Real-Time Joint Torque Estimation on Embedded system using EMG and Artificial Neural Network for Exoskeleton Robot

Recently, robots are being used in various fields, such as home, military, industrial robots, and in particular, the demand for exoskeleton robots that can be used in various industrial sites is gradually increasing by identifying human intentions and supporting insufficient muscle strength. Existing exoskeleton robots are controlled based on data such as estimated joint angles, torque, etc. using physical sensors such as force/torque sensors. This method of estimating torque is accurate but has a disadvantage of being slow. On the other hand, an EMG(Electromyography) signal is a biometric signal that is transmitted by electrical signals from the brain. Therefore, it is characterized by relatively inaccurate but faster measurement than actual muscle movement. In this paper, we propose a method to provide parameters in real time in a portable embedded system so that we can utilize EMG's features and utilize them directly in exoskeleton robots. This method use ANN(Artificial Neural Network) to enable estimation of faster speed as well as determination through precise mapping of EMG signals and target torque values. Finally, we propose a real-time torque estimation method that can be used complementary with existing physical sensors.