Abstract

Autonomous vehicles (AVs) are increasingly deployed in complex and safety-critical environments, raising fundamental questions about how ethical principles should be operationalized within real-time control systems. While existing research has extensively addressed ethical decision-making at the planning and behavioral layers, ethical intent is commonly translated into predefined trajectories and delegated to low-level controllers under the assumption of safe execution. This separation introduces a critical gap between ethical reasoning and physical feasibility, particularly in scenarios involving nonlinear vehicle dynamics, actuation limits, constrained perception horizons, and time-critical interactions with vulnerable road users. This thesis proposes an Ethical Model Predictive Control (E-MPC) framework that embeds ethical reasoning directly into the motion-control layer of autonomous vehicles. Ethical principles are formalized through a conservative deontological rule set and mapped to spatial motion envelopes that explicitly constrain the admissible trajectory space. A rule-based scenario classification mechanism distinguishes between normal driving, avoidable accident, and unavoidable accident contexts. Based on the active ethical context, the MPC objective function adapts its weighting structure to balance nominal tracking performance, ethical constraint satisfaction, and harm minimization under dynamic feasibility constraints. By enforcing ethical compliance at the optimization level, the proposed framework ensures that ethical intent is preserved under nonlinear vehicle dynamics and actuator limitations, without relying on post hoc arbitration or predefined maneuver selection. The proposed E-MPC is implemented using a nonlinear receding-horizon MPC formulation and integrated with ROS 2. The framework is evaluated in the CARLA simulation environment across a comprehensive set of scenarios, including nominal driving on curved roads, ethically motivated lane-change maneuvers, and unavoidable accident situations with constrained perception and limited stopping distance. Each scenario is executed multiple times under identical initial conditions to assess repeatability and robustness. In normal driving, the controller achieves sub-decimeter lateral tracking accuracy, with an average cross-track error of approximately 0.036 m and smooth control behavior characterized by average steering rates below 10−3 rad/s. In avoidable accident scenarios, the E-MPC consistently selects ethically admissible adjacent-lane envelopes and executes controlled lateral shifts of approximately 3.4 m while maintaining bounded steering and yaw-rate activity. In unavoidable accident scenarios with limited detection ranges (10–20 m), the controller reliably transitions to emergency braking, avoiding collisions at speeds up to 30 kph when feasible and reducing impact speeds to below 15 kph when collisions cannot be avoided. Overall, this work demonstrates that ethical reasoning can be embedded directly into real-time vehicle control without compromising stability or dynamic feasibility. By treating ethics as a constrained optimization problem rather than a symbolic decision layer, the proposed E-MPC provides a physically grounded, interpretable, and scalable approach to ethical autonomous driving under real-world constraints.

School

School of Sciences and Engineering

Department

Robotics, Control & Smart Systems Program

Degree Name

MS in Robotics, Control and Smart Systems

Graduation Date

Fall 2-4-2026

Submission Date

2-12-2026

First Advisor

Amr El-mougy

Committee Member 1

Dr. Mervat Abou El Kheir

Committee Member 2

Dr. Seif Eldawlatly

Committee Member 3

Dr. Alia El Bolock

Extent

105 p.

Document Type

Master's Thesis

Institutional Review Board (IRB) Approval

Not necessary for this item

Disclosure of AI Use

Thesis text drafting; Thesis editing and/or reviewing

Download

Share

COinS