Several problems in planning and reactive synthesis can be reduced to the analysis of two-player quantitative graph games. {\em Optimization} is one form of analysis. We argue that in many cases it may be better to replace the optimization problem with the {\em satisficing problem}, where instead of searching for optimal solutions, the goal is to search for solutions that adhere to a given threshold bound. This work defines and investigates the satisficing problem on a two-player graph game with the discounted-sum cost model. We show that while the satisficing problem can be solved using numerical methods just like the optimization problem, this approach does not render compelling benefits over optimization. When the discount factor is, however, an integer, we present another approach to satisficing, which is purely based on automata methods. We show that this approach is algorithmically more performant -- both theoretically and empirically -- and demonstrates the broader applicability of satisficing overoptimization.
翻译:规划和被动合成方面的几个问题可以归结为对双玩量图游戏的分析。 ~ 优化} 是一种分析形式。 我们争论说, 在许多情况下, 最好用 ~ spatisficing 问题取代优化问题, 而不是寻找最佳解决方案, 目的是寻找符合给定阈值的解决方案。 这项工作定义并调查了使用折扣和成本模型的双玩图形游戏中的讽刺问题。 我们显示, 虽然使用像优化问题一样的数字方法可以解决讽刺性问题, 但这种方法不会给优化带来令人信服的好处。 但是, 当折扣系数是整数时, 我们提出另一种配方方法, 纯粹基于自动磁数据方法。 我们显示, 这种方法在理论上和实验上都更具有演化性, 并展示了对调的过度优化的更广泛适用性。