Verbal instructing of sensor-based robots (VerbBot)

机器人的编程是其启动和在各种应用(例如，工业、实验室、服务)中进一步使用的关键步骤。机器人程序主要由具有相关参数(例如，目标位置、力值)的指令序列(例如，机器人运动、工具命令)和用于控制程序序列(例如，重复、分支)的控制结构组成。创建这样的项目是耗时的，需要专业知识。更糟糕的是编程，如果另外必须评估传感器数据以感知当前环境状况(例如，工件位置、接触力)以完成任务。即使是高级编程概念，如图形编程或演示编程，在这里也只能起到有限的帮助。因此，基于传感器的机器人系统的编程是它们在当前和未来的应用(例如家庭、工作车间、娱乐)中高效和灵活使用的瓶颈。本项目的愿景是通过自然语言对基于传感器的机器人进行直观的编程。作为第一步，这里用人类语言中的动词(例如，抬起或搬运)替换通常的机器人命令(例如，moveStraight、Open Grigper)。为此，首先必须更详细地理解动词的语义，以确保对机器人的技术支持。(例如，推和推之间的区别是什么？)此外，要用作机器人命令的动词必须提供参数。(例如，您想要将工件推到哪里？)它还必须从情况中提取额外的、通常只能隐含地获得的上下文信息。(例如，要推的物体的重量是多少？)如果有疑问，向用户询问或采取主动行动可能会有所帮助。最后，指令将基于传感器执行，任何错误情况都需要纠正(可能在对话中)。

The programming of robots is a key step in their start-up and further use in a variety of applications (e.g., industrial, laboratory, services). A robot program consists mainly of a sequence of instruc-tions (e.g., robot movements, tool commands) with the associated parameters (e.g., target position, force values) and a control structure for controlling the program sequence (e.g., repetitions, branches). The creation of such programs is time consuming and requires expertise. Even worse is the programming, if additionally sensor data must be evaluated to perceive the current environmen-tal situation (e.g., workpiece position, contact forces) to fulfill the task. Even advanced program-ming concepts, such as Graphical Programming or Programming By Demonstration, are only of limited help here. Programming of sensor-based robot systems thus represents a bottleneck for their efficient and flexible use in present and in future applications represent (e.g., household, work-shop, entertainment).The vision of this project is the intuitive programming of sensor-based robots through natural lan-guage. As a first step, here the usual robot commands (e.g., moveStraight, openGripper) are re-placed by verbs from the human language (e.g., TO LIFT or TO CARRY). For this purpose, first the semantic meaning of the verbs must be understood in more detail, to ensure that technical sup-port for the robot. (What is, e.g., the difference between TO PUSH and TO SHOVE?) Also a verb, to be used as a command for the robot, must be provided with parameters. (Whereto, e.g., do you want a workpiece to be pushed?) It must also be extracted additional, often only implicitly available context information from the situation. (What is, e.g., the weight of the object to be shoved?) In case of doubts, queries to the user or proactive actions may help. Finally, the instruction is to be executed sensor-based and any error situations need to be corrected (possibly in dialogue).