Principles of Metareasoning for Everyday Activities ((07) P05*)

这个子项目研究特定推理方法和表示集的控制，以及系统内部查询回答框架的设计，以解决完成人类规模的日常操作任务的机器人代理的推理任务。它解决了研究问题，如不同的推理方法应该如何-以及何时-应该交换数据和结论，以及推理工作流在多大程度上是可学习的。

This subproject investigates the control of specific ensembles of reasoning methods and representations and the design of a system internal query answering framework for addressing the reasoning tasks of a robotic agent accomplishing human-scale everyday manipulation tasks. It addresses research questions such as how – and when – should different inference methods exchange data and conclusions, and to what extent are reasoning workflows learnable.