SurgURDF-CalibLearning Robust Geometric Optimization for Surgical Robot Camera Calibration

We present SurgURDF-Calib, a geometry-aware learned optimizer for surgical robot–camera calibration that bridges robot kinematics and visual observations through structured geometric reasoning. Instead of directly regressing camera poses, SurgURDF-Calib learns how to select, weight, and optimize geometric constraints while preserving an analytical SE(3) solver. Together with SurgURDF-100K, a large-scale geometry-consistent benchmark with URDF and calibration annotations, our framework enables robust calibration across diverse surgical instruments and real-world environments.

🌟 Key Highlights

1. 🔗 Geometry-Structured Optimization

   Integrates URDF kinematics, joint states, and visual observations into a unified calibration framework, enabling geometry-consistent camera–robot alignment.

2. 🎯 Observability-Aware Frame Scheduling

   Selects informative observations from dense surgical videos through D-optimal geometric reasoning, reducing redundant constraints under limited frame budgets.

3. ⚙️ Learned Robust Sparse LM Refinement

   Learns geometric factor reliability and parameter-wise solver conditioning while preserving an analytical SE(3) optimization process.

4. 🦾 SurgURDF-100K Benchmark

   Provides 100K geometry-consistent surgical observations with synchronized URDF models, motion states, poses, and camera calibration annotations.