move cursor to steer · scroll to drive
Servinar Cheema
Photo of Your Name

Servinar Cheema

PhD Candidate, Centre of AI for Assistive Autonomy · University of Edinburgh

Safety verification & validation of automated driving systems: scenario generation, behaviour rulebooks, and closed-loop evaluation.

Research

Safety Evaluation of Black-box Systems

Addressing safety evaluation frameworks for Automated Driving Systems (ADS) that produce traceable evidence, advancing from binary pass/fail or flattened scores. Integrating simulation, structured testing, and formal analysis to provide formal guarantees with observed, simulated behaviour. Focus areas include characterising system failure modes, defining Operational Design Domain (ODD) boundaries, and constructing objective safety cases for deployment readiness.

Behaviour Rulebooks and Accountability

Formalising normative driving behaviours and traffic regulations into machine evaluable rulebooks. This research translates high-level traffic laws into symbolic logic constraints and predicates for runtime monitoring and system verification. This framework enables objective fault attribution during multi-agent interactions, bridging the gap between legislative expectations and verifiable ADS planning.

Scenario Generation and Edge Cases

Developing generative models for scenario-based testing which systematically expose safety-critical edge cases and long-tail events. By applying adversarial generation and parameter exploration across road topologies, dynamic agents, and traffic rules, this work synthesises complex variations. The objective is to maximise systematic coverage and identify conditions where ADS kinematic or behavioural stability degrades.

Verification, Validation, and Closed-Loop Testing

Investigating ADS performance within closed-loop Verification and Validation (V&V) architectures. Assessing SOTA driving operating systems in generated environments, where the ego-vehicle's decisions continuously influence surrounding traffic. Core objectives include assessment of the V&V system against different driving system architectures.

Programmes

  1. PhD, Perception Action and Behaviour

    School of Informatics, University of Edinburgh · 2023 – present

    Safety Verification and Validation of Automated Driving Systems.

    Supervised by Prof. Subramanian Ramamoorthy & Dr. Craig Innes.

  2. MMath, Mathematics with Statistics

    School of Mathematics, University of Southampton · 2018 – 2022

    Machine Learning · Statistical Computing · Numerical Methods.

Contact