To verify our proposed method, we try to find the acupressure points of the different subjects with different sizes and compare their results with the real acupressure points which are defined by a massage therapist in advance. The method for determining the real acupressure points is that the massage therapist first recognizes the acupressure points of each subject; and then we use the color labels to mark these points and get the real acupressure points through Algorithm 1 as mentioned before. We put the laser range finder in front of the massage chair 1m away and ask the subject to sit upright in the massage chair for scanning. There are ten subjects in our experiments. Fig.
5-9 shows the 3D model of the hefty subject and the slim subject respectively. The green dots on their backs indicate the real acupressure points and the red dots are the mapped acupressure points from the method we propose. We find that the mapped points are affixed to the surface accurately and the results are very close to the real ones.
The average distance error of the acupressure points from ten subjects is 10.7mm that is smaller than the contact radius (17mm) of the sphere-shaped end-effector and the standard deviation of these errors is 2.02mm which means our algorithm can cover all
acupressure points. The root mean square error of the position and the average distance error of the acupressure points for three body types are displayed in Table 5-2. We observe that larger people have bigger errors because their body shape is much more different with average sized people and we should give more features in morphing step to get better fitting. Since the subjects cannot sit exactly upright in the massage chair, it will have a slight rotation along z-axis and yield more significant deviations on the x-axis than y-axis. The experimental results prove our method can precisely generate adaptive trajectory for different people and can be applied in practical applications.
Besides the acupressure points for pressing, we generate the other trajectories for rubbing and stroking through the 6 points as shown in Fig. 5-10 and Fig. 5-11. The radius of rubbing is 3cm and its trajectories are indicated as yellow lines. The trajectories for stroking as magenta lines are planned from the upper points to the lower points and these paths have many acupressure points which are good for our health.
(a) Hefty Subject (Subject #1)
(b) Slim Subject (Subject #3)
Fig. 5-9 The 3D model of the subjects with different body shape. The green dots on their backs indicate the real acupressure points and the red dots are the mapped acupressure points from the method we propose.
Subject
Table 5-2 Errors of trajectory generation
We also record the time that our method takes. The average consuming time of teach and play is more than 75s and the consuming time of our method is 35s. It figures out our method can save more than half the time needed with teach and play.
Fig. 5-10 The rubbing trajectories for different subjects.
Compare to our previous work [16] as shown in Fig. 5-12. We can apparently find that the trajectories generating from our previous work are very tortuous which are difficult to be applied to practical applications. If the end-effector exactly follows these trajectories, there will be control problems and it will also affect the effectiveness of massage. Because our previous work only use two planes as features for trajectory planning, the trajectories have large offset which are not corresponding to the real trajectories used in general therapeutic massage.
Fig. 5-12 The trajectories generating from our previous work.
In contrast to the previous work, the trajectories from the method in this paper have higher accuracy and are more flat. Furthermore, the acupressure points in this paper are certified by physician that our previous work does not have.
Finally, we apply the self-developed LRF and the adaptive trajectory generation algorithm to the dual robot for massage. There are two scenarios. One is upper body massage that the subject sits upright in the massage chair during massage. Another is full body massage that the subject lies in the massage bed during massage. The dual robot determines the position of acupressure points according to the results of scanning.
The relation between the coordinate of the LRF and the coordinate of the dual arm robot is found by camera calibration. The translation is measured by tape measure and the orientation is calculated through the intrinsic parameters and extrinsic parameters of camera. Fig. 5-13 shows the process of the upper body massage.
Fig. 5-13 Apply self-developed LRF to upper body massage. Left: The LRF is mounted in front of the dual arm robot. Upper Right: The LRF is scanning the subject’s back.
Lower Right: The dual arm robot executes pressing on subject’s back.
Because each standard model in our algorithm is suitable for only one pose, we rebuild the 3D model for full body massage as shown in Fig. 5-14. In order to be closer
to the real situation, the subject is asked to be bare to the waist. Fig. 5-15 shows the process of the full body massage.
Fig. 5-14 The 3D model of the subject lying on the massage bed.
Fig. 5-15 Apply self-developed LRF to full body massage. Left: The LRF is mounted with a tilt angle in front of the massage bed. Upper Right: The LRF is scanning the subject’s back. Lower Right: The dual arm robot executes pressing on subject’s back.
Chapter 6 In-Situ Tissue Stiffness Estimation
The main issue of physical human-robot interaction in robotics massage is how to adjust the control strategies according to tissue stiffness to provide patients more comfortable and friendlier service. In this chapter, we first figure out the characteristics of soft tissue by using a 7 DoF robot arm and build an intuitive soft tissue model which have the information of hardness and thickness. Then we present an in-situ tissue stiffness estimation method which can estimate the characteristics of patient’s soft tissue during massage. The algorithm is based on adaptive law introduced in this chapter as well and it is verified through simulation and experiment.