M. Kakikura's research while affiliated with Institute of Electrical and Electronics Engineers and other places

The simulation framework we propose for complex path planning problems with multiagent systems focuses on the sheepdog problem for handling distributed autonomous robot systems – an extension of the pursuit problem for handling one prey robot and multiple predator robot. The sheepdog problem involves a more complex issue in which multiple dog robot chase and herd multiple sheep robot. We use the Boids model and cellular automata to model sheep flocking and chase and herd behavior for dog robots. We conduct experiments using a Sheepdog problem simulator and study cooperative behavior.

This paper addresses a new assistive scheme for enhancing the operation performance. We focus that the suitable assistive elements depend on variable operator's characteristics as individuality, skill, robot and environmental condition etc. Effects of the assistive elements, therefore, are different according to the operators and circumstances. However, the assistive effects were generally contemplated by system designers without concern of variable human operators' characteristics. Furthermore, required performances depend on the task which often include exclusive factors like promptness and safeness. This has possibilities to give adverse effects to the other elements against the designer's intention. And also, costs of assistive elements should be considered because the resources of information presentation is limited. In order to improve these problems, in this paper, a planning technique to achieve adequate assists is proposed. The technique deals with an optimization problem between on-line estimated assistive effects and their assistive costs. Finally, abilities of the proposed technique are verified by some experiments

Legged robots are expected to walk on irregular terrain with their high ground adaptability. Working in irregular environments, the robot is required to keep safe, that is, it is necessary to consider the terrain. This ability is difficult to be installed to the robots due to not enough intelligence and experience. Here, we focus that human can guess and estimate the safety foot position with global environmental recognition and prediction. We propose leg selectable interface for walking robots on irregular terrain. This time, we introduced this system into a quadruped robot. Proposed interface employ four fixed cameras above each foot to show the ground condition of around foot. The operator selects the object (each leg or main body) with the touch panel, and operates its position with the joystick. Our interface displayed a lot of information as robot condition. Additionally, to improve the stability of the robot, we install the two kinds of corrections, one is correction by robot's center of gravity and another is correction by leg angle. We verify the effect of above compensation with an experiment by a real machine. As a result, it was possible to walk by avoiding the obstacle, the robot was able to be stabilized by the correction.

This paper addresses a real time human sensitivity estimation technique on GUI for human adaptive teleoperation system. Although human operators have valuable abilities as global recognition, prediction and planning, they also have possibility to make mistakes. Especially, an overlooking among lots of information, is serious. In order to solve this problem, adequate GUI display planning for emphasising important feedback information is required. We therefore attempt to consider human perceptional characteristics. In other words, robustness for human individuality, skills, fatigue an important alert information window is put with perceptive properties consisted of colors, size, frame line width and so on. At the same time, unimportant optional windows put without disturbing the alert notice. However, the sensitivity depends on operator's individuality and GUI circumstances, namely, it is variable. Unfortunately, human operators often make careless mistakes that is not essential for the sensitivity. To evaluate the variable and including unessential mistakes sensitivity, in this paper, a realtime human sensitivity identification technique with statistical method is proposed and experimented.

Measurement of gaze is effective to evaluate human operator's attention, one's operation skills, perceptional capability and so on. Especially, gaze duration, called fixation time, is often utilized. Generally, it is said that long fixation time is detected when the operator pays attention to something intentionally. However, the duration also depends on saliency of displayed image, especially humans' perception characteristics are sensitive to intensities of an image. Although a lot of researchers have presented models of visual attention with the saliency map, the high saliency may attract a human gaze even if he/she do not have attention. Therefore, in order to estimate the human attention, we consider human vision characteristics with foveal vision. The foveal vision is used for scrutinizing highly detailed objects, and it also may relate to the attention. In this paper, we propose a new approach to estimate human visual attention by checking gaze duration and a saliency map considering human foveal vision characteristics. The estimation technique was experimented with five participants, and as the results, we found the technique makes aware of the attention more than conventional technique which considers only gaze duration.

The purpose of this paper is to propose a simulation framework for complex path planning problems with multi-agent systems. The paper focuses on the "sheepdog problem" that handles distributed autonomous robot systems. This problem is an extension of the "pursuit problem" that handles one prey robot and multiple predator robots. The sheepdog problem challenges to handle more complex issue, that is, multiple dog robots try to chase and capture multiple sheep robots. In this paper, we propose the use of Boids model and cellular automata to model the flocking behavior for the sheep robots, and the chasing and capturing behavior for the dog robots. We demonstrate an experiment of the sheepdog problem.

We propose an application of robotics technology to the motion analysis of motorcycles in order to decrease the risk of serious traffic accidents. This paper presents the experimental results with the electric motorcycle model equipped with the DCT mechanism (DCT ; Damping unit, Counter-steering angle, Temporary input), and also shows the results of simulation based on the Sharp* refined model that include the rolling resistance of the front tyre, round profile of tyres and the centrifugal force. In addition, we propose a measuring system for rolling angle of motorcycles with wireless camera. The experimental result of counter-steering phenomenon coincides with the result of simulation qualitatively, and confirms that the DCT mechanism is very useful equipment for data sampling of motorcycles.

Although the manipulator equipped on a mobile robot is extended its work space, it also brings about redundancy for the positioning due to the base mobile robot. To reduce the redundancy of manipulator, the manipulability is often applied as one of indexed for manipulator motion planning. In teleoperation with the mobile manipulator, it is load for human operator to consider the manipulability during a task. Consequently, we proposed an assist technique to improve the manipulability for teleoperated mobile manipulator. Furthermore, the viewability on the display monitor with the robot camera view is considered, and the robot motion to improve the viewability is adopted for the assist technique. Finally, the technique is experimented with four subjected in teleoperated manipulator positioning task and verified the effects.

Consideration of human skill is necessary to enhance total performance of human-machine system and to assist the human operation. We guess that human skill level may be evaluated by checking the task scheduling of the operator on a general machine manipulation. Based on this hypothesis, in this paper, we propose a vision-based method to evaluate ability of human task-scheduling by monitoring a hand motion of a human operator who controls a machine console. This method utilizes a Kalman filter, and extracts hand's position by changing the detection threshold dynamically. It was confirmed that the tracking to the hand motion is done successfully. Furthermore, it was found that checking difference of predicted hand motion and actual motion, gives useful information about the micro-slip.

This paper addresses a new technique of teleoperational assist to adapt to human operation. Although effects of the teleoperational assist functions are generally contemplated by system designers, the effect depends on variable operator's characteristics as individuality, skill, robot and environmental condition etc. Furthermore, required performances depend on a task which may include exclusive factors like promptness and safeness. Costs of assistance, also, should be considered for the teleoperation. Therefore, to improve these problems, a planning method for adequate assistances with online estimation of the assist effects is proposed.