Koichi Hidaka

Human skill elucidation based on gaze analysis for dynamic manipulation

To realize new kinds of human-machine system proposed as human adaptive mechatronics (HAM), experimental analyses to elucidate humans skill is reported in this paper. For skilled operation, we has been thinking that an adequate self-switching of reference targets on the manipulation is important. To prove this idea, we designed a special task by using virtual computer graphics of a hovercraft, measured gaze behavior of the operators, and analyzed their skill that seems to be related with the switching references. From the analyses of behavior of the skilled operator, it was confirmed that early switching of sub-controllers / reference signals plays a significant role in skill.

Evaluation of Human Skill in Teleoperation System

In this paper, an experimental analysis to evaluate human skill on a machine manipulation is introduced for realization of human adaptive mechatronics(HAM). HAM is a novel concept of intelligent mechanical systems that adapt themselves to assist and improve the users skill. Using a tele-operated wheel mobile robot and a maze task, we acquired data of which human became skilled, and analyzed them in order to know what index can evaluate human skill level. We applied three kinds of analyses, that are eye-gaze behavior analysis, analysis of feasible operation based on observed camera image and a pattern recognition by a Markov model.

Modeling of a radio-controlled car with camera based on system identification — Comparison between open-loop and closed loop methods

This paper describes a modeling of movement object. A auto driving of the car is studied to reduce traffic accidents and traffic jam in late years. The experiment tries there control with the radio controlled car which did movement same as a car. The control of the radio controlled car controls the steering voltage on speed uniformity this time. The control method usually uses model predictive control. The precision of the model becomes important for the model predictive control. To give the precision of model, a system model of straight line and curve line is necessary for the model of the radio controlled car. Since these movements of the radio controlled car are different, these model do not become the same models. The model of the car on curve needs closed loop identification with impossibility by open loop identification. Therefore, we examine the model of straight line using closed loop identification model to consider proper model.

Skill analysis of wheel mobile robot operation

This article reports result of experimental analysis in order to declare humans skill on a machine manipulation for global research on human adaptive mechatronics(HAM), that supports human and enhances human-machine total system. Using tele-operation wheel mobile robot with maze-task, the process of the manipulation of operators were measured and were analyzed via identification of the operators behavior. It was confirmed that experts control characteristics are well expressed by multi-switched linear ARX-models with prediction ahead.

Route space detection of industrial indoor vehicle with joint extended histograms of oriented gradients(EHOG)

This paper introduces the method of detecting a route space for an industrial indoor vehicle. The vehicles work in many industrial fields, e.g., a semiconductor production and a car assembly factory. The detection of the route space under illuminant disturbance is an important problem for the industrial vehicle robot. The industrial vehicle has to move to the same areas in a factory. For these works, the usual industrial vehicle, e.g., an automated guided vehicle (AGV), is transported in path line. However, this conventional system of AGV is not a flexible method to change a goal position. On the other hand, a mobile robot, e.g., a wheeled robot, can move without a line or rail for the path, and an auto vehicle runs on the urban road. These robots use multi-sensors. However, the industrial vehicle intent to decrease the number of sensors for the cost down. For the reason, we propose a detection of a passable route space for an AGV. We proposed the detection method using a stereo camera in order to move the AGV without path line. Our proposed method is based on the joint extended HOG (joint EHOG) and AdaBoost algorithm. EHOG has the robustness for the illuminant disturbance, however, the EHOG is not robust for variable image size. The proposed method has the robustness to the illuminant disturbance and to variable image sizes of the target. The proposed method decreases the number of the wrong detection by weighting factors to a weak classifier and combines the many detection areas by the mean shift clustering. The experimental results show the effectiveness of the proposed method for a route space detection and we finally show the future works.

Recognition of object by extended Histograms of Oriented Gradients (EHOG) on route for a mobile robot

This paper presents a recognition of obstacle and objects for an industrial a mobile robot, e.g., an automated guided vehicle (AGV), by using monocular camera. The mobile robot moves for transporting same parts in a factory where the robot has to pass a production line. An accurate recognition of object on the production line is required for moving the robot automatically. In addition, the robustness to luminance changes is required. During the past decades, some robust features, such as Scale Invariant Feature Transform(SIFT), Speeded Up Robust Features(SURF), Histograms of Oriented Gradients(HOG), or Extended HOG(EHOG), have been proposed in computer vision and machine learning. In this paper, we focus on the robustness of EHOG and we propose a decision algorithm of objects on a path by using the machine learning based on EHOG.We show that experimental results are provided and the usefulness of the proposed algorithm is introduced by these results.

Closed-Loop Identification and Robust Control Design of Vehicle via Modeling

Abstract This paper presents a modeling approach to control a vehicle by visual feedback and H ∞ control; this approach is based on two models designed using a closed-loop identification. The models are constructed from data for a curved path and straight path. Closed-loop identifications such as Closed-loop subspace model identification method (CL-MOESP) have been proposed recently and validated through simulations. However, consideration based experimental data have rarely been studied. We investigate the most effective identification approach of a vehicle model by experimental data. The most useful vehicle identification using visual feedback control is CL-MOESP. To evaluate its effectiveness, we use the fitting rate (FIT). Next, we design a robust controller based on the constructed models. Specifically, we formulate a model error for each models based on the curved and straight paths and use an H ∞ control algorithm for the vehicle controller within a given error. A nominal plant is regarded as the model given by the curve data. The model error is estimated as the difference between the models given by the straight and curved paths.

Adaptive control design of an omnidirectional autonated guided vehicle in consideration of mass change

Automated Guided Vehicle (AGV) is used in many factories. Wheeled mobile robot (WMR) is applied to the AGVs. The structure can not move in all directions freely. On the other hand, the structure that can be moved in all directions is important in the limited place such as factory. Recently omnidirectional mobile robots are studied in the robot field. One type of the mobile robots can move for all directions with special structure such that omniwheel. However the wheel damages floors, the carrying capacity of the robot is low and the ability of climbing bump is inferior. These problems are very weakness when the robot is used the AGV in factory such that assembly car plant. For the reasons, the AGV with omnidirectional type mechanism is desired. In this paper, we proposed a new control method of Omnidirectional AGV. The proposed control uses the adaptive control technique for considering mass change of load.

Relation between skill revel and input-output time delay

In this paper, we try to obtain knowledge about human skill on machine operation, using data measured by simple test, and analyze up-skilling. In order to obtain experimental data for human manipulation, we utilize a tele operated robot system. For the tele operation, it is necessary that human operates machine by using image information from display monitor. We aim at the acquisition of evaluation quantity about state of skill, and pay attention to delay time during the machine operation. For the skill, we investigated correlation with response delay time of human operation and control characteristics of the operations. We confirmed a possibility of quantification of progress of the up-skilling.

Adaptive feed-forward loop connection based on error signal

In this paper, we investigate effect of changing the connection of feed-forward loop based on error signal. Our motivation of this work is solution to progress of human skill. For the skill model, we study a human simple action such as arm motion. Many models that describe the human arm dynamics have been proposed in recent year. While one type does not need an inverse model of human dynamics, the system based on the model does not include feed-forward loop. On the other hand, another type model has a feed-forward loop and feedback loop systems. This type assumes feed-forward element includes an internal model by repeating action or training and this loop progress our skill. Then we usually have to exercise to get a good performance. This says that we design the internal motion model by training and we move on prediction for motion. Under the assumption, Kawato model is well known. The model proposed that learning of feed-forward element is promoted in brain so that the error of feedback loop decreases. Furthermore, we assume the connections in feedback loop and feed-forward loop are changed. We show numerical simulations and consider that the position error given by our vision changes the skill element and we confirm that the position error is the one of the estimate function for the improvement in our skill.