Master Thesis, 30 hp: Valid and Reachable Counterfactual Explanations for Reinforcement-Learning Policies in Pursuit–Evasion (AI)

Location: Linköping, Sweden

Company

hirify.global is a defense and security company developing advanced systems in aeronautics, weapons, command and control, sensors, and underwater systems.

What you will do

• Investigate whether minimal counterfactual state changes provide faithful, feasible, and stable explanations of reinforcement-learning agent decisions in a sequential pursuit–evasion environment.
• Implement a baseline method that identifies minimal state changes needed to alter an agent’s action.
• Implement and compare a dynamics-constrained method that generates feasible and reachable counterfactuals.
• Evaluate counterfactual explanations for action validity, reachability, stability, intervention faithfulness, and trajectory-level consequences.
• Develop and evaluate a user interface for visualizing counterfactuals through user studies.
• Deliver an XRL demonstrator and guidelines for using counterfactual explanations to understand autonomous decision-making.

Requirements

• Be in the final stage of a civil engineering programme in Computer Science and Engineering, Systems and Controls, Autonomous Systems, Robotics, or an AI and Machine Learning profile, or an equivalent programme.
• Be about to begin a 30 hp master thesis project.
• Have an interest in reinforcement learning, explainable AI, and interface design using visualization techniques.
• Have experience with PyTorch, TensorFlow, or JAX.
• Have a good understanding of Markov decision processes, policies, value functions, and deep reinforcement learning.
• Be able to work with an existing research codebase using Git.
• Have basic knowledge of statistics, including confidence intervals, multiple random seeds, hypothesis testing, and interpreting variance.
• Be interested in analysing why an agent behaves as it does.
• Pass security vetting under current security-protection regulations; additional citizenship obligations may apply where security clearance is required.

Culture & Benefits

• Work within hirify.global’s research project on unmanned collaborative aircraft systems.
• Collaborate with the Planning, Optimization and Reinforcement Learning unit through joint presentations and informal discussions with its RL evaluation and debugging team.
• Receive support and guidance while translating theoretical knowledge into practical solutions.
• The resulting XRL tool will be added to the unit’s toolbox for evaluating and debugging advanced AI techniques.

Hiring process

• Applications are reviewed on an ongoing basis, and the position may be filled before the closing date of 11 November 2026.
• Security vetting is required for the position.