Multi-agent systems
We develop logics, model checking techniques and tools for systems of interacting agents, covering temporal, epistemic and strategic properties: what agents know, and what they can achieve alone or together. This includes the model checker MCMAS and its extensions, parameterised verification of swarms with any number of agents, strategy synthesis, and safe coordination of multiple agents.
We also verify agents whose behaviour is given by neural networks, such as recurrent and memoryful agents acting in partially observable, non-deterministic environments. A network verified on its own is under Robust and verifiable machine learning; an agent and its environment verified together are here.


