Hunt Lab

Effective Patrolling Behaviours

We want to deploy teams of robots into real-world environments to monitor activity to ensure anything anomalous can be detected and investigated. In a multi-robot / swarm context, there is a clear opportunity to exploit cooperation between robots to verify anomalous findings by sending follow-up patrols.

There is a class of ‘adversarial patrolling’ algorithms that have already been developed, but not benchmarked robustly so far. We are interested in using technologies such as Graph Neural Networks (GNNs) to control swarms.

We investigate patrolling behaviours in simulation (e.g. ROS ‘Patrol Sim’) before testing controllers on real robots.

Evolving Behavioural Plasticity

Based on our understanding of behavioural biology, some animals demonstrate phenotypic plasticity (flexibility) in their behavioural types, for example to be ‘bold’ or ‘shy’ in how they respond to risk. 

We are investigating whether and how behavioural plasticity in these fundamental ‘personality types’ can emerge in evolutionary swarm robotics. We are evolving controllers in simulated environments, and then plan to combine this with morphological (shape) plasticity in real robots. 

Behavioural Heterogeneity for Enhanced Collective Decision-Making

Making decisions collectively is an important capability for many animal groups, such as ant or honeybee colonies. Bioinspired collective decision-making algorithms have been studied in simulated and real robot swarms.

We are investigating whether and how heterogeneity in behavioural traits will improve collective decision-making in robot swarms, in the context of the distributed patrolling problem and deciding where anomalies have been found.