PhD Studentship: Efficient Long-Horizon Task Execution in Physical AI (Deep learning, Computer Vision, Robotics)

PhD Studentship: Efficient Long-Horizon Task Execution in Physical AI (deep learning, computer vision, robotics)

Number of awards: 1

Award information: Fully funded PhD studentship covering Home tuition fees and a tax-free stipend. Available to Home fee-status applicants only due to funding structure. Delivered in collaboration with Intel, including industrial co-supervision. The student will work with Durham University academic supervisors and an Intel co-supervisor on research in Physical AI, edge AI and intelligent autonomous systems.

Closing date: 15 August

Overview
AI is shifting from passive prediction to systems that reason, plan, interact and act in the physical world. This PhD addresses efficient long-horizon task execution in Physical AI—complex tasks needing sequences of decisions, subgoals, corrections and adaptations over time. Robots must interpret instructions, decompose them, navigate dynamic environments, manipulate objects, detect failures and replan while tracking overall goals.

The core challenge is enabling high-level reasoning for long-horizon tasks without compromising the speed, reliability and efficiency required for real deployment. Current systems often reason effectively but act slowly, or act quickly but lack robust planning. Frontier models typically depend on heavy compute, cloud inference or controlled settings, limiting real-world use under compute, latency and energy constraints.

The project will explore orchestrating reasoning models, vision-language models, world models and efficient control policies for reliable robot behaviour. A key focus is deciding when to deliberate versus act reactively, using reasoning only when the gain in success justifies the cost in time, energy and compute.

Research directions may include long-horizon planning with monitoring and recovery; efficient hybrid reasoning; vision-language task decomposition; world models for prediction and planning; reactive-deliberative architectures; compute-aware evaluation; edge/cloud orchestration; and real-robot testing. Work could involve new algorithms, reasoning pipelines, foundation model evaluation, benchmarks, platform integration and practical deployment limits. It suits candidates interested in machine learning, computer vision, robotics, efficient inference or embodied AI, prioritising practical, reliable systems.

Durham offers strong facilities: Bede HPC (128 GPU), GPU cluster (90+ GPU), LiDAR, RADAR, drones, cameras, embedded devices and robots (including Unitree G1 humanoid, quadrupeds, UGVs and aerial platforms). Intel collaboration provides industrial insight into edge AI and deployable Physical AI.

Supervision: Dr Amir Atapour-Abarghouei (Durham), Prof Toby Breckon (Durham) and Dr Samet Akcay (Intel Principal Engineer, Edge Computing). The team publishes in top venues (CVPR, ICCV, ECCV, ICML). Students receive regular guidance, research training, publication support, hardware/compute access and Intel engagement opportunities.

Entry requirements: Relevant undergraduate or master’s degree in computer science, AI, engineering, mathematics, physics or related field; strong programming skills; interest in ML, computer vision, robotics, embodied AI or autonomous systems; motivation for independent research and high-quality publications; ability to combine theory, implementation and evaluation; and meet Durham’s English language requirements. Experience in robotics, RL, foundation/vision-language models or efficient inference is desirable but not essential.

Durham University is a leading Russell Group institution in a historic North East England city with excellent quality of life. Intel leads in computing innovation for efficient, real-world edge AI and robotics.

How to apply
Email your CV, transcripts and supporting documents to Dr Amir Atapour-Abarghouei at amir.atapour-abarghouei@durham.ac.uk for an initial discussion. Applications must be sent before 15 August. Interviews will follow shortly after for a proposed October 2026 PhD start date.

Advert information