Home services by a mobile manipulator system 移動マニピュレータシステムによるホームサービスに関する研究

博士論文

2011

In recent years, there has been considerable research performed in the area of mobile manipulation in home environments. The lack of accurate environment modeling and the uncertainty inherent to sensing technology avoid the direct application of approaches that have been classically adopted in more structured scenarios. As such, autonomous mobile manipulation in human environments is challenging and requires perceptual and manipulation skills which are robust against sparse, incomplete and noisy information. The main objective of this thesis is to explore new strategies for addressing the mobile manipulation challenges in home environments with an emphasis on the intersection of perception and control to create flexible robust systems. In this thesis, a general purpose mobile manipulation platform is first constructed to enable the exploration of new methods for perception and control in home environments. The developed platform characterizes both redundant kinematic degrees of freedom and a comprehensive sensor suite. Thus, it provides an adequate level of competency to address various mobile manipulation challenges, where both adequate sensorimotor capabilities and acquisition and fusion of a variety of exteroceptive information are required. Next, a novel RFID-based sensor fusion approach is proposed in our mobile manipulator system for home services, to realize robust and fast recognition and mobile manipulation of various objects. Our philosophy is to manipulate a wide range of objects of different shapes and sizes, as well as materials by utilizing all information including the geometrical, physical and additional information, to realize robust and efficient manipulation of various objects. As a case study, the mobile manipulation of various chairs by the mobile manipulator system is described. The effectiveness of the proposed methods for recognition and mobile manipulation has been shown by experimental demonstrations. To the best of our knowledge, this is the first implementation of its kind where mobile manipulations of various complexities are realized through an RFID-based sensor fusion system. Moreover, exploration of high speed tactile sensors for mobile manipulation in cooperation with humans is for the first time proposed and experimentally demonstrated. This opens the doors for a new class of physical human robot interaction techniques in the sense that they provide both reactive control systems and natural interaction patterns through advanced perception concepts. Tactile sensing-based reactive controllers have been proposed to realize safe physical human-robot interaction during the stand-up motion support. A novel tactile sensing-based fuzzy trajectory generation method for stand-up motion support is presented, which results in a natural style of interaction and is a promising approach for applications where service robots will have direct physical contact with the assisted people. Finally, an adaptive and robust control strategy is presented for simultaneous tracking tasks fulfilled by a wheeled mobile manipulator with its suitable reduced dynamic model. The proposed controller has the capability of compensating for unmodelled dynamic uncertainties and additive external disturbances, guaranteeing the states of the system asymptotically converge to the desired trajectory while ensuring the boundedness of the constrained force with controllable bounds, and being easily implemented in wheeled mobile manipulator systems. This is a very useful feature for applications. Asymptotic stability in the Lyapunov sense is acquired and rigorously proven, and simulation studies are presented to illustrate effectiveness of the developed method. As a result, the proposed controller is capable of simultaneous motion and constraint force tracking in constrained mobile manipulation tasks and is especially suitable for applications in home environments where unmodelled dynamic uncertainties and additive external disturbances must be addressed.