Wearable Master Device Using Optical Fiber Curvature Sensors for the Disabled

Transcription

1 Proceedings of the 2001 IEEE International Conference on Robotics & Automation Seoul, Korea May 21-26, 2001 Wearable Master Device Using Optical Fiber Curvature Sensors for the Disabled Kyoobin Lee*, Dong-Soo Kwon** * : Dept. of Mech. Eng. KAIST (Tel: ; Fax: ; leekb@robot.kaist.ac.kr) ** : Dept. of Mech. Eng. KAIST (Tel: ; Fax: ; kwonds@me.kaist.ac.kr) Abstract This paper addresses a wearable master device for physically handicapped persons whose arms are disabled. Optical fiber curvature sensors are used to measure the human body motion. For the developed wearable master device, a calibration and mapping method of the sensors is proposed to extract 2-DOF human shoulder motions. An experiment shows that the wearable master device can be used for a 2-DOF input device effectively for handicapped persons. It has also shown that a subject can control a mobile robot with the wearable master device. 1. Introduction In order to aid handicapped persons in the basic tasks of their daily lives, various assistant devices have been studied and developed. Such devices assist handicapped persons to move themselves to another place, to handle an object or to eat meal. Several rehabilitation robot manipulators have been developed[1][2] for handicapped persons. Although the most popular type of master device for rehabilitation robotic system is the joystick type, the joystick is not appropriate for physically handicapped persons who cannot move their arms. Therefore, it is necessary to develop a convenient and portable master device that can measure human body motion. Optical fiber curvature sensors are used to measure human body motion. Although we can use image processing systems or absolute position sensor systems to measure the angle of body joints, these systems are expensive and bulky. Optical fibers seem to be well-suited for sensors that measure body motions of handicapped persons because of their flexibility, lightness, small size and low cost. Figure 1. A conceptual sketch of wearable master device A conceptual sketch is shown in Fig. 1. Optical fiber curvature sensors are attatched on an upper garment. 2. Optical fiber curvature sensor Optical fiber curvature sensors are used for measuring human body motions. Transmitted light power in the optical fiber decreases exponentially when the fiber is curved [3][4]. This characteristic of the optical fiber is essential for curvature sensing. By using several optical fiber curvature sensors, it is possible to measure two or more degrees of freedom of motion. The optical fiber curvature sensor designed for this research is composed of five parts: a light source, a light detector, an optical fiber, an electric circuit and a power source /01/$ IEEE 892

2 attached and two sensors are attached as shown in Fig. 5. The positions of sensors were experimentally determined where maximum variation of the sensor outputs is met. Figure 2. Assemblage of optic fiber curvature sensor The light source and the detector are an IR LED and a phototransistor, respectively and the electric circuit consists of one OP-amp and several resistances and capacitors. Power loss of the light in the optical fiber is related with the amount of bent angle, regardless of the bending direction. Methods to measure both the amount of bending and the bending direction are proposed. Figure 3. Configuration for measuring the bending direction First, if the fiber is pre-bent as shown in figure 3, we can distinguish the bending direction. Figure 5. Locations of curvature sensors The upper garment should be tight to transmit human body motion to the optical fiber curvature sensors. If the upper garment is loose, the position of the optical fiber curvature sensor relative to human body can be changed every moment. It makes it impossible to measure the human body motion from data of optical fiber curvature sensors. The position of the optical fiber curvature sensor is selected by searching the position where the amount of change of sensor data is met. 4. An Artificial Neural Network for Mapping from Sensor Data to Available Position Commands A simple artificial neural network is used for mapping function from sensor data measured from the human shoulder motion to available position commands. The neural network makes it possible that the sensor controller adjusts its mapping function according to different users. A system block diagram is shown in Fig. 6. Figure 4. Surface treatment for measuring the bending direction Second, if one side of the cladding is taken off as shown in Fig. 4, the amount of the transmitted light power varies according to the bending direction. When the fiber is bent to the harmed side, the transmitted light power is increased and when the fiber is bent to the opposite side, the transmitted light power is decreased[5]. 3. A Wearable Master Device Using Optical Fiber Curvature Sensors A tight upper garment is made to which sensors are Figure 6. System block diagram The ANN(artificial neural network) is used in the sensor controller block. The sensor controller generates the position command from sensor data. Inputs of the ANN are two sensor data. Outputs of the ANN are coordinates in the workspace. The coordinates have to be continuous for proportional control. Although we can use only direction information for controlling 2-DOF motion, the amount of direction is useful for more efficient command inputs. For example, analogue joysticks are more helpful in computer tasks than digital joysticks (directions only). Therefore, the ANN must be the inverse model of the fiber sensors block. Weights of the ANN are learned by the mechanism as shown in Fig

3 1 2 E ( n) = e j ( n) 2 j C e j ( n) = d j ( n) y j ( n), neuron j is an output node. d j (n) is the j-th desired output. y j (n) is the j-th output of the ANN. Figure 7. Learning of the artificial neural network The ANN is a multi-layered perceptron. It has one hidden layer and 6 neurons in the hidden layer (Fig. 8). Figure 8. Signal-flow graph of the ANN 5. Experiments Two experiments have been carried out with the wearable master device (upper garment) shown in section 3. The first experiment is to move a cursor to five target points. The second experiment is to control a mobile robot. In experiments, the first process is calibration. Every time the subject puts on the wearable master, the position of the wearable master w.r.t the human body is slightly changed. Since the amount of bending of a human shoulder is quite small, the change of the position of the wearable master causes the change of the coordinates of the position command dramatically. So calibration process is essential when the subject puts on the wearable master. Short calibration time is important for human-machine interface. For instance, the calibration of joysticks is very short and easy. For the wearable master, the calibration points are chosen as shown in Fig. 9. In the calibration process, the subject moves his shoulder according to the instruction of the computer. The learning-rate parameter, η is set to 0.1. The momentum constant, α is set to 0.1. The number of epochs is The activating function is as follows. 1 ϕ j ( v j ( n)) = (1) 1+ exp( v j ( n)) where v j (n) is the weighted sum of all synaptic inputs plus the bias. Teachers of the ANN are position data and sensor data that are measured during the calibration process. The learning algorithm of the ANN is a steepest descent rule including a momentum term as follows[6]. E( n) w ji ( n) = η + α w ji ( n 1) (2) w ji ( n) where w ji is the weight. η is the learning-rate parameter. α is the momentum constant Figure 9. Calibration points The first experiment is carried out as follows. To show the performance of the wearable master, a computer monitor displays five target points and one cursor point that corresponds to the movement of a right shoulder of a subject. The purpose of this experiment is that the subject should move his right shoulder to move the cursor to the 894

4 target points. An experimental result is shown in Fig. 10. Figure 10. A result of an experiment on a human body Figure 12. Mapping from position command to robot motion Fig. 10 shows that the trajectory of the cursor can be moved to the five target points very closely. The average position error is 0.32% of the workspace. The resolution is 0.14%. The elapsed time is 14 seconds. This result indicates with promise that the proposed wearable master device with optical fiber curvature sensors can be used as an input device for the disabled. The second experiment is to control a mobile robot. The robot used in this experiment is a two-wheeled mobile robot developed in Intelligent Control laboratory in KAIST as shown in Fig. 11. It has a radio frequency (RF) wireless communication module. The motors of the robot have encoders. The main computer sends velocity commands to the robot and a PID-controller in the robot drives the two motors in rate mode. The rate of the movement is proportional to exponent of the distance from neutral position, R as follows. 4R Linear velocity, V = 0.5( e 1) cm / s (3) 4 Angular velocity, ω = 0.3( e R 1) rad / s (4) The exponential mapping makes it possible that the subject can control the robot finely. The experimental result of Fig. 13 is made by overlaying frames captured every 1.2 second. The elapsed time is 24 seconds. The subject can control the speed and direction of the mobile robot comfortably. Figure 11. Two-wheeled mobile robot Fig. 12 shows how to map from the human shoulder motion to the robot motion. When the subject moves his right shoulder up, down, forward and backward, the robot moves forward and backward and turns left and right, respectively. The number of direction is fixed by four. Of course, we can make the mapping function to control the direction finely, but it makes the subject to feel more difficult to control the mobile robot. Figure 13. An experiment to control a mobile robot 6. Conclusions The development of a wearable master device for handicapped persons depends on having sensors that are small, light, and attached simply to garments. Optical fiber curvature sensors are used to measure 895

5 the human body motion. Despite the problems of ambiguousness and coupled output of optical fiber curvature sensors, useful 2 DOF motion is extracted with 2 sensors through a sensor controller using an artificial neural network. The resolution of the wearable master device is lower than 0.2% of the workspace. An experiment to control a two-wheeled mobile robot like a wheelchair is carried out successfully. With these promising results, physically handicapped persons whose arms are disabled will be able to control motorized wheelchairs or rehabilitation robotic systems with lightweight and simple wearable master device that is equipped with optical fibers. Acknowledgment The authors are grateful for the support provided by MOST, National R&D Program (Critical Technology 21), Program ID: 99-J A (Development of Service Robot Technology). REFERENCES [1] Won-Kyung Song, He-Young Lee, Jong-Sung Kim, Yong-San Yoon, and Zeungnam Bien, KARES: intelligent rehabilitation robotic system for the disabled and the elderly, Engineering in Medicine and Biology Society, Proceedings of the 20th Annual International Conference of the IEEE, Volume: 5, 1998, Page(s): vol.5 [2] J.C. Rosier, J.A. van Woerden, L.W. van der Kolk, B.J.F. Driessen, H.H. Kwee, J.J. Duimel, J.J. Smits, A.A. Tuinhof de Moed, G. Honderd, and P.M. Bruyn, Rehabilitation robotics: the MANUS concept, Advanced Robotics, 'Robots in Unstructured Environments', 91 ICAR., Fifth International Conference on, 1991, Page(s): vol.1 [3] P. Halley, LES SYSTÈMES À FIBRES OPTIQUES, Editions Eyrolles. [4] V. Louis, P. Le-Huy, J.M. Andre, M. Abignoli, and Y. Granjon, Optical Fiber Based Sensor for Angular Measurement in Rehabilitation, Systems, Man and Cybernetics, Systems Engineering in the Service of Humans, Conference Proceedings., International Conference, Vol. 1,1993, pp [5] A. Djordjevich and YuZhu He, Thin structure deflection measurement, Instrumentation and Measurement, IEEE Transactions on, Volume: 48 3, June 1999, Page(s): [6] M. Hagiwara, Theoretical derivation of momentum term in back-propagation, International Joint Conference on Neural Networks, vol. I, 1992, pp , Baltimore 896