal logic trees are constructed using Hamilton-Jacobi reachability analysis, we are able to efficiently synthesize specification-compliant control sets that contain the control inputs a vehicle can implement to ensure it satisfies its requirements. Using the synthesized control sets, we design a shared autonomy system that allows a remote operator to safely control a connected vehicle in cases where automation is insufficient. By leveraging this methodology, we develop a framework that allows a remote human operator to change a connected vehicle's driving specification, automate the vehicle to complete the updated specification, and even intervene on the vehicle's operation, all with guarantees that the vehicle will comply with the specification. We validate both the technological feasibility and benefits of the developed framework on a small-scale connected vehicle testbed enabled with a 5G cellular network.