Detecção de escorregamento em tempo real para controle de uma garra robótica, utilizando Machine Learning

Abstract

The correct handling and gripping of objects depend on the precise control of the applied force, together with the slip detection of the object to be held. The implementation of tactile sensing for slip detection in the control of the gripping force can reduce the costs of handling objects in the industry and bring more safety to those who operate with these machines. Thus, an object gripping system was developed that acts in only one direction, it receives signals from two pressure sensors and an acoustic sensor, and it has a stepper motor that controls the gripping force exerted on the object. The developed system has its operating principle based on the slip detection, in real-time, on the contact surface of the object, through the analysis of the signal from the acoustic sensor, positioned on the surface, in addition to the detection of the grip strength. A supervised learning algorithm was used as a classifier and performed the function of detecting the presence of slipping on the device’s surface. Thus, the control system of the grip strength on the object to be held is based on the classifier’s response to the presence of slipping. Its answer, if positive, implies a fixed incremental gain in the force exerted during the grip. In the end, the performance of the online grip control system is analyzed for different objects and it is observed that the system can maintain the grip stable for objects with greater mass and high rigidity.