Reliable high precision positioning for intelligent machines

Abstract

In order for intelligent machines to proceed successfully from perception to action, the high-precision positioning accuracy of robotic manipulators is considered. First, accuracy estimates for two algorithms are derived and shown to depend upon the condition number of the manipulator Jacobian. Statistics for the positioning error are estimated using maximum-likelihood estimation. These statistics are used, along with a set of specifications, to generate entropy constraints which determine feasible algorithms. Reliability performance functions are then defined, and the reliability of each algorithm as a function of joint position is derived. The concepts are validated using a simulation of the Puma 560 kinematics. >