| Summary: | A personal-computer-based and a Raspberry Pi single-board computer-based virtual force sensor with EtherCAT communication for a six-axis robotic arm are proposed in this paper. Both traditional mathematical modeling and machine learning techniques are used in the establishment of the dynamic model of the robotic arm. Thanks to the high updating rate of EtherCAT, the machine learning-based dynamic model on a personal computer achieved an average correlation coefficient between the estimated torque and the actual torque feedback from the motor driver of about 0.99. The dynamic model created using traditional mathematical modeling and the Raspberry Pi single-board computer demonstrates an approximate correlation coefficient of 0.988 between the estimated torque and the actual torque. The external torque observer is established by calculating the difference between the actual torque and the estimated torque, and the virtual force sensor converts the externally applied torques calculated for each axis to the end effector of the robotic arm. When detecting external forces applied to the end effector, the virtual force sensor demonstrates a correlation coefficient of 0.75 and a Root Mean Square Error of 12.93 N, proving its fundamental competence for force measurement. In this paper, both the external torque observer and the virtual force control are applied to applications related to sensing external forces of the robotic arm. The external torque observer is utilized in the safety collision detection mechanism. Based on experimental results, the system can halt the motion of the robotic arm using the minimum external force that the human body can endure, thereby ensuring the operator’s safety. The virtual force control is utilized to implement a position and force hybrid controller. The experimental results demonstrate that, under identical control conditions, the position and force hybrid controller established by the Raspberry Pi single-board computer achieves superior control outcomes in a constant force control scenario with a pressure of 40 N. The average absolute error is 9.62 N, and the root mean square error is 11.16 N when compared to the target pressure. From the analysis of the results, it can be concluded that the Raspberry Pi system implemented in this paper can achieve a higher control command update rate compared to personal computers. As a result, it can provide greater control benefits in position and force hybrid control.
|
|---|