Delft Haptics Lab

Haptically Supporting Car Driving: Optimizing or Satisficing?

Abstract

This MSc project was performed by Bastiaan Petermeijer, under the supervision of David Abbink and Joost de Winter.

Background: A principal design choice in providing steering wheel force feedback to assist drivers during lane-keeping is whether feedback should be provided for any error with respect to the lane centre (continuous feedback) or only when acceptable tolerance limits are exceeded (bandwidth feedback). Problem: Research has shown that continuously assisting the driver by means of steering wheel force feedback improves driver performance and reduces mental workload. However, it has been suggested that continuous feedback induces aftereffects during subsequent manual driving. Method: In a fixed-base driving simulator study, we evaluated driver performance and mental workload, while drivers (n=32) received corrective force feedback on the steering wheel. Five designs were compared, namely no feedback (Manual), discrete bandwidth feedback (Band1), bandwidth feedback with a hysteresis filter (Band2), continuous feedback with feedback gain used in earlier studies (Cont), and continuous feedback with a feedback gain twice as strong (ContS). While driving the participants performed a secondary peripheral detection task. Participants were told that the feedback was not fully reliable. Towards the end of the trajectory the feedback algorithm was shut down in order to investigate short-term aftereffects. After each trial the participant would fill out a NASA-TLX and a Vanderlaan-questionnaire to measure workload, usefulness and satisfaction. Results: All four feedback designs eliminated the large lane centre deviations that occasionally occur during manual control. Both continuous feedback designs yielded smaller lane centre errors than bandwidth feedback and manual control. Shortly after the feedback was shut down, ContS resulted in worse lane keeping performance than Manual and bandwidth feedback. The secondary task showed a lower reaction time for Cont compared to Band1 and the NASA-TLX indicated lower workload for continuous feedback compared to manual control. The satisfaction scale of the Vanderlaan-questionnaire showed a significantly higher score for Cont compared to Band1. When the feedback shut down, drivers supported by ContS showed a lower minimum TLC, higher mean and maximum absolute lateral error, and more near lane departures. Conclusions: Providing the driver with continuous feedback is an effective way to improve the lane keeping performance and is experienced by drivers as more satisfactory and intuitive, as compared to bandwidth force feedback and manual steering. Bandwidth feedback is equally effective in preventing excessive deviation from the lane centre and did not suffer from the aftereffects that the ContS system did. Stronger continuous feedback resulted in better lane keeping performance, but also in larger aftereffects. The Cont design had a good performance, was perceived as satisfactory and showed no significant short-term aftereffects. Nevertheless, future research should point out if long-term behavioural adaptations lead to stronger aftereffects for continuous systems.