: Rapid introduction of new type robots, such as collaborative or mobile robots, to the industrial environment is anticipated, but the diffusion of these robots is difficult under the present condition. The reason is that conventional protective measures like fences cannot be applied and moreover new types of measures instead of fences do not have been developed for practical application. On the other hand, hazardous human operations required approaching machines without stopping the machine movable parts still remain and are difficult to prevent industrial accidents by applying conventional protective measures. These operations, such as maintenance, trouble-shooting, adjustment, etc., are called "hazardous point nearby operations" which are considered one of the forms of "man-machine cooperative working systems".
    To solve the above safety problems, a project research on the theme of the "Research on Fundamental Safety Technologies for Man-Machine Cooperative Working System" started in 2002, and it had tried to establish fundamental safety measures including methodologies and devices until 2007. This project research was carried out the following sub-themes:

• a) Clarification of the conditions of man-machine coexistence and collaboration and the inherent safety structure.
• b) Development a mobile object follow-up technique applied with environment recognition technology.
• c) Establishment of accident prevention measures for the hazardous point nearby operation.

This final report announces the sequel of the midterm report and is mainly concerned with new results obtained in the latter term of this project.
    Chapter 2 is related to the subject a) and describes a new safety device for human-collaborative robots. This device designed according to the safety guidance for human pain tolerance reported in the midterm report functions as a normally closed type clutch, and can realize safe torque control. Its fundamental characteristics of control and safety performance were found suitable for practical robot application.
    Chapter 3 and 4 are related to the subject b). Chapter 3 describes a new object detection method for man-machine coexistence systems using omni-directional vision sensors, which were proposed in the midterm report. In addition, a normality confirmation method with reference pattern comparison was discussed and experimented. Chapter 4 proposes a communication method for mobile robots by recognizing operator's gestures. This method was designed to reduce accidents caused by communication error, and the misrecognition risk of this method was evaluated.
    Chapter 5 is related to the subject c). In this subject, the industrial accident analysis, the accident prevention strategies for hazardous point nearby operations and for plural workers' operations in large scaled productions are reported in the midterm report. This chapter is a follow-up of the fundamental prevention consideration for hazardous point nearby operations. On the basis of this consideration, a new protective device of wood cutting machine (circular saw) which was a typical machine used in such hazardous operations was proposed.

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

: Recently, a new type robot sharing work space with persons has been under research and development for use in carrying heavy objects not only in factories but also at outdoor working site sand in welfare facilities. Since the general purpose of this robot is to substitute for conventional carriage supporting equipment, a working mode in which a human and a robot carry a heavy object together in a coordinated manner is being widely studied.
Since it is impossible to control the torque output of the load-lifting actuator not to exceed the human tolerance, especially those of vertical articulated type, a safety mechanism that always monitors the actuator torque output and stop the torque output immediately in case of excessive output occurrence is essential to ensure the safe use of robots. In addition, the robot must be capable of maintaining its posture with no supply of energy in emergency stop operation or failure occurrence.
    In order to realize the safety mechanism capable of fulfilling the above requirements, a normally closed type clutch using magnetorheological suspension and rare-earth permanent magnets is proposed in this study as a safety device for human-collaborative robots. This magnetorheological clutch functions as a variable torque limiter and a holding brake in robot joint axes. Aiming to achieve both good controllable characteristics and safety performance of the clutch, the magnetic poles of internal permanent magnets are faced asymmetrically for the improvement of magnetic field generation and the multi-sheet rotor stricture is adopted for the increase in the amount of torque output.
    An experimental result shows that the normally closed type of magnetorheological clutch can transmit driving torque up to approx. 120 Nm by an application of exciting current of -1.0 A and output holding torque of approx. 40 Nm which have enough applicability to the practical robot application.

Keywords;Human-Collaborative Robot, Risk reduction, Magnetorheological suspension, Actuator, Safe torque control, Asymmetrical failure characteristics, Emergency stop function

: The aim of this research is to develop a device to protect persons from possible collisions with machines in environments where machines move around working persons and where conventional protective measures for fixed machines such as guards are not applicable. Emphasis is placed on the development of an object detection method for image analysis with the aid of a computer and video cameras. This type of device is usually called a vision-based protective device (VBPD).
IEC/TR 61496-4 provides information related to the design, construction, and testing of electro-sensitive protective equipment (ESPE) that employs VBPDs for sensing purposes for safeguarding machinery. For writing this technical report, the results of a Japanese research project performed by JMF (The Japan Machinery Federation) were used. In the project report, the following were pointed out as problems encountered in the development of VBPDs. The integrity of the automatic image recognition technology currently available is insufficient for the application of the technology to protective devices.
First, object zone detection with a volume intersection method using omnidirectional vision sensors is proposed in this study. The detected object zone definitely contains the target object; however, the detection method has a tendency to estimate the zone to be larger than the target's real volume. In addition, false detection is unavoidable, and there is basically no solution for the ghost problem. These are features pertaining to the safety aspect. When more than three vision sensors are used, by comparing the inner areas enclosed by sensors, it is easy for the object zone of the object in the outer area to be overestimated. Therefore, the detection zone should be set in the inner area.
Second, a normality confirmation method using a reference pattern created by a light source instead of the floor or a wall filled with a reference pattern such as a checked pattern is proposed. The normality of the camera system is verified by taking a picture of the pattern projected onto the monitor by timesharing and checking the image. To take a picture of the reference light and the observation light alternately, a beam splitter is attached to the camera. If a checked pattern is used as the reference pattern created by a light source, a JMF normality confirmation algorithm can be used. Furthermore, the object detection method is examined by using an overlay image of the pattern light and observation light so as to improve the performance of the method. An experiment showed that the regularity of the pattern in the overlaid image varied depending on the complexity of the observed image. The experimental result indicates that object detection can be performed using an overlay image.

Keywords;Vision Based Protective Devices, Image sensor, Image analysis, Normality Confirmation

: As the robot technology advances, in stead of keyboards and control panels, which were previously used as the main communication method with a robot, new communication methods which uses voice recognition and gesture recognition are expected to become common as the robot will work cooperatively with human being and will be widely involved in people's daily life. That is to say, robot will act according to instructions from these kinds of interfaces, therefore the interest of our research is focused on risk assessment and risk reduction related to these new interfaces while they are used for robot control, to assure robot safety. Particularly in our research here, the type of the new interface is restricted to gesture recognition, and the research focus is on reducing the gesture misrecognition. As the first step for assessing and evaluating the different design selections according to the misrecognitionrisk for gestures, two models were established in the previous stage of our research. The first model was a three dimensional (3D) model for quantitatively expressing gestures, and second model was a quantity analysis model, called gesture distance model (GDM), to measure the difference between different gestures. The examples of applying these models with the average body-size data in the Japanese Body-size Database were given, and the basic data about standard gesture features of average body-size for Japanese people were also obtained. These results were reported previously in the reference¹⁾. In addition, the applications of these results to reduce misrecognition were briefly discussed there.
    Here in this report, design considerations are discussed about how to reduce the gesture misrecognition during the entire design work for a gesture recognition system. The design work is divided into three steps: basic design, function design and parameter design. In additional to the two models previously established, two principles are introduced, i.e., simplicity principle and distance principle. Based on these model and principles, the design method and policy are discussed individually for the three steps during the design work. In the basic design step, simplicity principle is mainly used.On the other hand, the distance principle, the 3D model and GDM model are mainly used in the function design steps and parameter design step. The results and the importance of our research are that the quantitative models and methods are provided for evaluating and comparing different design selections for implementing a gesture recognition system. Based on these results, we hope to provide a prototype to establish risk assessment and safety standards for the gesture recognition system to be used in industrial human-robot interfaces.

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

: It is a common knowledge for safety engineers to perform machine safe measures based on the risk reduction process described in ISO12100. Its process consists of the "Inherently safe design measure", "safeguarding" and other supplemental countermeasures. However, there are many "hazardous point nearby operations", such as the operation of process confirmation, adjustment, preparing, troubleshooting,maintenance, inspection, repairing, cleaning of the machine etc.. It is very difficult to apply above-mentioned safe measures for these operations because operators are close to machine moving parts without stopping it.
    The midterm report of the "Research on Fundamental Safety Technologies for Man-machine Cooperative Working Systems" published in 2005 proposed new risk reduction strategy against the "hazardous point nearby operation". New concepts of the risk management division, the accident prevention division and the supporting protective device are suggested in this midterm report.
    This paper proposes accident prevention conditions for "hazardous point nearby operations" and safety system for wood cutting machines based on above research results. Results obtained from this research are summarized as follows.

1. Clarification of accident prevention conditions
       The accident prevention condition is classified four types (0:ristriction of energy, 1:separation and fixation of human and machine space, 2:machine stop in hazardous conditions, 3:limitation of speed of machine moving parts (3a) or human body (3b)). Types of 3a and 3b mean accident prevention conditions for "hazardous point nearby operations".
2. Development of safety system for wood cutting machines
       The new safety system of wood cutting machine has a circular saw, two pieces of movable protective covers, locking device of these covers, IC tags for fixing to human fingers, loop-coil to detect IC tag sand control system. The circular saw of wood cutting machine is covered with rotated two pieces of (upper and lower) movable protective covers. IC tags are fixed to operator's fingers. Safe zones are formed to left and right of the circular saw. Safe zones are monitored by loop-coils to detect the presence of IC tags. The human absence is confirmed by detecting IC tags. These covers can be permitted to move only when the existence of operator's fingers in safe zones is confirmed. On the other hands, these covers are locked to move when the existence of operator's fingers in safe zones is not confirmed.

Keywords;Safety, Safety Control, Hazardous point nearby operation, Wood cutting machine, IC tag

: This specific research proposed new fundamental safety technologies for man-machine cooperative working systems. Results obtained from this research are summarized as follows.

1. Clarification of the man-machine collaborate conditions and the inherent safe structure
       In this subject, a pain tolerance was proposed as the dimension of force and the maximum deformation for examining the specifications of human-collaborative robots. The force output characteristics we redefined the safe force output characteristics of robot actuators on the basis of the pain tolerance, which was a minimum of 57N.
       According to the above result, a normally closed type clutch using magnetorheological suspension and rare-earth permanent magnets was proposed in this report (Chapter 2) as a safety device for human-collaborative robots. Examining the characteristics of this clutch under the various input current conditions, the experiment results showed sufficient control and safety performance.
2. Development of mobile object follow-up techniques applied with environment recognition technology
       In image recognition technology, one cause of failure in detection is error estimation of geometric features. The failure in detection is hardly acceptable to a vision system for safety. So object detection method without geometric features was developed. This method detects object zones with a volume intersection method using multiple omni-directional vision sensors. Another feature of this method is to estimate a larger zone than real area of target. It is considered that this method has safety side feature.
       Besides, a normality confirmation of a vision system is required for safety. The confirmation includes calibration of camera, function of signal processing circuit and so on. So the confirmation method using a reference pattern created by a light source instead of the floor or a wall filled with a reference pattern described in IEC/TR 61496-4 was proposed. The normality of the camera system is verified by taking picture of the pattern projected onto the monitor by timesharing and checking the image.
       A communication method for mobile robots by image recognizing gestures was also discussed.
3. Establishment of accident prevention strategy for hazardous point nearby operations
       In this subject, it is clarified that 44% cases of fatal accidents are caused by accidents related to"hazardous points nearby operations" and 36% cases related to operations in large product lines.
       A new risk reduction strategy which has new concept of the accident prevention division (0:ristriction of energy, 1:separation and fixation of human and machine space, 2:machine stop in hazardous conditions, 3:limitation of speed of machine moving parts(3a) or human body(3b)) and risk reduction division (I:safe, II:uncertain or ALARP, III:hazardous) was proposed. This result can be used as a compliment of ISO12100-1. A new safety system for wood cutting machines which has new movable guard and human detection equipment using RFID was also proposed.
       Results obtained from this research were applied to many safety standards such as ISO10218-1, ISO13855, IEC 61496-4 and article 28-2 of Japan industrial safety and health law.

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