Industry studentships at the EPSRC CDT in Autonomous Intelligent Machines and Systems, University of Oxford

Industry studentships

The CDT offers a number of industry funded studentships. These are either partially or fully funded, and will be advertised here.

When applying to one of these studentships, please quote the reference number in the application form, and inform the CDT Administrator that you have applied for this studentship when submitting your application.

Details on how to apply can be found here: https://www.ox.ac.uk/admissions/graduate/courses/autonomous-intelligent-machines-and-systems

Fully Funded 4-year Doctoral Studentship

Joint with the Oxford Martin AI Governance Initiative and the EPSRC CDT in Autonomous Intelligent Machines & Systems (AIMS)

Note: This studentship is fully funded.

Supervisor(s): Professor Michael Osborne & Professor Robert Trager

Start Date: October 2026

Autonomous systems powered by artificial intelligence will have a transformative impact on economy, industry and society as a whole. Our mission is to train cohorts with both theoretical, practical and systems skills in autonomous systems - comprising machine learning, robotics, sensor systems and verification- and a deep understanding of the cross-disciplinary requirements of these domains. Industrial partnerships have been and will continue to be at the heart of AIMS, shaping its training and ensuring the delivery of Oxford’s world-leading research in autonomous systems to a wide variety of sectors, including smart health, transport, finance, energy and extreme environments. Given the broad importance of autonomous systems, AIMS provides training on the ethical, governance, economic and societal implications of autonomous systems. For more information regarding the AIMS programme, see our web pages at: aims.robots.ox.ac.uk.

Title: Advancing Technical AI Governance and Safety

Abstract: This DPhil project aims to address problems within the emerging fields of technical AI governance and safety, building upon the foundational work of Reuel et al. (2024) in their survey "Open Problems in Technical AI Governance." The research will focus on three critical areas: identifying key intervention points in AI development and deployment, assessing the efficacy of potential governance actions, and enhancing governance options through novel technical mechanisms. Specific technical challenges to be addressed include developing robust methods for detecting and mitigating data bias, creating verifiable computation techniques for AI model training, designing privacy-preserving methods for external auditing of AI systems, and formulating technical approaches to ensure AI safety throughout the development pipeline.

The project's expected outcomes include an expanded taxonomy of technical AI governance and safety challenges, novel algorithmic approaches for monitoring and enforcing AI safety standards, and case studies demonstrating the real-world application of these governance tools. By bridging the gap between technical AI development, safety research, and policy-making, this research aims to contribute significantly to the responsible advancement of AI technologies in society. The findings will provide valuable insights for AI researchers, safety experts, policymakers, and governance bodies, ultimately supporting the creation of more effective, safer, and technically informed AI governance frameworks.

This project will be affiliated with the Oxford Martin AI Governance Initiative, will be jointly supervised by a policy supervisor from the Initiative, and will contribute to the AIGI/Oxford safety and technical governance community.

References: Reuel, A., Bucknall, B., et al. (2024). Open Problems in Technical AI Governance, https://www.governance.ai/research-paper/open-problems-in-technical-ai-governance

Award Value

The studentship covers the full course fees plus a stipend (tax-free maintenance grant).

Eligibility

Prospective candidates will be judged according to how well they meet the following criteria:

· Applicants are normally expected to be predicted or have achieved a first-class or strong upper second-class undergraduate degree with honours (or equivalent international qualifications), as a minimum, in computer science, engineering, physics, mathematics, statistics or other related disciplines. A previous master's qualification is not required.

· Excellent English written and spoken communication skills

Candidates will also need to demonstrate a broad interest in some or all of the four AIMS themes:

· machine learning, as a unifying core
· robotics & vision
· cyber-physical systems (e.g. sensor networks)
· control & verification

The deadline for applying is Wednesday 28^th January 2026. Candidates are therefore recommended to apply as soon as possible to and to inform wendy.adams@eng.ox.ac.uk when they have done so.

If you have any technical questions about the DPhil Studentship, please email wendy.adams@eng.ox.ac.uk

Please quote AIMS-AIGI-2026 in the studentship reference box.

aims.robots.ox.ac.uk

------

Fully Funded 4-year Doctoral Studentship

Joint with LIGHTSPEED and the EPSRC CDT in Autonomous Intelligent Machines & Systems (AIMS)

Note: This studentship is fully funded.

Supervisor(s): Xiaowen Dong & Jakob Foerster

Start Date: October 2026

Title: LLM driven Multi-agent Simulation for Social Networking in multiplayer online game

Abstract

The persistence of live-service games is not merely a consequence of continuous content updates but is fundamentally driven by the complex, self-reinforcing dynamics of multi-sided network effects. In these ecosystems, diverse user types—including regular users, user-generated content (UGC) producers, key opinion leaders (KOLs), and high-value customers—interact synergistically, creating a virtuous cycle where increased participation enhances the value and retention for all. However, traditional approach to interaction system design and operational strategies, which reply heavily on practitioner experience and controlled A/B experimentation, lack the predictive capability to systematically model the formation and long-term evolution of these social networks. This gap leads to challenges in forecasting social structure evolution, identifying early risk signals (e.g., virtual economy imbalance, user churn), and assessing the cascading impacts of operational decisions.

This research addresses these limitations by introducing a novel paradigm: a large language model (LLM) driven multi-agent simulation platform designed to model and analyze the social behaviors underpinning multi-sided network effects in video games. The core technical contribution is the development of a "LLM based Social Behavior and Networking Simulator". Its architecture comprises several key modules: 1) LLM-powered Agent Modeling, where each user type is represented by an AI agent with memory, goals, and social tendencies; 2) a Game World Simulator that models economic, social, and content creation subsystems; 3) a Multi-Agent Communication Protocol to facilitate interactions like trade and collaboration; 4) a Social Evolution Engine that simulates behavioral changes over time through mechanisms like social learning; and 5) a Policy Intervention API, allowing operators to input strategies (e.g., season systems, UGC incentives) and observe their long-term, counterfactual outcomes.

The technical pathway is structured in three phases. Phase 1 focuses on constructing a "Minimum Viable Society" (MVS) within a simplified game environment to observe emergent social phenomena. Phase 2, "Social Structure Evolution," integrates social network graphs and graph neural networks (GNNs) to simulate complex structures like guild formation and influencer-driven economies, creating an agentic learning flywheel where agent behaviors are refined through large-scale reinforcement learning. Phase 3 culminates in a "Social Strategy Intervention Platform," generating a reusable library of validated strategies by quantifying the impact of various interventions on ecosystem health metrics.

This work is distinctly differentiated from existing multi-agent economic simulations, such as the AI Economist, by its foundation on large language models (LLMs). This enables the simulation of rich, human-like social behaviors—including mentorship, guild dynamics, and live-streaming interactions—and a comprehensive virtual economy encompassing NFTs and UGC, far exceeding the scope of simple labor and tax strategies. The project offers significant value across multiple domains. For game design, it shifts the paradigm from experience-based intuition to predictive, system-level optimization for long-term ecosystem sustainability. For AI research, it pioneers a new frontier in explainable large-model social simulation, bridging AI, sociology, and behavioral science. For the industry, it aims to establish a foundational infrastructure—a queryable strategy library—that can reduce trial-and-error costs pre-launch and provide data-driven, long-term operational guidance, ultimately promoting standardization in the analysis of multi-sided network effects for the gaming industry.

Award Value

The studentship covers the full course fees plus a stipend (tax-free maintenance grant).

Eligibility

Prospective candidates will be judged according to how well they meet the following criteria:

● Applicants are normally expected to be predicted or have achieved a first-class or strong upper second-class undergraduate degree with honours (or equivalent international qualifications), as a minimum, in computer science, engineering, physics, mathematics, statistics or other related disciplines. A previous master's qualification is not required.

● Excellent English written and spoken communication skills

Candidates will also need to demonstrate a broad interest in some or all of the four AIMS themes:

● machine learning, as a unifying core

● robotics & vision

● cyber-physical systems (e.g. sensor networks)

● control & verification

The deadline for applying is Wednesday 28^th January 2026. Candidates are therefore recommended to apply as soon as possible to and to inform wendy.adams@eng.ox.ac.uk when they have done so.

If you have any technical questions about the DPhil Studentship, please email wendy.adams@eng.ox.ac.uk

Please quote AIMS- LIGHTSPEED-2026 in the studentship reference box.

aims.robots.ox.ac.uk

-----------------

Fully Funded 4-year Doctoral Studentship/Amazon Fellowship

Joint with Amazon and the EPSRC CDT in Autonomous Intelligent Machines & Systems (AIMS)

Note: This studentship is fully funded.

Supervisor(s): Ingmar Posner and Andrea Vedaldi

Start Date: October 2026

Abstract

Robots that act effectively in the real world need to perceive, plan and anticipate, they need world models: internal representations that capture the entities, dynamics, and interactions that underpin their environment. A central challenge is how to map raw sensory input into the structured representations on which such models can operate. While recent work in object-centric learning has shown promise in extracting entities and relations from video, these approaches remain limited in realism, robustness, and utility, particularly in physically embodied robotics settings.

This PhD project will investigate how to learn object-centric world models that not only capture predictive dynamics but also parse complex real scenes into interpretable, robot-usable representations. Research questions include: (i) how to extract persistent, object-based abstractions from noisy video streams; (ii) how to couple these abstractions to generative world models that support prediction, counterfactual reasoning, and planning; (iii) incorporating physical priors and constraints to ensure robustness particularly in contact-rich manipulation settings, and (iv) how to evaluate and refine these models through embodied tasks such as manipulation, interaction with humans, and navigation in dynamic settings.

By bridging scene parsing with world modelling, this project will advance the foundations of robot perception and decision-making. The outcome will be representations that are simultaneously learned from video, structured around objects, and actionable in robotics, a step toward robust, general-purpose robot intelligence.

Award Value

The studentship covers the full course fees plus a stipend (tax-free maintenance grant).

Eligibility

Prospective candidates will be judged according to how well they meet the following criteria:

● Excellent English written and spoken communication skills

Candidates will also need to demonstrate a broad interest in some or all of the four AIMS themes:

● machine learning, as a unifying core

● robotics & vision

● cyber-physical systems (e.g. sensor networks)

● control & verification

The deadline for applying is Wednesday 28^th January 2026. Candidates are therefore recommended to apply as soon as possible to and to inform wendy.adams@eng.ox.ac.uk when they have done so.

If you have any technical questions about the DPhil Studentship, please email wendy.adams@eng.ox.ac.uk

Please quote AIMS-AMAZON-2026 in the studentship reference box.

aims.robots.ox.ac.uk