Visual-Biased Observability Index for Camera-Based Robot Calibration

Abstract

Abstract Efficient robot integration can be realized by matching real and virtual robots, and accurate robot models can be generated by kinematic parameter calibration. End-effector pose selection for pose measurement to discover the positioning errors is critical in kinematic parameter calibration. Ideal pose selection maximizes calibration accuracy for a defined measurement uncertainty and optimizes measurement cost and utility. In the design of the pose selection process, observability indices are widely accepted criteria for effective pose selection to evaluate calibration performance. Observability indices represent the effect of uncertainty in the measured end-effector poses on the calibrated parameters. However, unlike expensive direct measurement using laser, low-cost camera-based kinematic calibration estimates the end-effector poses from the marker points in the captured image. The variance of the detected marker positions biases the end-effector poses and, eventually, the calibrated parameters. Therefore, this study proposes extended observability indices for pose selection based on this bias to realize accurate calibration with a low-cost camera. The target observability index is O1, a scale-free, reliable index used in kinematic calibration. Considering the visual bias, we extended it as Ov1. This study evaluated Ov1 by comparing the positioning accuracies calibrated on poses selected by maximizing it, original O1, O3 known as the best criterion to restrain the end-effector positioning uncertainty, and Ov3, which is the extended O3 for consistency. A ball-bar test showed that the poses selected by the index Ov1 exhibited higher positioning accuracy than the other indices.