Action-Feasibility-Based Pragmatic Understanding of Ambiguous Instructions by Service Robots
Action-Feasibility-Based Pragmatic Understanding of Ambiguous Instructions by Service Robots
复制标题
基于行动可行性的服务机器人对模糊指令的务实理解
DOI：
10.1109/sii52469.2022.9708746
复制
发表时间：
2022
期刊：
影响因子：
--
通讯作者：
Taniguchi Tadahiro
中科院分区：
文献类型：
--
作者：
Naito Taichi;Hirota Naoya;Hagiwara Yoshinobu;Iwahashi Naoto;Taniguchi Tadahiro
本研究的目的是开发一种方法，使服务机器人不仅能够理解明确的用户指令，如“从厨房给我拿可乐来”，而且还能理解模糊的指令，如“我渴了”，并根据环境规划可行的动作序列。为了实现这一点，我们提出了一种方法，使机器人预测的目标描述从用户的模糊指令使用的指令理解模块，并利用行动规划模块的目标描述和环境信息的基础上规划一个可行的行动序列。通过在两个模块之间设置目标描述，可以使用Seq2Seq结合BERT来学习从模糊指令到目标描述的映射。此外，可执行的动作序列可以使用ProbLog基于目标描述和环境信息作为概率逻辑推理的结果来规划。我们进行了一个实验，以评估所提出的方法在理解用户的模糊指令和规划一个可行的动作序列的有效性。在实验中，我们比较了所提出的方法，它认为目标描述和环境信息，与直接预测的动作序列，从用户的指令的方法。此外，进行了实验，以确定所提出的方法的有效性方面的组合。实验结果表明，所提出的方法表现出更好的预测动作序列的可行性方面比比较方法。研究证实，所提出的方法使服务机器人预测合适的目标描述用户的模糊指令，并根据环境规划一个可行的行动序列。
The aim of this study is to develop a method that enables a service robot to understand not only explicit user instructions, such as "Bring me the coke from the kitchen," but also ambiguous instructions, such as "I ’m thirsty," and to plan a feasible action sequence according to the environment. To achieve this, we propose a method that enables the robot to predict a goal description from the user’s ambiguous instructions using an instruction understanding module, and to plan a feasible action sequence based on the goal description and environmental information utilizing an action planning module. By setting the goal description between the two modules, map-ping from an ambiguous instruction to a goal description can be learned using Seq2Seq combined with BERT. In addition, an executable action sequence can be planned using ProbLog based on the goal description and environmental information as the result of a probabilistic logic inference. We performed an experiment to evaluate the effectiveness of the proposed method in understanding the user’s ambiguous instructions and plan-ning a feasible action sequence. In the experiment, we compared the proposed method, which considers the goal description and environmental information, with a method that directly predicts an action sequence from the user’s instruction. In addition, an experiment was performed to determine the effectiveness of the proposed method in terms of compositionality. The experimental results demonstrated that the proposed method exhibited performed better in terms of the feasibility of the predicted action sequences than the comparison method. The study confirmed that the proposed method enables the service robot to predict suitable goal descriptions from the user’s ambiguous instructions and plan a feasible action sequence according to the environment.