ScholarWorks@성균관대학교

초록

Conventional dual-robot calibration methods typically ignore uncertainty in robot-pose measurements and suffer from degradation of calibration accuracy. To overcome these limitations, we present a reprojection-based calibration framework for dual robotic systems that explicitly models robot-pose uncertainty within the extended kinematic chain formulation AXB=YCZD. The method jointly estimates the unknown transforms X, Y, and Z, as well as the robot poses. Simulations show the proposed method consistently outperforms a Lie derivative-based method in calibration accuracy. Furthermore, separately estimating the per-robot variance components significantly improves robot- pose estimation accuracy, especially when their noise characteristics differ. Real-world experiments using a Franka Research 3 dual-robot system validate the method’s practical applicability. Our approach outperforms the Lie derivative-based baseline in translation, rotation and reprojection accuracy on both training and test sets. These results confirm that modeling robot-pose uncertainty enables robust dual-robot calibration.

키워드