Augmented feedback for powered wheelchair training in a virtual environment

Abstract

Background: Powered wheelchair (PW) driving is a complex activity and requires the acquisition of several skills. Given the risks involved with PW use, safe and effective training methods are needed. Virtual reality training allows users to practice difficult tasks in a safe environment. An additional benefit is that augmented feedback can be provided to optimize learning. The purpose of this study was to investigate whether providing augmented feedback during powered wheelchair simulator training results in superior performance, and whether skills learned in a virtual environment transfer to real PW driving.

Methods: Forty healthy young adults were randomly allocated to two groups: one received augmented feedback during simulator training while the control group received no augmented feedback. PW driving performance was assessed at baseline in both the real and virtual environment (RE and VE), after training in VE and two days later in VE and RE (retention and transfer tests).

Results: Both groups showed significantly better task completion time and number of collisions in the VE after training and these results were maintained two days later. The transfer test indicated better performance in the RE compared to baseline for both groups. Because time and collisions interact, a post-hoc 2D Kolmogonov-Smirnov test was used to investigate the differences in the speed-accuracy distributions for each group; a significant difference was found for the group receiving augmented feedback, before and after training, whereas the difference was not significant for the control group. There were no differences at the retention test, suggesting that augmented feedback was most effective during and immediately after training.

Conclusions: PW simulator training is effective in improving task completion time and number of collisions. A small effect of augmented feedback was seen when looking at differences in the speed-accuracy distributions, highlighting the importance of accounting for the speed-accuracy tradeoff for PW driving.

Keywords: Augmented feedback; Powered wheelchair; Training; Virtual reality.

Fig. 1

Summary of experimental design

Fig. 2

Outline of PW task (VE). Not drawn to scale

Fig. 3

Real PW task outline. Not drawn to scale

Fig. 4

Terminal feedback regarding time and collisions given to feedback group. Additional feedback regarding the pattern of the movement was also provided. The feedback consisted of the participants’ trajectory overlaid on a 2D diagram of the task viewed from above (Fig. 5a). The locations where participants made collisions (red) or came close to making collisions (blue) were indicated on the diagram. The participants were then shown snapshots of their 3 worst moments (see Fig. 5b for example). The images showed what part of the PW made contact or came close to the obstacle

Fig. 5

Terminal feedback regarding the PW’s trajectory given to feedback group. Participants can review where in their trial they made collisions (panel a, red circles) or came close to making a collision (panel a, blue circles); they can then review what part of the wheelchair collided with an obstacle (panel b)

Fig. 6

Time to complete task in the VE for feedback and no feedback group at baseline, acquisition and retention. Overall, participants had better performance from baseline to acquisition and the effect was maintained at retention. No significant group differences were found. Error bars represent confidence intervals

Fig. 7

Mean task completion time results across trials according to task version for the a) Feedback group and b) No feedback group. Error bars representing confidence intervals are shown for the test task only. Confidence intervals for the other tasks were similar and are not shown for clarity. Task completion time generally decreased over the course of training

Fig. 8

Number of collisions in the VE for feedback and no feedback group at baseline, acquisition and retention. Overall, participants had better performance from baseline to acquisition and the effect was maintained at retention. No significant group differences were found. Error bars represent confidence intervals

Fig. 9

Time to complete the RE test task for all participants in the feedback and no feedback groups, at baseline and for the transfer test two days after training. Overall, participants had better performance from baseline to the transfer test. No significant group differences were found. Error bars represent confidence intervals

Fig. 10

Speed-accuracy distributions for the VE test task. a. Comparison of the baseline and acquisition distributions for the feedback group. b. Comparison of the baseline and acquisition distributions for the no-feedback group. c. Comparison of the baseline and retention distributions for the feedback group. d. Comparison of the baseline and retention distributions for the no feedback group. For baseline to acquisition, the 2D KS test is significant for the feedback group but not for the no-feedback group, indicating that participants in the feedback group had better speed and accuracy after training

Fig. 11

Speed-accuracy distributions for the VE training task number 3. a. Comparison of training trials 3 and 16 distributions for the feedback group. b. Comparison of training trials 3 and 16 distributions for the no-feedback group. The 2D KS test is significant for the feedback group but not for the no-feedback group, indicating that participants in the feedback group had better speed and accuracy at the end of training in comparison to the beginning of training