A Hybrid Adaptive Visual Servoing Approach for Robust Robotic Manipulation in Space

Visual servoing is a core capability for autonomous robotic manipulation in space applications like in on-orbit servicing (OOS) operations, yet traditional methods degrade under visibility changes, motion constraints, or unreliable depth. We propose a Hybrid Adaptive Visual Servoing (HAVS), combining coarse eye-to-hand (ETH) position-based with fine eye-in-hand (EIH) 2.5D servoing. A blending strategy is used to ensure that transitions avoid discontinuities, and pose estimation selectively fuses localized depth when reliable. HAVS is validated in a high-fidelity dynamic simulation of Redwire Space's STAARK manipulator for space operations. Results show that HAVS achieves smoother control, lower energy consumption, and faster convergence compared to classical hybrid-switching, while maintaining accuracy and feature visibility. Its depth strategy improves behavior over pure 2D and is more efficient than full 3D, making HAVS well suited for precision OOS.