Learning Heuristics with Graph Neural Networks for Numeric Planning: A Preliminary Study

We consider the problem of learning heuristics for numeric planning domains, using Graph Neural Networks. The problem has been approached multiple times, from different perspectives and with varying results for classical planning, but is relatively new for numeric planning. The goal is to extend the work proposed by Stalberg, ̇ Bonet, and Geffner [1] to handle numeric planning problems.