Research on development of task primitives for robots using programming by demonstration in virtual reality

点击翻译按钮获取中文摘要

Robotics research has been applied to various fields recently. Various motions and tasks are realized with various robots including humanoids. Robot programs are different from ordinary programs because some of them(i.e., programs for tasks) are made to achieve goal states for the tasks in the real world. These task programs pose a different "reusability problem" for equivalent execution of a task. The programs for those robots are developed by human programmers, which are developed by trial and error using their experience and knowledge. When other people apply such a program to a different robot and a different environment, it may cause difficulty. Since such a program has an unknown part for others than its programmer in order to reuse and adjust the program, it is difficult for others to modify the program. This problem is concerned with not only the difficulty of commentating appropriately for the program but also the essential difficulty of porting the program using knowledge of a specific robot body in a specific environment to other programs for other robot bodies in other specific environments. In order to deal with this problem, it is important to divide a part of programs that depends on the bodies of the specific robots and the specific environments, extract the part of programs that can be adjusted as parameters, and to determine objective methods of their adjustment. It is also important to visualize applicability of such programs in order to illustrate limits of the applicability of the programs for programmers. In this research, we develop a simulator of state transitions in a task using the pseudo contact point method, which is used for detection of contact states transitions in fine motion, and an example of determining a parameter of a program(primitive) for state transition detection systematically using random search technique.