: Recently, along with the technical innovation in robotics, human-collaborative robots, mobile robots and other similar robots have begun to be introduced in industrial sites. The introduction of these robots is expected to increase rapidly as the time passes. Since these robots require mutual approach and direct contact between humans and machines, the conventional safe measures, such as the isolation of humans and machines from each other with fences or enclosures, are no longer applicable and some new safety technology should be developed instead.
    On the other hand, the conventional operation of processing, adjustment, trouble-shooting, maintenance, inspection, repairing, cleaning or other type operations that require humans to approach machines without stopping the machine moving parts(hazardous point nearby operation)are also important operations of the man-machine collaborative working system. However, since the conventional safe measures are not applicable, fatal industrial accidents occur frequently even today, and the development of new safety technology calls for urgent attention.
    To meet the above needs, a project research under the theme of the "Research on Fundamental Safety Technologies for Man-Machine Cooperative Working Systems" started in 2002, promoting the following sub-themes :

1. Clarification of the man-machinecollaborative conditions and the inherent safety structure.
2. Development mobile object follow-up techniques applied with environment recognition technology.
3. Establishment of accident prevention measures for the hazardous point nearby operation.

    This midterm report describes experimental results and accident prevention strategies performed from 2002 to 2004 mainly. Chapter 2 and 3 are related to the research subject (1). Chapter 2 suggests a human pain tolerance as a safe design index for human-collaborative robots. An inherently safe design method based on its index is also suggested. Chapter 3 shows experimental results of the human pain tolerance. The worst limitation of pain tolerance is clarified to be 57.7N. Chapter 4 and 5 are related to research subject (2). Chapter 4 suggests an object zone detection algorism considering the occlusion problem. Chapter5 shows a gesture recognition method as a new type of human-robot communication.
    From Chapter 6 to Chapter 8 are related to the research subject (3). Chapter 6 shows a result of fatal accident analysis caused by industrial machines. The result of this analysis is that 44.2% cases of fatal accidents are related to hazardous point nearby operations, 35.7% cases are related to operations in large production lines and 12.4% cases are related to a erroneous machine start of co-operators. Chapter 7 suggests an accident prevention strategy of the hazardous point nearby operation. A risk management division, an accident prevention division and a supporting protective device are suggested as a new concept of its strategy. Chapter 8 suggests an accident prevention strategy for the operation executed by multiple operators in large production lines. The matrix expression of operator's "CLASS" and human's action "TYPE" is suggested as a new method for the hazard analysis of human actions.

Keywords; Man-machine system, Collaboration, Inherent safety, Environment recognition, Risk assessment,Protective measure, Safe design, Safety control

: Recently, there has been a tendency for robots having no guard but a moving mechanism to approach humans and for humans to positively utilize the abilities of robots. Such coexistent and collaborative relation between humans and robots is expected to have demand not only in industrial fields but also in service fields. However, while they are implementing their work, it is assumed that such new-mode human-collaborative robots contact and dynamically work on human bodies. Due to this assumption, the rules of isolation and the rules of stop, which have been regarded as the preconditions of the safety securement for workers when they are working with conventional-mode industrial robots, are no longer applicable. As a result, the safety securement for humans is one of the most important problems with their practical application. Particularly, there is no definite index of the relation between humans and robots, such as how much force output should be permitted to a robot according to the degree of contact required in the course of implementing the target work.
    In the situation in which humans and robots work together by collaborating with each other, it is impossible to avoid the physical contact and the interaction of force between robots and humans. Even if a robot in motion contacts a human, the robot is not always requested to stop. The situation in which the human receives the excessive impact strength caused by collision and subsequent excessive pinching force may be regarded as a dangerous event. Since it cannot count on the robot for stopping its operation or on the human for avoiding accidents to prevent disasters, it is necessary to reduce the robot force conveyed to the human to be within the allowable range.
    In this study, the authors propose a pain tolerance as an index for examining the specifications of human-collaborative robots from the human side, and try to define the force output characteristics of robot actuators based on this pain tolerance. The authors also examine the maximum allowable deformation of skin, which is another means of expression for the pain tolerance, and discuss the braking characteristics of robots as well.
    As the index for examining the specifications of human-collaborative robots from the human side, the authors propose the pain tolerance as the dimension of force and the maximum deformation, define the force output characteristics of the actuator from the former and the braking characteristics of the actuator from the latter, and examine the safety design strategy for human-collaborative robot. The force output characteristics of the robot actuator are defined for each of 3 stages to restraint it under the pain tolerance of 57.7 N and the mode of inching operation is selected to limit the overrun of the robot arm to the maximum allowable deformation.

Keywords;Robotics, Human-robot collaboration, Human-machine cooperation, Inherent safety, Safe design, Risk reduction

: Concerning the inherently safe design measures of a new type of robot which aims to contact with persons like a mobile robot not enclosed with a guard, this report focuses on the tolerant limits of the sense of pain that the persons feel while being contacted by such the human-collaborative robot. The pain tolerance is predictable information linked directly to the external injuries of human bodies and a significant parameter to determine the acceptable limits of mechanical stimuli that the human-collaborative robot can give to the persons in the process of a designated task. However, from the previous studies related to the level of human sense of pain, the great difficulty of numerically analysis of the pain tolerance has been pointed out because the distributions of the human subcutaneous tissues and pain sensing organs are greatly depending on the human body portions. Therefore, there still has not been enough data available to manifest the pain tolerance.
    With the view to the above, we focused on the sense of somatic pain felt by those who were contacted by the robots and then an index based on the human pain tolerance for the safety design of the human-collaborative robots was proposed in our previous studies. However, in the previous studies, the validity of adopting the pain tolerance as the safety evaluation index for the human-collaborative robots could not be demonstrated well because of the following problems: 1) the probe speed, 2) the probe angle to contact the subject's body surface, 3) the influence of the size and shape of the probe.
    This report describes the composition of a measuring system improved to solve these problems and the statistical results of the pain tolerance obtained from truly static measurements performed by a new measuring system. The movement of the contact probe at constant slow speed was achieved by adopting a ball screw mechanism and an AC servo motor. This, however, resulted in the substantial increase in the risk of this measurement and thereby a protective measure utilizing a 3 position enable device and an electro-magnetic clutch is newly devised.
    The measurement of the pain tolerance was carried out with 9 subjects limited within healthy male adults. Referring to the results of the preliminary tests, the spherical tip-end probe of 10mm diameter was set and the subjects put on defined clothes. From the result of the measurements, the lowest pain tolerances obtained from the 9 subjects were shown at the upper arm and the mean value and the 5th percentile at the upper arm were 112 N and 57.7 N, respectively. On the other hand, the highest mean value of 294N and the highest 5th percentile of 215N were observed at the back. With the range of this measurement, there is a tendency that the pain tolerances on the back of the body were higher than those on the front of the body, and there was no correlation at all between the apparent softness of the compressed region and the pain tolerance. Considering that the pain tolerance is the values based on the subjective judgment of the individuals, it can be said that very consistent values were obtained and regarded as effective guidance for the safe design of the human-collaborative robot.

Keywords;Camera monitoring system, Image processing, Safety, Omni-directional vision sensor.

: The purpose of this research is to develop the camera monitoring device which is applicable to safety protection devices used in mechanical safety, especially in human-machine cooperation systems. This kind of device is usually called Vision Based Protective Devices (VBPD). By using VBPD, the machines are stopped while people intrude the dangerous area of the machine monitored by video cameras.
    In this paper, the method for detecting presence area by using multiple omni-directional vision sensors is proposed for VBPD, and its applicability to collision estimation is examined by simulation. It is shown that the method robustly implements the presence-area detection even when occlusion occurs. In addition, an experiment system is constructed by using four omni-directional vision sensors, the demonstration for presence-area detection and collusion estimation is performed on AGV. From the demonstration, it is shown that tracking moving-objects and collision estimation in real-time are possible. Furthermore, it is assured that presence-area is detected without missing any objects, even when occlusion occurs.
    The results of this paper can be concluded as the follows :

1. The detected presence-area contains the moving-objects without error, but the true area of the moving-object is larger than the real area.
2. When more than 3 omni-directional vision sensors are used, comparing with the inner areas enclosed by sensors, the outer area is likely to be detected as extra larger area. If this is the problem, the monitoring area had better to be the inner areas.
3. In a simulation experiment in which four omni-directional vision sensors are set at the vertexes of a 5m x 4 m rectangular, a cylinder object with diameter of 30 cm can be detected as a presence-area with up to three times of the area, in the case that the method for presence-area detection is a simplex one for the purpose of real-time applications.
4. Although the presence-area detection is robust to occlusion, the missing is unavoidable if multipleobjects are crowded together such that there seems to be one object by all cameras.
5. Ghosts in the detection results are unavoidable by the proposed method. There is basically no solution for the ghost problem, therefore it is necessary to supplement with the extra sensors corresponding to the monitored objects. In addition, there is also the solution in which a result presence-area smaller than a threshold is thought to be a ghost and is deleted. However it is necessary for this solution to set the threshold smaller than presence-areas of human-bodies.

Keywords; Camera monitoring system, Image processing, Safety, Omni-directional vision sensor

: When the gesture recognition is used in robot control, as a new method for human-robot communication, it is required to do the risk analysis and the safety assessment for the method. Because the requirements on safety and on reliability are strict, only those simple and easily-recognizable gestures could be used in the current stage of gesture recognition technologies. In this paper, the gestures considered are restricted to those inputed from still images and those formed only by two arms. A three dimensional (3D) model is established for expressing gestures quantitatively in three dimensional space, and a quantity analysis model based on the distance between gestures, which we call the Gesture Distance Model (GDM), are proposed for evaluating about the possibility of gesture miss-recognition(or simply the miss-recognition risk). By applying to the GDM the average body-sizes in the Japanese Body-size Database, an analysis table about the standard distances between 16 basic gestures is obtained. In addition, an implementation for gesture recognition based on the GDM, by using the standard body-size as the one side in GDM and the gesture to be recognized as the other, is also proposed. Finally an explanation is also given about the gesture design method which causes small miss-recognition risk.
    The results and the importance of this paper are that the quantitative analysis of GDM, developed to reducing the miss-recognition risk, can also be applied as a general quantitative analysis expression for all possible arm gestures in still image. As the result,

1. This quantitative expression makes possible for defining and describing the gesture exactly without ambiguity.
2. When implementing a gesture recognition system by using GDM, it is not necessary to learn and save many image patterns and their variations to match whether an input gesture image is among the designed ones. Instead, the images corresponding to designed gestures and to their variations can be generated by calculation according to this model, and even the image match can be replaced by the match of features.
3. By quantitative model for gesture expression, it is also possible to establish a standard for gesture design and for safety evaluation used in human-robot interfaces.

Keywords;Human-machine Communication, Pattern Recognition, Gesture Recognition, Robot Safety, Risk Assessment

: When the fatal accident arises, many people declare that we never repeat more accident. However, it is forgotten as the time passes and a resemble accident repeat again. The reason why such problem arises seems to be because not only original causes are not perfectly investigated to each accident, but also information on the accident are not universalized as a general engineering knowledge.
    This report intends to investigate original causes of fatal accidents caused by industrial machines and to clarify a universalized engineering knowledge that is useful for machine designer or manufacturer. Fatal accidents of 129 cases caused by industrial machines are analyzed for that purpose. The important target of investigating original causes is to clarify a fault of safe measures and human operations. Two sorts of accident caused by "being crushed or drawn into a machine" and caused by "being impacted against a machine" are treated in this report.
    As a result of fault analysis for safe measures, it is clarified that 79.2% cases of fatal accidents are caused by the fault of safe measures, such as fixed guards (34.9%), movable guards (51.9%), protective devices (24.0%) or safety-related control systems (23.3%). This result indicates the possibility of drastically decreasing accidents by means of adequate safe measures described in ISO 12100 standard or in "Guidelines on comprehensive safety standard for machinery". As a result of fault analysis for human operations, it is clarified that (1) 44.2% cases of fatal accidents are caused by accidents related to "hazardous point nearby operations", that is the operation of process confirmation, adjustment, preparing, trouble-shooting, maintenance, inspection, repairing, cleaning of the machine etc. without stopping the machine moving parts; (2) 35.7% cases related to operations in large production lines; and (3) 12.4% cases related to co-operator's erroneous machine starts.
    It is also clarified that rate of cases related to (2) and (3) is 39.5%, and rate of cases related to (1) through (3) is 65.1%. From the result mentioned above, roughly two thirds of fatal accidents relate to any of (1) through (3). Consequently, there is a possibility of decreasing the number of cases of accidents by establishing the adequate accident prevention method by putting focus on hazardous point nearby operations related (1), and on operations made by multiple operators in large production lines, related to (2) and (3).

Keywords; Safety, Safety Control, Industrial Machine, Fatal Accident, Accident Prevention Method, Hazardous Point Nearby Operation, ISO12100

:It is a common knowledge for safety engineers to perform machine safe measures based on the risk reduction process described in ISO 12100. Its process consists of the "inherently safe design measure", "safeguarding" and other supplemental countermeasures. However, there are many "hazardous point nearby operations", that is the operation of process confirmation, adjustment, prepairing, trouble-shooting, maintenance, inspection, repairing, cleaning of the machine etc.. It is very difficult to apply above safe measures for these operations because a operator is close to machine moving parts without stopping it.
    This paper shows a new safety strategy not only for operations that the "inherently safe design measures" and the "safeguarding" can be used, but also for "hazardous point nearby operations". New concepts of the risk management division ( I:safe condition, II:uncertain condition, III:hazardous condition), the accident prevention division ( 0:ristriction of speed and force, 1:separation and fixation of human and machine space, 2:machine stop in hazardous condition, 3:hazardous point nearby operation) and the supporting protective device ( type A:periodic checking for protective devices, type B:priodic training for operators ) are suggested for that purpose.
    These results can be used as a complement of ISO 12100. Following measures are especially to be taken for the hazardous point nearby operation.

1. Improving the operation method to start a machine from the outside of safeguarded space. This causes transition of the accident prevention division to division 1.
2. Application of safety confirmation system to stop a machine in hazardous condition. This causes transition of the accident prevention division to division 2.
3. Even if 1) and 2) are difficult, trying to reduce a risk as far as reasonably practicable by applying the supporting protective device or realizing a restricted (low speed) machine operation.
4. In the case of application of the supporting protective device, clarify the matter to be informed from a manufacturer to a user. The manufacturer delivers information of residual risk to the user. It means the risk transmission.
5. Prepairing safe operation procedures and planning education and training for using the supprting protective device.

Keywords; Safety, Safety Control, Industrial Machine, Accident Prevention Method, Hazardous Point Nearby Operation, ISO12100

: There are many hazardous operations executed in large production lines such as the operation of trouble-shooting, maintenance or cleaning of machines. An operator enters in its line for these operations while a machine is stopped. In this case, when other operator restarts the machine by mistake, the accidental start of the machine could cause a serious disaster.
    An indirect monitoring device using a movable guard, key-switch or plug is adopted to prevent these accidents. However, this is not always effective because an operator may close to a movable guard while other operator remains in the line, or he enters in the line without extracting the key. To prevent these problems, another system is adopted so that the movable guard does not open unless the operator extracts the key or plug. However, even if this countermeasure is taken, it could not be effective if, after an operator who does not have a key enters in the line with an operator having a key at the same time, only the operator having the key exits the line and returns the key to the key box.
    This paper shows a new accident prevention strategy for the adequate risk reduction in large product lines having above mentioned problems. Results of this research are described as follows:

1. The hazard analysis method for complicated human actions is suggested by the matrix expression of operator's "CLASS" and human's action "TYPE".
2. The direct monitoring system such as the laser scanning sensor can achieve the adequate risk reduction (risk management division I). On the other hand, the indirect monitoring system using a movable guard, keyswitch or plug cannot achieve it(risk management division II) because they involve the problem that operators forget to extract the key or plug. The indirect monitoring system using an IC tag cannot also achieve it (risk management division II)because an operator sometimes forgets to use personal protective equipments (helmet, gloves etc.) that an IC tag is installed on.
3. The indirect monitoring system that uses both a key-switch and a monitoring device (photo-electric sensor, mat switch etc.) is regarded as the risk management division I or II because the machine restarting operation is relying on chief operator's attention. On the other hand, there is a possibility of achieving the risk management division I by using a both key and a counter gate.

    These results can be useful to the supplement of ISO11161 standard.

Keywords; Safety, Safety Control, Industrial Machine, Accident Prevention Method, Human Detection System