The reinforcement learning team works on both fundamental and applied AI research, with a particular focus on reinforcement learning. Reinforcement learning is the study of sequential decision making under uncertainty, and hence encompasses many problems, from robotic manipulation to coordinating multiple goal-based agents. Our aim is to improve our understanding of biological and artificial intelligence, and use these insights to develop AI systems that can be applied to complex tasks in the real world.
Brain-Computer Interface Robot Control
Brain-computer interfaces (BCIs) provide a direct connection between our nervous signals and computers, enabling humans with neurological/physiological disorders to interact with the world in ways that would otherwise be impossible. However, BCI inputs are challenging to work with, requiring sophisticated signal processing and machine learning techniques. Coupling BCIs with robots through the paradigm of shared autonomy would enable users to physically interact with the world. Unfortunately, operating robots in the real world (outside of highly-constrained factory settings) is challenging, requiring further state-of-the-art methods in artificial intelligence and control theory. We are currently working on integrating and developing research on all of these areas in order to achieve more generalisable BCI robot control.
Benchmarking Imitation Learning
Researchers have developed many imitation learning method over the last few years, each claiming state-of-the-art performance on various tasks. However, results can be easily influenced by factors outside of the claimed contributions of a method, such as data preprocessing, extensive hyperparameter tuning, or simply more computation. In A Pragmatic Look at Deep Imitation Learning, we broke down the contributions of many methods, and developed a unified framework for imitation learning methods, allowing us to perform a fair comparison between algorithms. We have also released an open source library for other researchers to research or apply imitation learning methods.
Procedural Content Generation for Space Engineers
Many video games rely on procedural content generation (PCG) in order to (semi-)autonomously create a large amount of game content. For some assets, such as vehicles, the generated content must be both functional and aesthetically-pleasing. In addition, the designer, or even player, typically wants a variety of options to choose from. To achieve this for the Space Engineers 3D sandbox game, we have been using a combination of techniques from evolutionary computation. In Evolving Spaceships with a Hybrid L-system Constrained Optimisation Evolutionary Algorithm, we developed a novel hybrid evolutionary algorithm to generate interesting and functional spaceships. We later improved upon this in Surrogate Infeasible Fitness Acquirement FI-2Pop for Procedural Content Generation, by developing a novel fitness function, and further applying it to the well-known MAP-Elites "quality-diversity" algorithm.