【摘要】 在开放世界中求解智能规划问题往往是比较困难的,这是由于在开放世界中,某些对象可能是未知的,因而在搜索规划解时需要考虑不同的可能性.一种解决的方法是使用传感器观察未知的对象,而该方法使用的前提是传感器能够保证获取规划所需的所有信息.与以往工作不同的是,本文考虑利用外部人士(Crowd)求解规划问题.假设存在一些外部人士可以为开放世界中某个规划问题提供必要的信息,然而在实际情况下,某些外部人士提供的信息可能是具有欺骗性的,如何使用此类信息求解规划问题是本文关注的重点.针对此类问题,本文提出了一个新颖的求解方法,首先获取一个求解开放世界下的规划问题所需的带有变量的命题公式集合,然后根据外部人士对命题公式的标注估计出变量所取的值,从而将开放世界中的规划问题转化为一般的规划问题求解.最后通过实验验证了该算法的有效性.更多还原

【Abstract】 Plan synthesis in an open world is challenging,since some objects in an open world might be unknown,then we need to consider various scenarios before planning. One way to solve this problem is to employ sensors to observe unknown objects,assuming the sensors are capable of correctly capturing all information needed for planning. Different with previous work,we turn to the crowd for help before doing planning. We assume there are abundant annotators available to provide information needed before planning,however there is possibly a substantial amount of discrepancy from the crowd in practice. It is thus challenging to solve the planning problem with possibly noisy information provided by the crowd. We propose a novel approach with two phases. We first build a set of propositions with variables,and collect values from crowd for those propositions. We then estimate the actual values of variables and transform the problem in an open world into a normal planning problem and solve it. Finally,we empirically exhibit the effectiveness of our approach.更多还原