Image-Based Composite Learning Robot Visual Servoing With an Uncalibrated Eye-to-Hand Camera

Adaptive visual servoing based on a depth-independent interaction matrix is effective for uncalibrated robot visual servoing when both the intrinsic and extrinsic parameters of a camera are unknown. But current results on this topic omit the analysis of parameter convergence that is beneficial to improving the overall performance and robustness of robot control systems. This article proposes an image-based visual servoing method for robot regulation under an uncalibrated eye-to-hand camera, where composite learning is integrated organically to enhance parameter convergence. The proposed method guarantees asymptotic convergence of pixel errors under no excitation and asymptotic convergence of both pixel errors and parameter estimation errors under a condition termed interval excitation that is strictly weaker than persistent excitation. To fully exploit the merits of composite learning, we introduce scaling to balance unknown camera parameters and regularization to improve estimation robustness. Simulations and experiments on a collaborative robot with seven degrees of freedom named Franka Emika Panda have verified the superiority of the proposed method in significantly enhancing parameter convergence and, in turn, considerably improving control performance.