Generative World Models for Games
Presenting how to train spatio-temporal transformers from scratch and how to adapt and fine-tune text-image-to-video foundation models into action-conditioned world models.
🌴Biography
Hello, I’m Aidan, a researcher working at the intersection of machine learning, sequential decision-making, and embodied AI. My research is driven by the goal of building autonomous agents that can learn and generalize behaviours across a wide range of tasks. I’m particularly interested in methods and architectures for learning world models, and in understanding how agents can leverage them to solve new tasks efficiently.
Bio:
I am a Research Associate at the University of Edinburgh in The Bayesian and Neural Systems Group working with Amos Storkey and Peter Bell. Previously I was a Finnish Center for Artificial Intelligence postdoctoral researcher at Aalto University in Joni Pajarinen’s Robot Learning Lab and Arno Solin’s Machine Learning Research Group. I obtained my PhD from the University of Bristol under the supervision of Arthur Richards and Carl Henrik Ek. During my PhD I developed methods for controlling quadcopters in uncertain environments by synergising methods from probabilistic machine learning, stochastic differential geometry and reinforcement learning.
Interests
World models
Generative models
Reinforcement learning
Representation learning
Embodied AI
Robotics
Recent News
[05.10.25] Just released our technical report Generative World Modelling for Humanoids: 1X World Model Challenge Technical Report detailing our appraoch that won the 1X world model challenge.
[03.10.25] We’re top of the leaderboard in the 1X Humanoid World Model Challenge. See here for more details.
[18.05.25] New preprint (led by Yi Zhao) Efficient Reinforcement Learning by Guiding Generalist World Models with Non-Curated Data.
[26.02.25] New paper accepted to ICLR 2025 Workshop on World Models: Understanding, Modelling and Scaling (led by Yi Zhao) Generalist World Model Pre-Training for Efficient Reinforcement Learning.
[22.01.25] New paper accepted to ICLR 2025 - “Discrete Codebook World Models for Continuous Control”.
Recent Publications
(2025).
Generative World Modelling for Humanoids: 1X World Model Challenge Technical Report.
arXiv preprint arXiv:2510.07092.
(2025).
Efficient Reinforcement Learning by Guiding Generalist World Models with Non-Curated Data.
arXiv preprint arXiv:2502.19544v2.
(2025).
Generalist World Model Pre-Training for Efficient Reinforcement Learning.
ICLR 2025 Workshop on World Models: Understanding, Modelling and Scaling.
(2025).
Discrete Codebook World Models for Continuous Control.
In The Thirteenth International Conference on Learning Representations (ICLR).
(2025).
Entropy Regularized Task Representation Learning for Offline Meta-Reinforcement Learning.
In AAAI 2025".
Recent & Upcoming Talks
Generative World Modelling for Humanoids: 1X World Model Challenge
Presenting our methods for winning both tracks of the 1X world model challenge.
Huawei-Edinburgh Joint Lab: Discrete Codebook World Models
Aidan Scannell
•
•
1 min readProjects
1X World Model Challenge
Introduction World models equip agents (e.g., humanoid robots) with internal simulators of their environments. By “imagining” the consequences of their actions, agents can plan, …
Aidan Scannell
•
•
5 min read
Discrete Codebook World Models
In reinforcement learning (RL), world models serve as internal simulators, enabling agents to predict environment dynamics and future outcomes in order to make informed decisions. …
Aidan Scannell
•
•
1 min read
Implicitly Quantized Representations for Reinforcement Learning
Learning representations for reinforcement learning (RL) has shown much promise for continuous control. In this project, we investigate using vector quantization to prevent …
Aidan Scannell
•
•
1 min read