Jitter removal in KUKA KR-5 using Modified Kalman Filter while tele-operating with Exoskeleton

摘要

This paper aims at removing noise/jitters with the help of Kalman filter technique. The application being considered is that of an industrial robot KUKA KR-5 to be remotely controlled by an exoskeleton. The exoskeleton is a wearable device, which senses the motion of the human limb, through the various sensors and potentiometers embedded in the design. Signals from the human subject’s elbow is reflected using a one DOF exoskeleton and its control has been discussed [3]. The external forces acting on the remote/virtual arm is replicated on the arm exoskeleton system developed in our laboratory [4]. This motion is then mapped to the KUKA. The motion of the human limb being sensed by the exoskeleton, is observed to be noisy. There is a significant amount of jittering observed even when the exoskeleton is maintained static and motion is not provided. This causes the KUKA to respond to the exoskeleton’s ‘no motion’ state through a visible vibration. A minute vibration in the exoskeleton (caused due to a slight activity in the human arm) causes noise propagation during exoskeleton tele-operation. It could be felt at the remote station manipulator by a noticeable displacement of 2mm in the KUKA KR5 robot. This causes disruption in the motion mapping between the two. This removal of this noise/jitter is being aimed at in order to precisely control the motion of the robot. There are standard filters that were implemented to take care of the noise propagation. Butterworth, Chebyshev and median filter were implemented which can be used for noise removal/reduction. This implies suppressing or removing certain frequencies causing interference. The standard filters generally introduce an initial delay in the system which varies as the order of the filter varies. Hence the system response is improved but lags behind the actual response by a few seconds (variable). Kalman filter is robust against uncertainty in process and noise covariance [2]. The Kalman filter works on a predict-update mechanism. It uses the previous state and the current measurement to predict the current state. Thus, the delay can be avoided by providing a good approximation of the previous (a-priori) state estimate. Hence, the Kalman filter can be considered to be a delayless system. Moreover, Kalman filter is computationally efficient and also minimizes mean square error [1] The Kalman filter algorithm works in two-step procedure: Predict and Update. An important requirement for implementing a Kalman filter is that the dynamics of the system under consideration should be known well in order to define a model that defines the dynamics of the system accurately. It mainly estimates the state of unknown variables using a series of consecutive measurements over a defined period of time and with a known sample time. These series of measurements give a precise estimate of the next state of the system as compared to that obtained using a single measurement. The estimate thus obtained as a result is more stable and thus the filter is also known to smooth out the output, removing variations due to the noise present. Kalman filter designed for real-time estimation of the orientation of human limb segments