具有时序逻辑目标偏好的随机系统的启发式定性规划
Opportunistic Qualitative Planning in Stochastic Systems with Preferences over Temporal Logic Objectives
摘要:
当不能同时满足所有约束时,偏好在确定要满足哪些目标/约束方面起着关键作用。在这项工作中,我们研究了基于偏好的规划建模为马尔可夫决策过程的随机系统中,受到可能的不完全的偏好在时间上扩展的目标。我们的贡献有三个方面:首先,我们引入了一种偏好语言来指定对时间扩展目标的偏好。其次,我们定义了一个新的自动机理论模型来表示由给定的偏好关系诱导的前序。偏好的自动机表示使我们能够开发一个基于偏好的随机系统规划算法。最后,我们展示了如何合成机会主义策略,实现了一个结果,提高了当前的可满足的结果,具有积极的概率或概率1,在随机系统。我们说明我们的解决方案的方法,使用机器人运动规划的例子。
英文摘要:
Preferences play a key role in determining what goals/constraints to satisfy when not all constraints can be satisfied simultaneously. In this work, we study preference-based planning in a stochastic system modeled as a Markov decision process, subject to a possible incomplete preference over temporally extended goals. Our contributions are three folds: First, we introduce a preference language to specify preferences over temporally extended goals. Second, we define a novel automata-theoretic model to represent the preorder induced by given preference relation. The automata representation of preferences enables us to develop a preference-based planning algorithm for stochastic systems. Finally, we show how to synthesize opportunistic strategies that achieves an outcome that improves upon the current satisfiable outcome, with positive probability or with probability one, in a stochastic system. We illustrate our solution approaches using a robot motion planning example.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。