I will present two algorithms that are relevant to the themes of this workshop.
The first one concerns planning a minimal sequence of locations that have complete visual surveillance of the domain. I will present a greedy algorithm and its theoretical performance guarantees and bottleneck.
The second topic concerns the learning of a game involving planning paths of a group of agents that maintain visual connection to a set of moving targets.
One of the main challenges come from light-of-sight's non-local dependence on the geometries of the domain.
A common goal for the two algorithms is that they would generalize to new but similar domains. We will discuss important aspects of our learning models and how training data should be generated.