Minimizing Human Tracking Error for Robotic Rehabilitation Device

Abstract

This paper explores the design of a novel robotic device for gait training and rehabilitation, a method to estimate a human's orientation within the rehabilitation device, as well as an optimal state space controller to actuate the rehabilitation device. The robotic parallel bars (RPBs) were designed to address the shortcomings of currently available assistive devices. The RPB device moves in response to a human occupant to maintain a constant distance and orientation to the human. To minimize the error in tracking the human, a complementary filter was optimized to estimate the human's orientation within the device using a magnetometer and gyroscope. Experimental measurements of complementary filter performance on a test platform show that the filter estimates orientation with an average error of 0.62 deg over a range of angular velocities from 22.5 deg/s to 180 deg/s. The RPB device response was simulated, and an optimal state space controller was implemented using a linear quadratic regulator (LQR). The controller has an average position error of 14.1 cm and an average orientation error of 14.3 deg when tracking a human, while the simulation predicted an average error of 10.5 cm and 5.6 deg. The achieved level of accuracy in following a human user is sufficiently sensitive for the RPB device to conduct more advanced, realistic gait training and rehabilitation techniques for mobility impaired patients able to safely support their body weight with their legs and arms.