Resource Registry: Reinforcement Learning

This research project explores the application of AI and machine learning technqiues in video games, focusing on procedural content generation and AI-driven advice. The research team (RAIL) aims to dynamically generate diverse game content and create intelligent virtual companions that offer personalised guidance and strategies to enhance player experiences. By integrating AI technologies, the research team strives to revolutionise video game desgin and contribte to the advancement of AI in the gaming industry.

In this line of work, the research team is interested in techniques that allow an agent to quickly leverage past knowledge to solve new tasks. In particular, the research team focuses on how agents can acquire behaviours that can be combined to generate interesting, novel abilities. One particular focus is applying Boolean operators to learned behaviours to generate probably optimal solutions to new tasks. These approaches are not only human-understandable but also result in a combinatorial explosion in an agent’s abilities, which is key to tackling the multitasking or lifelong learning setting.

Here, the RAIL research team focuses on learning abstract representations, which the team believes is an important component if we are ever to apply reinforcement learning to the real world. In particular, the RAIL research team focuses on skill- and symbol discovery, as well as the interplay between the two. The RAIL research team have applied their approaches to challenging pixel-based tasks that require high-level planning and has shown that symbolic representations can be learned directly from low-level sensor data.