Inertial measurement units (IMUs) are increasingly used to estimate movement quality and quantity to the infer the nature of motor behavior. The current literature contains several attempts to estimate movement smoothness using data from IMUs, many of which assume that the translational and rotational kinematics measured by IMUs can be directly used with the smoothness measures spectral arc length (SPARC) and log dimensionless jerk (LDLJ-V). However, there has been no investigation of the validity of these approaches. In this paper, we systematically evaluate the use of these measures on the kinematics measured by IMUs. We show that: (a) SPARC and LDLJ-V are valid measures of smoothness only when used with velocity; (b) SPARC and LDLJ-V applied on translational velocity reconstructed from IMU is highly error prone due to drift caused by integration of reconstruction errors; (c) SPARC can be applied directly on rotational velocities measured by a gyroscope, but LDLJ-V can be error prone. For discrete translational movements, we propose a modified version of the LDLJ-V measure, which can be applied to acceleration data (LDLJ-A). We evaluate the performance of these measures using simulated and experimental data. We demonstrate that the accuracy of LDLJ-A depends on the time profile of IMU orientation reconstruction error. Finally, we provide recommendations for how to appropriately apply these measures in practice under different scenarios, and highlight various factors to be aware of when performing smoothness analysis using IMU data.

Appropriately responding to mechanical perturbations during gait is critical to maintain balance and avoid falls. Tripping perturbation onset during swing phase is strongly related to the use of different recovery strategies; however, it is insufficient to fully explain how strategies are chosen. The dynamic interactions between the foot and the obstacle may further explain observed recovery strategies but the relationship between such contextual elements and strategy selection has not been explored. In this study, we investigated whether perturbation onset, duration and side could explain strategy selection for all of swing phase. We hypothesized that perturbations of longer duration would elicit lowering and delayed-lowering strategies earlier in swing phase than shorter perturbations. We developed a custom device to trip subjects multiple times while they walked on a treadmill. Seven young, healthy subjects were tripped on the left or right side at 10% to 80% of swing phase for 150 ms, 250 ms or 350 ms. Strategies were characterized by foot motion post-perturbation and identified by an automated algorithm. A multinomial logistic model was used to investigate the effect of perturbation onset, side, and the interaction between duration and onset on recovery strategy selection. Side perturbed did not affect strategy selection. Perturbation duration interacted with onset, limiting the use of elevating strategies to earlier in swing phase with longer perturbations. The choice between delayed-lowering and lowering strategies was not affected by perturbation duration. Although these variables did not fully explain strategy selection, they improved the prediction of strategy used in response to tripping perturbations throughout swing phase.