Learning-based robotic contact-rich manipulation for nuclear fusion remote maintenance operations

About the Project

3 year fully funded PhD linked to the UK Atomic Energy Authority

Future fusion power plants will require highly reliable, safe, and efficient remote maintenance, as direct human intervention will be infeasible due to radiation exposure, extreme temperatures, limited accessibility, and other environmental hazards. The UK Atomic Energy Authority (UKAEA) is at the forefront of developing such technologies, notably through the Remote Applications in Challenging Environments (RACE) facility. Future reactors will introduce tighter spatial constraints, higher component density, longer operational lifetimes, and more stringent safety requirements than current testbeds. Therefore, these tasks must be performed autonomously or semi-autonomously, and failures can lead to contamination spread, prolonged reactor downtime, or significant safety risks.

This research aims to support UKAEA’s mission by developing advanced robotic manipulation solutions capable of operating robustly in these challenging conditions. Robots operating in fusion environments must cope with cluttered, constrained, and unpredictable scenes, where objects are often irregularly shaped, composite, deformable, or entangled with neighbouring components. Tasks frequently involve close contact with the environment, uncertain object properties, and limited sensing due to occlusions or restricted viewpoints.

A fundamental technical bottleneck in these scenarios is contact-rich grasping and manipulation. Classical robotic manipulation methods typically assume isolated rigid objects and rely on simplified stability metrics, such as centre-of-mass alignment or force-closure conditions. While effective in structured industrial settings, these assumptions break down in cluttered fusion environments, leading to dropped objects, unintended collisions, or destabilisation of surrounding components. Learning-based approaches, including reinforcement learning (RL) and imitation learning, offer a promising alternative by allowing robots to learn manipulation strategies directly from interaction data. However, their application in safety-critical, cluttered environments remains limited by several factors: poor sample efficiency, limited generalisability across object types and configurations, sensitivity to unmodelled dynamics, and difficulties in guaranteeing safe behaviour.

Our prior work introduced the Grasp Affordance Manipulation (GAM) framework, which addressed some of these limitations by integrating affordance prediction into the learning loop. A manipulation affordance prediction model was used to estimate task success for candidate actions, while a manipulation-affordance-based grasp filter selected stable and task-relevant grasp poses. This approach demonstrated improved performance in simulation for challenging tasks such as object disentanglement.

Building on this foundation, the proposed PhD will develop a next-generation, generalisable GAM framework by combining world-dynamics reasoning, data-driven learning, and enhanced perception. The research will integrate learned dynamics models with affordance-driven RL and imitation learning to enable safer, more data-efficient manipulation. Particular emphasis will be placed on robust uncertainty-aware decision-making and failure recovery, ensuring reliable deployment in nuclear-relevant environments. The resulting framework will provide robots with the ability to reason about contacts, object interactions, and task outcomes, enabling robust manipulation in cluttered, hazardous settings.

The PhD research will be conducted jointly at the Robotics and Autonomous Intelligent Machines (RAIM) research group, School of Engineering, Cardiff University, and UKAEA’s RACE facility. Cardiff University will provide a strong academic environment for algorithm development, simulation studies, and foundational research, while RACE offers nuclear-relevant testbeds, operational expertise, and direct engagement with end users.

As a PhD student, you will receive comprehensive training and support. You will have access to state-of-the-art robotic facilities, including GPU workstations, robotic manipulators, mobile platforms, and advanced sensors such as LiDAR, depth cameras, IMUs, and GNSS. You will join a vibrant, interdisciplinary research group from both Cardiff and UKAEA, and benefit from supervision by experienced academic staff, early-career researchers, and industrial partners, who will provide mentoring, specialist training, and exposure to real-world fusion challenges.

RACE (Remote Applications in Challenging Environments) was founded in 2014 as part of the UK Atomic Energy Authority (UKAEA) fusion research and development programme - to create robots for operating in some of the most challenging environments imaginable.

UKAEA’s wider mission is to lead the commercial development of fusion power and related technology, and position the UK as a leader in sustainable energy. Based at Culham Campus near Oxford and at a new technology facility in South Yorkshire, UKAEA runs the UK’s fusion research programme and operated the Joint European Torus (JET) fusion experiment on behalf of scientists from 28 European countries, until its closure in 2023. Now UKAEA is responsible for decommissioning JET.

UKAEA is expanding its technology capabilities, to ensure the UK remains at the forefront of fusion research and development - providing key research support for the UK’s own future fusion powerplant - STEP Fusion.

Academic Criteria

Candidates should hold or expect to gain a first-class degree or a good 2.1 (or their equivalent) in Engineering or a related subject.

Desirable skills include robotics, Deep Learning, programming in Python, mechatronics, and related.

Applicants whose first language is not English will be required to demonstrate proficiency in the English language (IELTS 6.5 or equivalent)

Contact for further information

Please contact Dr Ze Ji (jiz1@cardiff.ac.uk) to informally discuss this opportunity

How to apply

Applicants should submit an application for postgraduate study via the Cardiff University webpages (http://www.cardiff.ac.uk/study/postgraduate/research/programmes/programme/engineering) including;

• an upload of your CV
• a personal statement/covering letter
• two references (applicants are recommended to have a third academic referee, if the two academic referees are within the same department/school)
• Current academic transcripts

Applicants should select Doctor of Philosophy (Engineering), with a start date October 2026.

In the research proposal section of your application, please specify the project title and supervisors of this project and copy the project description in the text box provided. In the funding section, please select "I will be applying for a scholarship / grant" and specify that you are applying for advertised funding, reference ZJ RACE 26

Deadline for applications

30th June 2026. We may however close this opportunity earlier if a suitable candidate is identified.

Funding Notes

Tuition fees at the Home rate are covered by the University (£5,006 in 2025/6) and an annual stipend equivalent to current Research Council rates is provided (£20,780 stipend for academic year 2025/6)

Eligibility Full funding is currently available to Home applicants only on a 3 years full time basis.