Research on recognition-manipulation system of deformable object for humanoid robot with adaptive high-resolution vision and dynamic slipping sensors.

本研究的目的是通过硬件和软件方面的创新，包括多指手、高分辨率立体视觉、可变形物体建模、共享识别、操作规划和运动控制，了解不仅能够处理简单刚性物体（如杯子），还能够处理复杂和可变形的毛巾和衣服的识别-操作方法。除了在原研究计划中列出的可变形物体仿真、彩色深度图像传感器感知算法、毛巾抓取任务的操作动作控制等研究成果外，我们还研究了机器人从实验中学习的新型智能机制，包括动作描述和控制机制，用于故障检测和恢复。最后，这个研究的影响是如此之大，以至于有几个研究者受到这个主题的启发开始了布料操纵的研究。

The objective of this research is to understand recognition-manipulation methodology that is able to handle not only simple-rigid object such as a cup, but also complex and deformable towels and clothes through an innovation in both hardware and software aspects including multi-fingered hand, high-resolution stereo vision, deformable object modeling, share recognition, manipulation planning and motion control. In addition to the research results which are listed on the original research plan, including deformable object simulation, perception algorithm using color-depth image sensor, manipulation action control on towel grasp task, we have investigated new type of intelligence mechanism that the robot learns from experiments with action description and control mechanism for failure detection and recovery. Finally, influence of this research is so great that there are several researchers who started cloth manipulation research inspired by this subject.