Acceleration-Bounded Position-Based Visual Servoing for Non-Overshooting Convergence and Responsive Tracking

抄録

This letter proposes an acceleration-bounded position-based visual servoing controller that suppresses overshoot during large-error convergence while preserving responsive target tracking. The proposed method introduces a `proxy` feature vector as an intermediate reference for a proportionalderivative feedback law. The proxy feature vector converges toward the desired feature with overdamped, first-order dynamics as long as the acceleration is within a prescribed admissible set, while it stays close to the current feature vector when the acceleration is saturated. The continuous-time representation of the controller is given as a differential-algebraic inclusion, from which a closed-form discrete-time implementation is derived via backward Euler discretization. Experiments using a sixdegree-of-freedom UR3e manipulator with an eye-in-hand camera illustrate non-overshooting convergence under large initial errors, while retaining tracking responsiveness comparable to a conventional PD controller for a manually moved target.