Evolving Initial Heuristic Functions for Agent-Centered Heuristic Search

摘要: Heuristic functions guide search algorithms and have a profound impact on their performance. In the context of agent-centered real-time heuristic (RTHS), represents agent’s initial domain knowledge which agent then updates as it explores graph. An ideal should capture some specific to effectively yet be general enough for broad class problems. It also computationally efficient, compact in its representation human-interpretable. Traditionally heuristics RTHS been designed by humans (e.g., Manhattan distance). this paper we explore alternative building machines. To keep them portable human-interpretable represent each closed-form algebraic formula. Yet make problem specifics thus more effective guiding search, automatically build tailored achieve both objectives, propose evaluate searching space functions. As preliminary demonstration, find that outperform distance grid-based pathfinding. We develop an insight how such formula-based are able exploit characteristics certain pathfinding maps.