Driving with driver assistance systems utilising haptic shared control (HSC) can lead to annoyance, or even disuse, when the intent of both entities differs. In order to increase compliance and acceptance of these systems, this study explores two types of visual information projected onto the outside scenery. One visualisation portrays only the intended trajectory of the HSC system, whereas the other visualisation complements this by showing the manoeuvring boundaries of the car. These visualisations were evaluated in a human-in-the-loop simulator study focused on supporting an early and late car overtake manoeuvre, initiated by HSC. Although most participants reported a preference for visualisation of both the intended trajectory as manoeuvring boundaries, results in terms of torque conflicts and position conflicts indicated no significant differences between the visualisations. Subsequent analysis, however, indicated that this was probably caused by the variability in how participants used the visualisations in combination with their apparent preferences in performing the overtake manoeuvre. In conclusion, supporting HSC with visual information does not improve compliance, but shows an improvement in acceptance. For future work it is recommended to further investigate the impact of visualisations on intra-driver variability.

An issue in driving simulation is that behaviour displayed in simulation does not exactly replicate behaviour in real-life. For example, roadside vegetation density impacts a driver’s speed and lateral position in on-road studies, but not in driving simulator studies. In this study it is investigated if the increase in fidelity and presence that head mounted display based virtual reality bring forth, yields behavioural adaptation as shown in real-life by evaluating the effect of three different roadside vegetation density conditions in a novel head mounted display based virtual reality driving simulator. Twenty-nine participants drove a 2m wide car over a 11km long 3.5m wide winding road. Participants completed three trials driving through three different roadside vegetation density conditions per trial, the density conditions being light (one tree per 40m), medium (one tree per 20m) and dense (one tree per 5m). The effect of the density conditions was evaluated with respect to speed, lateral position, steering reversals, subjective workload and self-reported risk. Increased vegetation density increased self-reported risk, but did not affect speed or lateral positioning. This finding is congruent with findings in con- ventional simulator studies, which could indicate that despite the advantages of head mounted display based virtual reality regarding fidelity and presence the resulting driving behaviour still has a closer connection to conventional simulated driving than real-life driving. In future work the built and implemented simulator can be improved and optimized, furthermore the relative validity of the simulator should be investigated.

Lane change manoeuvres are known to vary widely in lane change duration. This is thought to be an effect of the surrounding vehicles and personal preference of drivers. However, little is known about the effect on steering behaviour during a lane change manoeuvre. Moreover, the relation of the effect of traffic to inter- and intra-driver variability is unknown. This study focuses on quantifying inter- and intra-driver variability in lane change duration and steering behaviour during lane changes in two different traffic scenarios. In an exploratory study, 21 participants drove 30 lane change manoeuvres in a 6 DoF moving base driving simulator. Two scenarios were used: a closing gap in the target lane and a constant gap in the target lane, with 15 repetitions per scenario. The results show high inter-driver and intra-driver variability, for both lane change duration (M=6.34 s SD-inter=0.90 s SD-intra=1.26 s) and steering behaviour (e.g. maximum steering wheel angle M=4.14 deg SD-inter=1.62 deg SD-intra=1.34 deg). The effect of the scenario was not significant for lane change duration and maximum steering wheel angles. Additionally, it was shown that lane change duration only has a medium correlation with the maximum steering wheel angle (Pearson R(585)=-.48, p\textless0.001). Furthermore, the mean and variability of the lane change duration decreased when lane changes were initiated with a shorter distance to the slow lead vehicle. Concluding, the lane change duration does not fully determine steering behaviour during a lane change, making it an unsuitable metric for determining human-like lane change trajectories. It is therefore proposed to create trajectories based on steering behaviour. It seems that drivers exhibit high variability in lane change behaviour when spatio-temporal criticality with respect to traffic is low. Higher spatio-temporal criticality limits the mean and variability of the lane change duration. Future work should determine whether this variability is the result of driver preference or indifference. Additionally, future work should implement and test human-like lane change trajectories based on steering behaviour as opposed to lane change duration.

The arrival of highly automated vehicles introduces a new interaction between the vehicle and driver. System limitations during highly automated driving require the driver to be ready to take back control at request.
Previous studies on the take-over process concluded that the driver requires a transition period to stabilize vehicle control after resuming manual control. These studies used traded control to instantaneously transfer control back to the driver, causing an abrupt switch in control authority. Therefore, this study explores Haptic Shared Control as a different transition approach. By varying the level of haptic authority, a smooth connection between automation system and driver can be realized. The aim of this study is to investigate if Haptic Shared Control improves the take-over performance compared to the traded control approach. A total of 30 participants drove two trials in a driving simulator, one for each transition approach. Each trial consisted of 10 take-over scenarios divided into two levels of time-criticality. During autonomous driving the participants were engaged in a secondary task. The take-over performance was assessed based on safety performance, lateral vehicle control, controller performance and subjective measures.
Results showed a significant decrease in the standard deviation of the lateral position evaluated over the mean trajectory per participant for the Haptic Shared Control approach compared to traded control. Haptic Shared Control also showed a significant decrease for the mean lateral obstacle clearance. The analyses on torque conflicts revealed a significant increase for critical take-over maneuver compared to non-critical take-over maneuvers.
This suggests that haptic shared control can assist the driver in stabilizing lateral vehicle control after resuming manual control. On the other hand, the driver is limited in performing a sharp evasive maneuver, and this relationship is discussed. More research is needed on using an adaptable human compatible reference.

The technological advancements in the automotive industry have enabled the automation of numerous routine driving tasks. As a result, the art of driving has become a control task with a strong supervisory character, including the common human factor issues. Haptic Shared Control has been shown to be useful in keeping the driver in-the-loop by providing continuous haptic guidance on the steering wheel. Nonetheless, it has been reported that haptic support often induces control conflicts caused by the limited amount of information that can be conveyed through haptic forces. As a consequence, it is often burdensome to develop a correct mental model of the underlying controller and to establish accurate situation awareness. This study presents the results of the conceptual development of a novel visual feedback system inspired by the principle of Ecological Interface Design. By displaying the future trajectory with respect to both the physical limitations of the vehicle and the intentional constraints imposed by the road, the driver is able to establish a better understanding of the space of possibilities in a certain driving scenario. The new visual feedback is combined with current haptic feedback solutions, with the goal of guiding the driver through force-feedback, while visualizing the space of possibilities as defined by the work domain constraints. A human-in-the-loop experiment was conducted to evaluate the effects of the novel feedback system on driver behavior and acceptance during an obstacle avoidance task. In analogy with previous findings, the results showed that haptic guidance was beneficial during obstacle avoidance in terms of response time, control activity and effort, where the addition of EID-inspired visual feedback revealed an improved task execution, together with a significant reduction of control activity and conflicts.

Driving with Haptic Shared Control (HSC) provides an alternative for traditional traded control in human-controller interaction. Whilst driving, control is shared between the driver and the controller by translating the controller's desired steering input to additional torques on the steering wheel.
Literature provides several guidelines for the tuning of these torques but not for the tuning of the interaction stiffness around the controller's desired steering input: the Level of Haptic Authority (LoHA), this is usually kept constant.
High LoHA tunings have beneficial effects on the driver performance but result in high conflict torques and negatively impact the driver acceptance.
In this study two adaptive LoHA algorithms are proposed based on Time to Lane Crossing (TLC). By increasing the LoHA in critical, low-TLC, scenarios these should improve performance while also allowing for a larger steering freedom and low conflicts in safe scenarios.
The adaptive LoHA is applied symmetrically (bi-directional) and asymmetrically (only in the direction of the low TLC).
The adaptive algorithms are compared to manual driving, a low and a high static LoHA tuning in a within-subject driving simulator study. Fourteen participants performed a lane keeping task in which lane width varied to influence the safety margins and TLC.
While driving with adaptive LoHA, the mean conflict torque was significantly lower for the adaptive algorithms than for the high stiffness controller and participants experienced lower workloads. However, no difference was found between the symmetric and asymmetric LoHA controller. These results show that adaptive LoHA based on TLC is an effective way to achieve a similar performance as with static LoHA but with lower conflicts and a lower workload.

Haptic shared control (HSC) is a method to combine the abilities of humans and machines, in which human and automation jointly exert forces on an input device. According to human-centered design, the underlying controller for HSC should closely resemble human behavior. This paper aims to continuously adapt HSC based on an online identified operator model. This approach is named coadaptive HSC. Co-adaptive HSC is hypothesized to decrease conflicts when the human adjusts his control behavior to changing requirements of the task.
In a compensatory tracking experiment with eleven participants, co-adaptive HSC is compared to time-invariant HSC. During the experiments, the participants control a time-varying controlled element, which changes from a single integrator to a double integrator and back unannounced. Time-invariant HSC was designed for controlling a single integrator, whereas co-adaptive HSC adapted to the human by continuously estimating the time-varying parameters of the operator using an Extended Kalman Filter (EKF). This online identified model of the operator was directly used in the shared controller, therefore, the co-adaptive HSC imitates human behavior.
A 36% decrease in conflict rate (the percentage of time in which the human force and the controller force have opposite directions) was found for co-adaptive HSC compared to time-invariant HSC when the controlled element was a
double integrator. However, for some participants incorrect EKF-estimation of the neuromuscular damping resulted in undesired oscillations in control force during co-adaptive HSC. This indicates that the performance of co-adaptive HSC can be further improved.
Overall, this study proves that co-adaptive HSC is a promising method to reduce conflicts, and is able to adapt to operator behavior in unforeseen situations.

Taking over control from an automated vehicle may take a substantial amount of time if the driver is not engaged in the driving task. Take-over requests containing directional information of hazardous surrounding cars could aid the driver in taking over the vehicle faster. However, whether the directional information should be presented ipsilaterally or contralaterally is still inconclusive. In this study, 34 participants were presented with animated video clips of traffic situations on a three-lane road, ending with a near-collision in front after 1,3, or 6 seconds. In each video, one lane was free to maneuver to safely. Participants were instructed to make a safe lane-change by pressing the left or right arrow key. At the start of each video, participants were provided with verbal auditory feedback: (1) ‘Go left/right’ (ipsilateral), (2) ‘Danger left/right’ (contralateral), and (3) Non-directional beeps as a baseline. 80% of the trials provided valid auditory feedback (i.e., relevant to the video situation). 20% of the trials provided invalid auditory feedback (i.e., feedback opposite to the video situation, so left instead of right and vice versa). Auditory feedback ‘Go/Danger left’ was always presented from the left speaker, and ‘Go/Danger right’ was always presented from the right speaker, whereas the non-directional beeps were presented from both speakers. Participants’ keyboard responses and first gazes were recorded in each trial. It was hypothesized that when there was 1-second to respond, ipsilateral feedback (‘Go’) would lead to fastest responses because little time is available to detect the hazard. For 3 and 6-second-to-respond situations, it was hypothesized that contralateral feedback (‘Danger’) would lead to a faster detection time of the hazard, because it facilitates visual detection. The results showed that for 1 and 3-second videos, ipsilateral feedback led to significantly faster responses compared to the baseline, and for 3 and 6-second videos contralateral feedback ‘Danger’ led to significantly faster responses compared to the baseline. ‘Go’ and ‘Danger’ did not yield a significant difference in response time between each other in all videos. First fixations seem to be placed on the most salient visual stimuli, independent of the audio feedback. In 1-second time-to-respond videos this was the center of the road (the location where the potential collision is happening. In 3-seconds, this was the hazard coming from the left or right lane. And for 6 seconds the first fixations were more distributed. In conclusion, verbal auditory feedback ‘Go’ and ‘Danger’ can aid in taking over a vehicle, by reducing the take-over response time compared to baseline warnings. However, this may not be the result of facilitation of the visual detection of the hazard, because visual behavior seems to be influenced mainly by visual stimuli, independent of auditory stimuli.

Designing lane-keeping assistance (LKA) systems that are both effective and well-liked by drivers is a highly challenging process, that is not well understood. This is illustrated by a wide variety of market releases of LKA systems, and a large body of literature illustrating different designs and various evaluation methodologies that are often performed in driving simulators. As a result, it is unclear how design choices impact driver steering behaviour and acceptance, and extrapolate this to real-world driving where driver distraction is often a reality. This study presents a detailed evaluation methodology to compare three haptic LKA designs in a single real-world truck driving study. These designs are constructed based on two haptic LKA approaches found in literature; continuous support and bandwidth support. It is hypothesized that continuous support is favored over bandwidth support but that both LKA approaches are effective and well-liked when implemented in real-life and show to be particularly effective for distracted drivers. Two of the evaluated systems were triggered to generate haptic torques only when the predicted lateral error exceeded a bandwidth of 0.4 m: a single-bandwidth system (SB); that shuts off the guidance when the predicted lateral error returned within the bandwidth, and a double bandwidth system (DB); that shuts off the guidance when a second inner bandwidth (close to lane center) is reached [38]. The third evaluated system generated continuous haptic torques towards the lane center: a continuous double bandwidth system (CDB). Sixteen participants drove four trials on a private test-circuit; one trial without and three trials with haptic support. For each support system, participants drove both a distracted and a non-distracted condition. The results show that compared to manual control, all three support systems provided equal benefits in terms of accuracy and prevention of large lateral errors (>0.7m). When a lane departure did occur both DB and CDB support systems showed shorter lane departure times with smaller lateral errors compared to manual driving. The DB and CDB support systems showed high driver acceptance and reduced large swerving behaviour that was observed during distracted driving. All three support systems however lacked effectiveness and driver acceptance during curves. Ultimately the CDB support system was the overall preferred haptic LKA design. This study shows the potential for DB and CDB support to be used in real-life, however higher driver satisfaction can be achieved when the support systems are able to cope with humans’ driving behaviour during curves.

An important design choice for Haptic Shared Control is the magnitude of assisting forces: high assisting forces are beneficial during agreement between operator and controller, but also result in larger conflict forces in case of disagreement. In order to use higher forces without increasing conflict forces, literature proposes Adaptable Haptic Shared Control: using real-time operator grip force measurements to smoothly scale the magnitude of assisting forces. For full hand grip force no objective measurements comparing Haptic Shared Control versus Adaptable Haptic Shared Control are known and only very limited data on this comparison is available for finger pinching and subjective measurements. In order to prove that, using grip force to adapt the magnitude of assisting forces indeed leads to reduced conflict forces while maintaining the regular Haptic Shared Control performance during agreement, an experiment is required. It is hypothesized that Adaptable Haptic Shared Control will reduce conflict forces during disagreement and that the larger the disagreement the higher the grip force. Both an Adaptable and regular Haptic Shared Controller are designed and implemented on an actuated joystick, which is extended with a 2D dynamometer to allow real-time grip measurements. Eighteen subjects participated in an experiment where they used the Triar joystick to steer a virtual object along a path consisting of a multisine (agreement) interspersed with straight sections containing obstacles that needed to be avoided (conflict). After the adaptable and regular haptic shared control condition a Vanderlaan questionnaire was provided to the participant with the question to score them compared to manual control. During the path following task, both the Adaptable Haptic Shared controller and regular haptic shared controller provide similar increased performance in lateral deviation compared to manual control (p<0.01). During obstacle avoidance, the Adaptable Haptic shared control significantly decreased conflict forces compared to haptic shared control (p<0.01), but at a price of brief increased grip force. No significant difference in grip force was found between obstacles sizes (conflict). The subjective usefulness and satisfying score of adaptable and regular haptic shared control are both positive compared to manual but do not significantly differ from each other. Result show that with Adaptable Haptic Shared Control humans prefer increasing their grip force to lower the Haptic Shared Control forces during disagreement. Additionally during non-conflict situations they maintain the beneficial forces of Haptic Shared Control.