This article deals with the subject of increasing the positioning accu
racy of a manipulator beyond what is typically obtained via Denavit-Ha
rtenberg (D-H) calibration. It is commonly accepted that one method fo
r accomplishing this requires including ''extra'' terms (such as stati
c deflection) in the D-H model. Another technique, which is experiment
ally verified and quantified in this article, is that of local calibra
tion. Rather than continually introducing more and more modeling terms
, we restrict the worlvolume of the manipulator. One contribution of t
his article is an experimental investigation of local calibration. Mos
t data indicates that local calibration offers advantages and it quant
ifies that the approximate improvement on one particular PUMA 560 robo
t is as much as four times better than global calibration. Some data i
ndicates that it is not always true that ''localizing'' the calibratio
n is an improvement. The article also discusses the problem of recogni
zing bad calibration data. We demonstrate a technique for identifying
and removing ''bad'' data. Finally, the article shows that the localit
y of workvolume must include a consideration of the joint motions as w
ell. For example, it presents results when data is taken with the arm
in a single configuration, and in multiple configurations. (C) 1995 Jo
hn Wiley & Sons, Inc.