Momina Rizwan's PhD defence

Thesis title: Safety and Reliability for Autonomous Robots in Dynamic Environments

Author: Momina Rizwan, Department of Computer Science, Lund University

Faculty opponent: Professor Nico Hochgeschwender, Universität Bremen, Germany

Examination Committee:

• Professor Ulrik Pagh-Schultz Lundquist, University of Southern Denmark, Denmark
• Professor Federico Ciccozzi, Mälardalen University
• Assistant Professor Marie Farrell, University of Manchester, United Kingdom
• Deputy: Adjunct Professor Johan Eker, Lund University

Session chair: Professor Björn Regnell, Lund University

Supervisors:

• Senior Lecturer Christoph Reichenbach, Lund University
• Professor Volker Krueger, Lund University

Location: E:B, E-huset, Klas Anshelms väg 10/Ole Römers väg 3, Lund

Here is a link to download the thesis at LU Research Portal

Abstract

Autonomous robots must operate reliably and safely under uncertain, dynamic conditions over extended periods.  To ensure such operational robustness, it is vital that both developers and operators can clearly and verifiably specify functional requirements and safety constraints throughout the robot software lifecycle. My research targets different layers of robot operational safety: early error detection, real-time safety enforcement, and adaptive failure recovery. First, we extend the DeROS language to develop ROSSMARie, a DSL to generate a runtime safety monitor for enforcing safety rules and enabling autonomous recovery. ROSSMARie ensures functional safety through real-time rule monitoring and resume-capable interventions, validated on an industrial robot control platform in scenarios involving human proximity, terrain hazards, and contact instability. Second, we present EzSkiROS: an embedded DSL framework in Python that supports early fault detection during the pre-deployment (launch) phase of robotic skills. This DSL checks the consistency between Behavior Tree (BT) implementations, high-level symbolic skill contracts, and ontology-based world models. By performing symbolic and dynamic checks before execution, this approach identifies latent faults that would otherwise manifest at runtime. Third, we introduce a safety monitoring architecture Reflex-Plan, which enables communication between the runtime safety monitor and the deliberate high-level planner. This dual-layer design enables ”fast thinking” for immediate hazard response and ”slow thinking” for recovery planning. Reflex-Plan is validated in a mock hospital environment using a mobile manipulator, demonstrating measurable improvements in task continuity, response latency, and hazard mitigation. Together, these contributions form a safety pipeline that uses DSL-based robotic programming. Our results demonstrate significant improvements in operational safety and code maintainability, enabling autonomous robots to handle failures proactively and recover adaptively in complex, real-world settings.