Assessment of lower limbs kinematics during human–robot interaction using inertial measurement units

Abstract

Introduction: The kinematic structure of wearable robots designed for lower limb rehabilitation usually replicates that of the human body (anthropomorphism), thus kinematic incompatibilities may arise from the mismatch between robotic and human joint axes of rotation. Conversely, non-anthropomorphic structures can be designed to be robust against misalignments, thus minimizing undesired interaction forces [1]. A non-anthropomorphic treadmill-based robot for hip and knee flexion/extension assistance, using compliant actuators [2] and deriving human joint kinematic patterns with forward kinematics, has been previously presented [3] and [4]. Inaccuracies in the mapping of rotations from the robot joint space to the human joint space may arise from: (1) human segments length estimation errors, which affect the human–robot parallel kinematic chain, (2) non-rigid human–robot connections and (3) uncertainty associated to the position of fastening points. The aim of this work is to validate the human joints kinematics estimated by the robot (RK) using that provided by inertial measurement units (IK) as reference. Methods:Ahealthy subject (male, 28 y.o.) walked on a treadmill at 0.55, 0.7 and 0.85 m/s wearing the robot. Four inertial measurement units (IMUs – Xsens, MTX, sampling frequency 50 Hz) were attached to the pelvis, thigh, and shank cuffs and to the foot on the right side (Fig. 1). The RK of hip and knee were estimated from the encoders measurements (sampling frequency 200 Hz) using the third-order polynomial approximation of the forward kinematic function derived in [3]. At the beginning of each trial, while the subject was standing (reference configuration), thigh and shank IMUs were mathematically aligned to the pelvis IMU. Hip and knee IK were estimated in terms of sagittal components of the relevant orientation vectors provided by the IMUs. Gait cycles were identified using the acceleration peaks at heel strike. Data from the encoders and IMUs were synchronized using an external trigger. The root mean square deviation (RMSD) between hip and knee RK and IK was computed for the three trials over 10 gait cycles. Results: Representative patterns of hip and knee angles in the sagittal plane (treadmill velocity: 0.55 m/s) are reported. RMSD values averaged over the three different gait speeds were 2.9±0.2◦ for the hip and 4.9±1.2◦ for the knee (mean±sd). In all the trials, maximum discrepancies were found at the peak values (RK peak extension angles were overestimated). Discussion: Non-anthropomorphic robots can offer several advantages in terms of wearability but they may introduce some inaccuracies in the estimation of human joints kinematics. However, results showed that hip and knee RK errors were within levels acceptable in the majority of clinical applications. Moreover, the systematic errors observed can be corrected through a proper calibration of the robot-to-human mapping. The estimate of lower limb kinematics from IMUs integrated in the robotic cuffs could facilitate the control of the robot.