Model-Based Dynamic (Re-)Design of Reactive Control Agents

Problem
Proactive agents that plan ahead require world knowledge that often is not available, and, in any case, tend to quite slow. Reactive agents often get caught in local minima, and, more generally, provide few guarantess about the quality of solutions. An alternative framework is to use a functional model of reactive control to dynamically (re-)design the control structure. This may requires a lot less knowledge than proactive methods but may provide better guarantees than reactive ones.

Students will first read relevant papers and then design a model of a reactive agent (e.g., a light-following robot).