Tetsuro Yabuta

Cooperative control algorithms for a redundant finger-arm robot were proposed in a previous study, based on the movement of the human hand-arm system. With these algorithms, the finger-arm robot arm was able to complete a constrained task by integrating admittance control and impedance control with the manipulability control of the fingers. The manipulability of the fingers was controlled using both an approximate global search (i.e., the top search method) and a local optimization method. However, the dynamic characteristics of these methods have not yet been studied. In this study, the conventional cooperative control algorithms were developed further by extending them to dynamic movement in the admittance control of a redundant multi-finger-arm robot with manipulability control of the fingers. We clarified the relationship between the feasibility of the virtual dynamics and the motion speed, as well as the range of the achievable virtual dynamics. This paper evaluates and discusses the proposed methods in terms of the finger manipulability control. The experimental results show that the algorithms responded to more dynamic movement, and a broader range of virtual dynamics was achievable.

In our laboratory, we have succeeded in acquiring forward actions to various robot systems using Reinforcement Learning. We have also succeeded in acquiring a giant swing motion as dynamic task by devising its rewards. Then, the purpose of this study is to clarify probabilistic behavior of giant swing. Although the giant swing robot has a continuous dynamic motion such as its angle and angler velocity, its state of the motion must be divided into 216 states in order to apply the reinforcement learning. For this reason, this robot shows probabilistic behaviors. Consequently, it became clear that collapse of the learning knowledge may happen which is defined as Q value that could acquire the giant swing motion become impossible to acquire this motion by increasing learning count. The results show that 100% successful rotation cannot be obtained even when its learning converges sufficiently.

The exercise to move a whole body from a certain place to other places is one of basic and important activities for all creatures. In our laboratory, on the view point of forward locomotion-pattern, we have succeeded that various robots acquire locomotion pattern by using Q-learning method that is one of the reinforcement learning. However all locomotion-patterns of our previous studies are assumed to be quasi-static, whose motion velocity is very slow. For that reason, this paper attempts to acquire dynamic and consecutive locomotion-pattern of qudarupedal robot. Hoyt et al. researched relation between running velocity and oxygen consumption per unit distance when horses walk, trot and gallop at any speed. According to these results, each gait has a certain velocity to minimize oxygen consumption, and has the almost same minimum value among each gait. Therefore, this paper attempts to reproduce this phenomenon of a mobile robot.

Although eyes and upper limbs work together over the whole course of arm movement, vision usually dominates in visual-haptic perception. The dependence of the brain on visual information is demonstrated by the wide variety of visual illusions, such as “shade and shape,” that have been developed. In order to quantitatively measure upper-limb motion, a haptic device can be used to obtain numerical data. Although, as the name suggests, such a device is capable of transmitting force, human perception mechanisms tend not to measure force accurately owing to interference. In this paper, we analyze the role played by visual information in line drawing. In experiments where participants were asked to estimate the length of a line using a haptic device operated by an experimenter and then to draw a line on their own, there was a tendency to draw a longer line when their eyes were open. These results reinforce the importance of visual information to length recognition in drawing.