PhD Studentship: Causal Reinforcement Learning

About the Project

Phaidra builds autonomous AI control systems for data centre and industrial infrastructure. We deploy reinforcement learning in production on some of the world's most complex physical systems. The hard, unsolved research problems are the same ones that matter in practice. This studentship is an opportunity to work on foundational RL research while staying grounded in real-world challenges.

Reinforcement Learning (RL) has emerged as a powerful framework for sequential decision-making. Yet a fundamental limitation remains: agents trained on historical data under fixed policies often exploit spurious correlations that break at deployment time, especially when the environment shifts or the new policy explores previously unseen regions of the state-action space.

This PhD project tackles that limitation by integrating causal reasoning into RL. Causal inference provides a formal language (causal graphs, interventional queries, counterfactuals) for distinguishing stable structural relationships from incidental correlations. The research will investigate how these tools can make RL agents more robust and generalizable, particularly in real-world industrial settings.

The project will proceed in three phases:

Theoretical Foundations: formalising policy learning from biased, small datasets through a causal lens; characterising how confounding and mediators affect offline RL.

Algorithm Development: building RL algorithms that leverage known or learned causal structure to improve out-of-distribution generalisation and provide policy guarantees.

Benchmarking & Evaluation: evaluating proposed methods on controlled simulated environments with known causal structure, benchmarked against standard and offline RL baselines.

Supervisors

Academic Supervisor: Prof. Alessandro Abate, Department of Engineering, University of Cambridge

Industrial Co-supervisors: Dr. Miguel Suau and Dr. Alec Edwards, Phaidra

The student will be based primarily at the University of Cambridge, with the opportunity to spend time at Phaidra.

Funding & Duration

This is a fully funded 4-year PhD studentship, expected to start January 2027, co-funded by Phaidra and administered by the University of Cambridge.

Who You Are

You are a curious and technically rigorous researcher who wants to work at the intersection of causal inference and sequential decision-making. You are excited by foundational questions with real-world stakes and want your PhD to contribute both to the academic literature and to the practical deployment of intelligent systems.

Key Qualifications

A first-class or upper second-class honours degree (or equivalent) in Computer Science, Mathematics, Engineering, Statistics, or a related technical field.

Strong background in at least one of: reinforcement learning, machine learning, probabilistic modelling, or control theory.

Proficiency in Python and standard ML libraries (PyTorch, NumPy, SciPy, scikit-learn).

Clear scientific writing skills and the ability to communicate research to both academic and applied audiences.

Eligibility to study at the University of Cambridge (international students welcome; English language requirements apply).

Preferred Skills & Experience

Familiarity with causal inference, causal graphical models, or structural equation models.

Prior research experience (undergraduate thesis, MSc dissertation, research internship, or publications).

Experience with offline RL, batch RL, or safe RL.

Exposure to applying ML to real-world physical or industrial systems.

How to Apply

There are two parallel steps, both required:

Apply through Phaidra's careers portal at https://job-boards.greenhouse.io/phaidra. You will be asked to submit a CV and a short cover letter describing your research interests and motivation.

Apply to the University of Cambridge through the postgraduate application portal for the PhD in Engineering programme. Name Prof. Alessandro Abate as your proposed supervisor and reference this studentship in your application. You are also encouraged to email Prof. Abate directly at aa2807@cam.ac.uk with your CV and a one-page statement of research interest.

Both applications must be submitted. We encourage you to apply as soon as possible. Applications close 30 July 2026.

The studentship is expected to start January 2027.

Interview Process

Technical and research discussion with an industrial supervisor (60 minutes)

Meeting with the academic supervisor (60 minutes)