4-D body posture in sagittal plane for care-receiver patient can be estimated by using only two low-reolution tactile sensors mounted on robot arms

Abstract

In order to contribute in a more convenient transfer of the mobility impaired patients for both patients and caregivers, a two-armed nursing-care assistant robot for transferring patients between bed and wheelchair has been developed in our center. For the safety of movement it is essential for the robot to have an accurate perception of how the patient is located on its arms, from the beginning of the lift up till the end of put down. This study is a contribution in accuracy of the posture estimation, given very limited sensor information. We have developed a flexible model-free method for real-time estimation of the care-receiver's posture by this robot. This algorithm uses no sensor but a pair of soft low-resolution tactile sensors installed on the robot's forearms. The performance of this algorithm is evaluated by experiment, showing good accuracy (average and root mean squared error of about 0.4 and 2.4 degrees, respectively) as well as acceptable real-time computation time with respect to the robot control cycle (less than 10 milliseconds).