Reinforcement Learning for Trustworthy and Explainable AI in Decision Support for the Aerospace domain
This project aims to develop a Trustworthy AI (TAI) demonstrator, specifically leveraging Explainable AI (XAI) techniques. The core challenge is to design, prototype, and evaluate an approach that is not only effective in performance, but also demonstrably and appropriately trustworthy.
More specifically, the objective is to explore opportunities for reinforcement learning within this space to address aerospace decision making problems. RL is a powerful paradigm in machine learning where an “agent” learns to make sequential decisions by interacting with an “environment.” The agent’s goal is to maximise a cumulative reward signal, discovering an optimal policy - a mapping from states to action - through trial-and-error. This approach is highly effective for complex decision-making problems, such as navigation, resource allocation, and control systems, which are common in the aerospace domain.
The challenge: Select a public domain simulation environment (e.g. Gridworld or similar AI Gym environment) that is representative of the decision-making problems in the aerospace domain. Then train Reinforcement Learning agents using the environment(s) and select existing algorithms as a benchmarking exercise. Develop a trustworthy AI / Explainable AI technique, aimed to build appropriate trust of the agent and its behaviour with users. Then evaluate the approach. Demonstrate how the technique could be applied to other cases (i.e. simulation environments, problems, etc).
Researchers: Gonzalo Cancio, Tom Heslin, Guoda Laurinaviciute, Sanjukta Bhattacharya and Noah Barker