Lower limb exoskeletons require the correct support magnitude and timing to achieve user assistance. This study evaluated whether the sign of the angular velocity of lower limb segments can be used to determine the timing of the stance and the swing phase during walking. We assumed that stance phase is characterized by a positive, swing phase by a negative angular velocity. Thus, the transitions can be used to also identify heel-strike and toe-off. Thirteen subjects without gait impairments walked on a treadmill at speeds between 0.5 and 2.1 m/s on level ground and inclinations between −10 and +10°. Kinematic and kinetic data was measured simultaneously from an optical motion capture system, force plates, and five inertial measurement units (IMUs). These recordings were used to compute the angular velocities of four lower limb segments: two biological (thigh, shank) and two virtual that were geometrical projections of the biological segments (virtual leg, virtual extended leg). We analyzed the reliability (two sign changes of the angular velocity per stride) and the accuracy (offset in timing between sign change and ground reaction force based timing) of the virtual and biological segments for detecting the gait phases stance and swing. The motion capture data revealed that virtual limb segments seem superior to the biological limb segments in the reliability of stance and swing detection. However, increased signal noise when using the IMUs required additional rule sets for reliable stance and swing detection. With IMUs, the biological shank segment had the least variability in accuracy. The IMU-based heel-strike events of the shank and both virtual segment were slightly early (3.3–4.8% of the gait cycle) compared to the ground reaction force-based timing. Toe-off event timing showed more variability (9.0% too early to 7.3% too late) between the segments and changed with walking speed. The results show that the detection of the heel-strike, and thus stance phase, based on IMU angular velocity is possible for different segments when additional rule sets are included. Further work is required to improve the timing accuracy for the toe-off detection (swing).

This paper presents a method for tailoring a parametric controller based on human ratings. The method leverages supervised learning concepts in order to train a reward model from data. It is applied to a gait rehabilitation robot with the goal of teaching the robot how to walk patients physiologically. In this context, the reward model judges the physiology of the gait cycle (instead of therapists) using sensor measurements provided by the robot and the automatic feedback controller chooses the input settings of the robot so as to maximize the reward. The key advantage of the proposed method is that only a few input adaptations are necessary to achieve a physiological gait cycle. Experiments with non-disabled subjects show that the proposed method permits the incorporation of human expertise into a control law and to automatically walk patients physiologically.