When the joint position, velocity, and acceleration dataare free of noise, which is for practical purposes realizedby applying the above-mentioned signal processing, theidentification matrix in (2) is free of noise, and the maximumlikelihood estimation simplifies to weighted linearleast squares estimation (WLSE). The complexity of WLSEis comparable to that of LLSE, in fact, the only difference isthat WLSE weighs the data with the inverse of the covarianceof the actuator torque measurement noise and there-OCTOBER 2007 « IEEE CONTROL SYSTEMS MAGAZINE 63FIGURE 4 A KUKA IR 361 industrial robot in the robotics laboratoryat KU Leuven. The position controller of the robot is provided byOrocos software [24], [25]. This software facilitates the applicationof periodic commanded trajectories, while synchronizing encoderand actuator current measurements.fore discriminates between accurate and inaccurate data.