Safe Artificial Intelligence Lab

The overarching aim of the Safe Artificial Intelligence Lab is to develop novel computational methods and tools for providing safety guarantees to a wide range of autonomous systems, including autonomous vehicles, robotic systems, and swarm systems.

We are particularly active in the following topics:

• Scalable methods and tools for the verification of neural networks, including CNNs, and RNNs.
• AI-based specification languages and logic-based verification methods for reasoning about agent-based systems.
• Safe reinforcement learning for agent-based systems.

Our work is guided by a passion for Artificial Intelligence and the belief that AI should be safe and secure for society to use.

We have a history of development and maintenance of open-source state-of-the-art toolkits for Safe AI and international collaboration both with academia and the industry.

We presently benefit from strong links with the Assured Autonomy DARPA program, the Centre for Doctoral Training in Safe and Trusted AI and the Royal Academy of Engineering