Estimation of Wrist Force/torque for Robot Gripper using Neural Network

This paper proposes a kind of estimation method of wrist force/torque for robots. The method adopts the data fusion technique according to the output variations of the finger force sensors installed in the gripper. The finger force sensors are used to measure the clamping force of the gripper in the design. When the accuracy of measurement is not required exactly and there is the limitation of weight and volume in space robots, we utilize the existing devices to estimate the wrist force/torque without the wrist force/torque sensor, which not only meets the practical requirement but also decreases the weight and cost of robots. An experimental bench is developed and the calibration experiments are conducted to detect the relationship between the wrist force/torque and finger forces. The experimental data are used to train a Back Propagation (BP) artificial neural network, and the construction and parameters of the network are obtained. The results of data fusion of the wrist force/torque are consistent with the practical calibration values, and the effectiveness of the wrist force/torque estimating technique is proved.

I. Introduction

In the space, robots often assist or replace the astronaut to do some tasks out of the spaceship, for example, checking the space station, grasping and operating the work-pieces, and catching the floating objects, etc. When robot takes these operations, the wrist of its gripper will receive the force. If the force exceeds its limitation, the flange, at the root of gripper, will be fractured, even the robot probably separates itself from the experimental platform and becomes the space rubbish[1]∼[3], this situation is very danger. Therefore, to ensure the safety of operation of robots, it is necessary to monitor the wrist force/torque for robot gripper, in this situation the high measuring accuracy is required. For this purpose wrist force/torque sensors are developed [4]∼[6]. In theory, the wrist/torque sensor can measure the force precisely. However, for the specialty of space robot, the wrist force sensor sometimes can not be equipped on it because of the limitations of its weight and dimension. Can other sensors on the gripper be utilized to measure the wrist force/torque? We found that, the output variations of finger force sensors can reflect the value of the wrist force/torque, but their relationship is very complex and difficult to be described using a mathematical model by the analytical method, therefore, this paper presents a novel method that utilize neural network to describe the relationship since ANN possesses the good property of approximating any functions.