Efficient Physical AI for Real-World Robotics – Funded PhD StudentshipUniversity

Fully funded PhD supported by Intel

[Covers Home tuition fees and a tax-free stipend. Open only to Home fee-status applicants. Co-supervised by Durham academics and an Intel co-supervisor. State your UK Home eligibility in all communications and applications.]

Physical AI moves beyond analysis to systems that perceive, reason, learn and act in the world, combining foundation models, computer vision, machine learning and robotics. Deployment is limited by dependence on very large models, extensive data, costly compute, cloud inference and large-scale robot data collection, while robots must run continuously under constraints on compute, memory, latency, power and training data.

This PhD will develop capable, efficient, deployable Physical AI for real robots, shaped to the student’s strengths. Topics may include: robot foundation and vision-language-action models; efficient reasoning and planning for long-horizon, real-time tasks; data-efficient learning with fewer demonstrations and less physical interaction; reinforcement and continual learning for adaptation and recovery; world models and predictive learning; multimodal perception; learning from simulation, synthetic data, human video, demonstrations and unstructured experience; compact models (compression, distillation, efficient inference); edge AI on embedded hardware; hybrid intelligence linking reactive control with higher-level reasoning; and the gap between simulation, benchmarks and physical platforms.

Research may cover new algorithms, foundation-model evaluation, learning pipelines, benchmarks, robot integration, or computational and practical limits. Focus may be machine learning, computer vision, reinforcement learning, robotics, embodied AI, efficient inference, or a mix. The aim is capable, efficient, adaptive, reliable, practical systems.

Durham hosts Bede (UK regional supercomputer), a departmental GPU cluster, and labs with LiDAR, RADAR, EEG, drones, cameras, embedded devices and robots, such as the Unitree G1 humanoid, Unitree Go2 quadruped, unmanned ground vehicles and aerial platforms. Intel collaboration targets edge AI, compact models, efficient deployment, intelligent sensing and Physical AI.

Supervisors: Dr Amir Atapour-Abarghouei and Prof Toby Breckon (Durham); Dr Samet Akcay, Principal Engineer, Edge Computing, Intel Client Computing and Physical AI. Expertise spans computer vision, machine learning, robotics, deep reinforcement learning, autonomous systems and efficient AI, with papers in CVPR, ICCV, ECCV, ICML and related venues.

Training covers regular supervision, help building an independent profile, training in writing, publication and presentation, access to GPUs, robots and sensors, and contact with Intel. Style: independent, problem-driven research in a supportive setting.

Durham is a Russell Group university in historic North East England, with a collegiate system, strong community and affordable living costs. Intel’s Edge Computing team brings efficient AI and robotics intelligence to devices outside the cloud.

Applicants need a relevant undergraduate or master’s degree (computer science, AI, engineering, mathematics, physics or related); strong programming; interest in machine learning, computer vision, robotics, embodied AI, foundation models or autonomous systems; motivation to research and publish; ability to link theory, implementation and evaluation; and Durham’s English language requirements. Robotics, reinforcement learning, foundation or vision-language models, efficient inference or embedded AI experience is useful, not essential.

Email CV, transcripts and supporting documents to Dr Amir Atapour-Abarghouei ([email protected]). Applications are considered on a rolling basis until the post is filled.

Subjects: Computer Science, Artificial Intelligence, Mathematics, Machine Learning, Computer Vision, Robotics, Engineering.

Covers Home tuition fees and a tax-free stipend.