A novel step-by-step procedure for the kinematic calibration of robots using a single draw-wire encoder

TL;DR

This work tackles the problem of improving robot positioning accuracy by introducing a novel step-by-step kinematic calibration that uses only 1D distance data from a single draw-wire encoder. The method fixes the encoder location in joint space and designs calibration-point sets that isolate each unknown parameter, enabling sequential, non-iterative parameter identification within a DH-based kinematic model with $m=10$ error terms $\bigl(\delta\theta_2,\delta\theta_3,\delta\theta_4,\delta\theta_5,\delta\theta_6,\delta a_1,\delta a_2,\delta a_3,\delta d_4,\delta d_6\bigr)$. Experimental validation on a 6-DOF Adept Viper S650 shows substantial improvements: after calibration, the maximum distance discrepancy across configurations drops by ~84% and plane-fit residuals shrink from <0.32 mm to <0.05 mm, with corresponding reductions in mean and standard-deviation errors. The approach is cost-effective, computationally light, and implementable directly in robot controllers, offering a practical alternative to laser-tracker-based calibration in industrial settings.

Abstract

Robot positioning accuracy is a key factory when performing high-precision manufacturing tasks. To effectively improve the accuracy of a manipulator, often up to a value close to its repeatability, calibration plays a crucial role. In the literature, various approaches to robot calibration have been proposed, and they range considerably in the type of measurement system and identification algorithm used. Our aim was to develop a novel step-by-step kinematic calibration procedure - where the parameters are subsequently estimated one at a time - that only uses 1D distance measurement data obtained through a draw-wire encoder. To pursue this objective, we derived an analytical approach to find, for each unknown parameter, a set of calibration points where the discrepancy between the measured and predicted distances only depends on that unknown parameter. This reduces the computational burden of the identification process while potentially improving its accuracy. Simulations and experimental tests were carried out on a 6 degrees-of-freedom robot arm: the results confirmed the validity of the proposed strategy. As a result, the proposed step-by-step calibration approach represents a practical, cost-effective and computationally less demanding alternative to standard calibration approaches, making robot calibration more accessible and easier to perform.

Figures (8)

• Figure 1: System at the configuration specified by the joint coordinates $\bm{\theta_0}$ (left) and at the $i$th calibration configuration, specified by the joint coordinates $\bm{\theta_i}$ (right).
• Figure 2: Flowchart of the proposed step-by-step calibration procedure
• Figure 3: Experimental setup during the calibration procedure: system at the configuration specified by the joint coordinates $\theta_0$ (left) and at the $i$th calibration configuration, specified by the joint coordinates $\theta_i$ (right).
• Figure 4: Experimental setup during the first validation procedure, where the robot is moved through a set of poses in the workspace, changing its configuration: "above" and "noflip" (left), "below" and "flip" (right)
• Figure 5: Experimental setup during the second validation procedure: the touch probe detects the contact with the granite surface plate (the LED light turns red) and the corresponding end-effector coordinates are recorded.