Driving performance was evaluated in terms of speed, path root-mean-square error (RMSE), maximum path error, and path smoothness, which was calculated using the Spectral Arc Length (SPARC) of the wheelchair velocity [
8]. Arm movement performance was evaluated in terms of arm movement amplitude, arm movement frequency, arm movement synchrony, and interaction force. Arm movement synchrony was measured calculating the cross-correlation coefficients of the left and right arm movement signals, and interaction force was calculated as the average peak force for each arm movement repetition. Finally, exercise intensity was evaluated in terms of heart rate increase, which was measured using a wearable chest strap sensor (Polar H10, Polar Electro Oy, Finland). Resting heart rate was measured at the beginning of the experiment by asking the participant to relax for five minutes and calculating the average value over the last minute. Heart rate increase was calculated taking the average heart rate value during a driving task and subtracting the resting heart rate from it.
All the driving performance metrics were calculated from the path data of the wheelchair simulator for the virtual driving tasks. For the real driving tasks, the path data was measured using a motion capture system with nine cameras (Optitrack Flex 3, NaturalPoint Inc., USA) and two reflective markers mounted on the head rest of the powered wheelchair.
To compare the performance metrics (
Score) of the two
Groups of participants (Unimpaired and DMD) with the two control
Inputs (joystick and MOVit 2.0), a linear mixed-effects analysis was conducted on all metrics for each driving task (i.e., virtual and real driving tasks). For the driving performance metrics, we modeled
Input and
Group (and their interaction) as fixed effects, and used an error term with random intercepts grouped by
Subject (Eq.
1). For the arm movement performance metrics, we modeled
Group as fixed effect, and used an error term with random intercepts grouped by
Subject (Eq.
2). Analysis of variance (ANOVA) tests were used to compare the performance scores for each of the performance metrics, and Bonferroni tests were applied for pairwise comparison. Statistical analysis of the questionnaire results was performed with paired t-tests. We used
\(\alpha = 0.05\) as the level of significance. Statistical analyses were carried out using R 3.5.0 [
9] with lme4: Fitting Linear Mixed-Effects Models [
10], lmerTest: Tests in Linear Mixed-Effects Models [
11], and lsmeans: Least-Squares Means [
12].
$$\begin{aligned}&Score \sim Group * Input + (1|Subject) \end{aligned}$$
(1)
$$\begin{aligned}&Score \sim Group + (1|Subject) \end{aligned}$$
(2)