IBVS vs PBVS for Aerial Robotics: Key Differences and Real-World Challenges

Visual servoing is essential for aerial robotics tasks where GPS alone falls short—such as landing on moving platforms, grasping objects, or close-up turbine inspection. The two main classical approaches are Image-Based Visual Servoing (IBVS) and Position-Based Visual Servoing (PBVS). IBVS computes control signals directly from pixel coordinates, avoiding 3D reconstruction, which makes it robust to camera calibration errors and computationally efficient. However, it suffers from image singularities and unpredictable trajectories, including the counterintuitive "camera retreat." PBVS first estimates the target's 3D pose before sending Cartesian-space error to the controller, offering intuitive trajectory planning but demanding highly accurate pose estimation and being sensitive to calibration noise.

Real-world deployment exposes several critical problems ignored by common simulator assumptions. Lens shading (vignetting) and miscalibrated gain, exposure, or focus systematically shift pixel coordinates, introducing steady-state errors that degrade CV-based detection and tracking. Limited field of view (FoV) is another major challenge—during agile maneuvers with underactuated drones, losing the target breaks the visual control loop entirely. Additionally, real cameras suffer from vibration and non-instantaneous shutter effects. Advanced solutions like Model Predictive Control (MPC) with visual penalty terms or online trajectory replanning are needed to maintain visibility and robustness. These gaps underscore that classical visual servoing must be adapted for high-speed, safety-critical aerial platforms.