Intelligent Visual Servoing Using Learning-Based Grasping Configurations and Adaptive Controller Gain

For the traditional image-based visual servoing (IBVS) system, the desired configurations of grasping are predefined for the robot and the gain $\lambda$ in the control law is fixed. While training the robots with new skills, the predefined grasping configurations may increase the workload; meanwhile, the fixed controller gain $\lambda$ may slow down the convergence of the system. To this end, in this paper, an intelligent IBVS system using learning-based grasping configurations and adaptive controller gain is presented. With the Gaussian Mixed Model (GMM), the robot is capable of learning the desired grasping configurations through a certain number of demonstrations. To accelerate the convergence of the IBVS system, a fuzzy logic (FL) unit is applied to adjust the gain $\lambda$. The simulation results demonstrate that the proposed approach enables the IBVS system gain the learning ability on grasping configurations, besides accelerating the convergence.