Biography
Prof. Ingmar Posner leads the Applied Artificial Intelligence Lab (A2I) at the University of Oxford. A founding director of the Oxford Robotics Institute (ORI), Ingmar’s career is shaped by a vision to create machines that constantly improve through experience, with a core research focus on representation learning and structured world models for robotics and beyond. He co-led the team behind the UK’s first autonomous car, the Oxford RobotCar (now part of the permanent collection of the London Science Museum), and is a co-founder of Oxa, an award-winning provider of autonomous vehicle software. In 2023, the ORI was awarded a Queen’s Anniversary Prize, the highest national honour for UK educational institutions. An ELLIS Fellow, Ingmar currently serves as an Amazon Scholar.
Research Interests
Prof. Posner's research focus spans the algorithmic foundations and real-world applications of AI and reflects his long-standing interest in how intelligent agents can robustly perceive, ground, and interact with complex environments. His work has produced seminal contributions to robot learning, spanning large-scale learning from demonstration, unsupervised learning of scene dynamics, object-centric generative modelling, and 3D object detection. It has been recognised with multiple Best Paper Awards at leading international venues such as ICAPS, IROS, and ISER.
Central to his group's research is the development of structured representation learning and world modelling. Rather than relying on purely task-specific learning, this paradigm is designed to enable machines to capture re-usable, interpretable representations of the world's underlying structure and dynamics. By moving beyond traditional perception boundaries, these models help autonomous systems, and increasingly scientists, reason about complex environments, improving generalisation and interpretability across diverse domains ranging from robot manipulation to the natural sciences.
Recent Publications
Compact Symbolic Representation Learning for Efficient Planning from Unsupervised Interaction
Geiss H-J, Hollenstein J, Agostini A, Posner I & Piater J (2027), 16944, 503-521
BibTeX
@misc{compactsymbolic-2027/,
title={Compact Symbolic Representation Learning for Efficient Planning from Unsupervised Interaction},
author={Geiss H-J, Hollenstein J, Agostini A, Posner I & Piater J},
year = "2027"
}
Coherent Off-Policy Improvement of Large Behavior Models with Learned Rewards
Scherer C, Watson J, Gruner T, Palenicek D, Posner I et al. (2026)
BibTeX
@misc{coherentoffpoli-2026/6,
title={Coherent Off-Policy Improvement of Large Behavior Models with Learned Rewards},
author={Scherer C, Watson J, Gruner T, Palenicek D, Posner I et al.},
year = "2026"
}
No Epoch Like the Present: Robust Climate Emulation Requires Out-of-Distribution Generalisation
Stanley-Clamp B, Lei A, Christensen HM & Posner I (2026)
BibTeX
@misc{noepochlikethep-2026/5,
title={No Epoch Like the Present: Robust Climate Emulation Requires Out-of-Distribution Generalisation},
author={Stanley-Clamp B, Lei A, Christensen HM & Posner I},
year = "2026"
}
XQCfD: Accelerating Fast Actor-Critic Algorithms with Prior Data and Prior Policies
Palenicek D, Vogt F, Watson J, Posner I, Kragic D et al. (2026)
BibTeX
@misc{xqcfdaccelerati-2026/5,
title={XQCfD: Accelerating Fast Actor-Critic Algorithms with Prior Data and Prior Policies},
author={Palenicek D, Vogt F, Watson J, Posner I, Kragic D et al.},
year = "2026"
}
Posterior Sampling Reinforcement Learning with Gaussian Processes for Continuous Control: Sublinear Regret Bounds for Unbounded State Spaces
Flynn H, Watson J, Posner I & Peters J (2026)
BibTeX
@misc{posteriorsampli-2026/3,
title={Posterior Sampling Reinforcement Learning with Gaussian Processes for Continuous Control: Sublinear Regret Bounds for Unbounded State Spaces},
author={Flynn H, Watson J, Posner I & Peters J},
year = "2026"
}
Disentangling Dynamical Systems: Causal Representation Learning Meets Local Sparse Attention
Baumgartner MW, Lei A, Watson J & Posner I (2026), Proceedings of Machine Learning Research, 323, 119-165
BibTeX
@inproceedings{disentanglingdy-2026/1,
title={Disentangling Dynamical Systems: Causal Representation Learning Meets Local Sparse Attention},
author={Baumgartner MW, Lei A, Watson J & Posner I},
pages={119-165},
year = "2026"
}
An End-to-End Framework for Modeling Pneumatic Soft Robots Based on Differentiable Finite Element Methods
Zhong S, Yao Y, Maiolino P & Posner I (2025), IEEE Robotics and Automation Letters, 10(12), 13003-13010
BibTeX
@article{anendtoendframe-2025/12,
title={An End-to-End Framework for Modeling Pneumatic Soft Robots Based on Differentiable Finite Element Methods},
author={Zhong S, Yao Y, Maiolino P & Posner I},
journal={IEEE Robotics and Automation Letters},
volume={10},
pages={13003-13010},
publisher={Institute of Electrical and Electronics Engineers (IEEE)},
year = "2025"
}
LUMOS: language-conditioned imitation learning with world models
Demoss B, Hawes N & Posner H (2025), 2025 IEEE International Conference on Robotics and Automation (ICRA), 8219-8225
BibTeX
@inproceedings{lumoslanguageco-2025/9,
title={LUMOS: language-conditioned imitation learning with world models},
author={Demoss B, Hawes N & Posner H},
booktitle={IEEE Robotics and Automation Society (RAS 2025)},
pages={8219-8225},
year = "2025"
}
The Complexity Dynamics of Grokking
DeMoss B, Sapora S, Foerster J, Hawes N & Posner I (2025), Physica D Nonlinear Phenomena, 482, 134859-134859
BibTeX
@article{thecomplexitydy-2025/8,
title={The Complexity Dynamics of Grokking},
author={DeMoss B, Sapora S, Foerster J, Hawes N & Posner I},
journal={Physica D Nonlinear Phenomena},
volume={482},
number={134859},
pages={134859-134859},
publisher={Elsevier},
year = "2025"
}
Enhancing Joint Human-AI Inference in Robot Missions: A Confidence-Based Approach
Nguyen D-A, Colombatto C, Fleming S, Posner I, Hawes N et al. (2025)
BibTeX
@misc{enhancingjointh-2025/8,
title={Enhancing Joint Human-AI Inference in Robot Missions: A Confidence-Based Approach},
author={Nguyen D-A, Colombatto C, Fleming S, Posner I, Hawes N et al.},
year = "2025"
}
