Reaching to inhibit a prepotent response: A wearable 3-axis accelerometer kinematic analysis

The present work explores the distinctive contribution of motor planning and control to human reaching movements. In particular, the movements were triggered by the selection of a prepotent response (Dominant) or, instead, by the inhibition of the prepotent response, which required the selection of an alternative one (Non-dominant). To this end, we adapted a Go/No-Go task to investigate both the dominant and non-dominant movements of a cohort of 19 adults, utilizing kinematic measures to discriminate between the planning and control components of the two actions. In this experiment, a low-cost, easy to use, 3-axis wrist-worn accelerometer was put to good use to obtain raw acceleration data and to compute and break down its velocity components. The values obtained with this task indicate that with the inhibition of a prepotent response, the selection and execution of the alternative one yields both a longer reaction time and movement duration. Moreover, the peak velocity occurred later in time in the non-dominant response with respect to the dominant response, revealing that participants tended to indulge more in motor planning than in adjusting their movement along the way. Finally, comparing such results to the findings obtained by other means in the literature, we discuss the feasibility of an accelerometer-based analysis to disentangle distinctive cognitive mechanisms of human movements.

The acceleration calibration and preprocessing analysis has been run on the data collected by an external experimenter (not part of the cohort involved in the trials) who repeated multiple selection tasks, just as a participant. Within each task, the experimenter answered to a central cue stimulus by tapping a central response key below the cue. In this way, the displacement remained roughly the same for each trial. In particular, the experimenter performed 40 trials: 1 anticipation, 2 omissions, 37 valid answers. The subsequent analysis focused on the raw acceleration signals that started when the sensor was pressed for the first trial and ended when the last valid answer was given.
The accelerometer data were sampled at 100 Hz (i.e., data sampled every 10 ms) and data were stored in g units for offline analyses.
Considering the 3-axis accelerometer, the principal output was, for each axis, the measured signal, which may be broken into the following components [1]: acquired acceleration = effective acceleration + gravity acceleration + noise.
In order to examine the true movements of the participants, we processed the acquired acceleration components to obtain their corresponding effective acceleration ones, as raw acceleration signals also contained noise, which could include an offset error, and gravity. In particular, the separation of the latter components becomes increasingly difficult during rotational movements. In fact, in the case of rotational movements (which were observed during our experimental task), the frequency domains of the movement-related component and the gravitational component can overlap, thus their separation can become challenging [1].
Resorting to state of the art approaches [1], the effective acceleration was extracted implementing the following two key steps (Table 1): (a) a band-pass filter, and, (b) an offset estimation and subtraction step. We now proceed expanding the discussion regarding their use in this work. raw accelerometer data ↓ band-pass filter ↓ offset estimation and subtraction ↓ conversion from g to m/s 2 ↓ calibrated and preprocessed acceleration Table 1. Acceleration calibration and preprocessing.
Following [1], a 4 th order Butterworth band-pass filter with cut-off frequencies equal to 0.2-15 Hz was applied to the signal. The filter cut-off frequency of 0.2 Hz was chosen on the presumption that most daily movements of human body parts occur at frequencies higher than 0.2 Hz. The cut-off frequency of 15 Hz was instead chosen to remove the effect of high-frequency noise. Also the 1-20 Hz cut-off frequencies were evaluated, considering other choices made in literature [1][2][3][4], however it was not possible to observe any meaningful difference with respect to the 0.2-15 Hz band. Comparing now the acceleration signals in Figure 1, it is possible to see that the raw acceleration components were shifted with respect to 0 g because of gravity. The z component, for example, would fall as low as −g. After applying the band-pass filter, all acceleration components adjusted to lie around 0 g.  To estimate the offset error, data was collected from the accelerometer while at rest with the x, y and z axes pointing towards the ground (Figure 2). Accelerometer at rest positions [5].
From the filtered signal, for each of the three components, we computed the mean of the differences between actual accelerometer readings and the 0 g value expected from an accelerometer at rest. We hence obtained an offset value for each of the three axes. Successively, such values were removed from the acceleration data components, according to the pseudocode reported in Algorithm 1.

2/4
Algorithm 1 Accelerometer offset 1: procedure (for each axis x, y, z) 2: i in (x, y, z)  Finally, we obtained an estimate of the effective acceleration, adopting g = 9.80665 m/s 2 for the conversion from g to m/s 2 units.