The risk of motion sickness is considerably higher in autonomous vehicles than it is in human-operated vehicles. Their introduction will therefore require systems that mitigate motion sickness. We investigated whether this can be achieved by augmenting the vehicle interior with additional visualizations. Participants were immersed in motion simulations on a moving-base driving simulator, where they were backward-facing passengers of an autonomous vehicle. Using a Head-Mounted Display, they were presented either with a regular view from inside the vehicle, or with augmented views that offered additional cues on the vehicle's present motion or motion 500ms into the future, displayed on the vehicle's interior panels. In contrast to the hypotheses and other recent studies, no difference was found between conditions. The absence of differences between conditions suggests a ceiling effect: providing a regular view may limit motion sickness, but presentation of additional visual information beyond this does not further reduce sickness.

@en

Previous literature suggests a relationship between individual characteristics of motion perception and the peak frequency of motion sickness sensitivity. Here, we used well-established paradigms to relate motion perception and motion sickness on an individual level. We recruited 23 participants to complete a two-part experiment. In the first part, we determined individual velocity storage time constants from perceived rotation in response to Earth Vertical Axis Rotation (EVAR) and subjective vertical time constants from perceived tilt in response to centrifugation. The cross-over frequency for resolution of the gravito-inertial ambiguity was derived from our data using the Multi Sensory Observer Model (MSOM). In the second part of the experiment, we determined individual motion sickness frequency responses. Participants were exposed to 30-minute sinusoidal fore-aft motions at frequencies of 0.15, 0.2, 0.3, 0.4 and 0.5 Hz, with a peak amplitude of 2 m/s² in five separate sessions, approximately 1 week apart. Sickness responses were recorded using both the MIsery SCale (MISC) with 30 s intervals, and the Motion Sickness Assessment Questionnaire (MSAQ) at the end of the motion exposure. The average velocity storage and subjective vertical time constants were 17.2 s (STD = 6.8 s) and 9.2 s (STD = 7.17 s). The average cross-over frequency was 0.21 Hz (STD = 0.10 Hz). At the group level, there was no significant effect of frequency on motion sickness. However, considerable individual variability was observed in frequency sensitivities, with some participants being particularly sensitive to the lowest frequencies, whereas others were most sensitive to intermediate or higher frequencies. The frequency of peak sensitivity did not correlate with the velocity storage time constant (r = 0.32, p = 0.26) or the subjective vertical time constant (r = − 0.37, p = 0.29). Our prediction of a significant correlation between cross-over frequency and frequency sensitivity was not confirmed (r = 0.26, p = 0.44). However, we did observe a strong positive correlation between the subjective vertical time constant and general motion sickness sensitivity (r = 0.74, p = 0.0006). We conclude that frequency sensitivity is best considered a property unique to the individual. This has important consequences for existing models of motion sickness, which were fitted to group averaged sensitivities. The correlation between the subjective vertical time constant and motion sickness sensitivity supports the importance of verticality perception during exposure to translational sickness stimuli.

@en

Percepts of verticality are thought to be constructed as a weighted average of multisensory inputs, but the observed weights differ considerably between studies. In the present study, we evaluate whether this can be explained by differences in how visual, somatosensory and proprioceptive cues contribute to representations of the Head In Space (HIS) and Body In Space (BIS). Participants (10) were standing on a force plate on top of a motion platform while wearing a visualization device that allowed us to artificially tilt their visual surroundings. They were presented with (in)congruent combinations of visual, platform, and head tilt, and performed Rod & Frame Test (RFT) and Subjective Postural Vertical (SPV) tasks. We also recorded postural responses to evaluate the relation between perception and balance. The perception data shows that body tilt, head tilt, and visual tilt affect the HIS and BIS in both experimental tasks. For the RFT task, visual tilt induced considerable biases (≈ 10° for 36° visual tilt) in the direction of the vertical expressed in the visual scene; for the SPV task, participants also adjusted platform tilt to correct for illusory body tilt induced by the visual stimuli, but effects were much smaller (≈ 0.25°). Likewise, postural data from the SPV task indicate participants slightly shifted their weight to counteract visual tilt (0.3° for 36° visual tilt). The data reveal a striking dissociation of visual effects between the two tasks. We find that the data can be explained well using a model where percepts of the HIS and BIS are constructed from direct signals from head and body sensors, respectively, and indirect signals based on body and head signals but corrected for perceived neck tilt. These findings show that perception of the HIS and BIS derive from the same sensory signals, but see profoundly different weighting factors. We conclude that observations of different weightings between studies likely result from querying of distinct latent constructs referenced to the body or head in space.@en

In dynamic driving simulators, the experience of operating a vehicle is reproduced by combining visual stimuli generated by graphical rendering with inertial stimuli generated by platform motion. Due to inherent limitations of the platform workspace, inertial stimulation is subject to shortcomings in the form of missing cues, false cues, and/or scaling errors, which negatively affect simulation fidelity. In the present study, we aim at quantifying the relative contribution of an active somatosensory stimulation to the perceived intensity of self-motion, relative to other sensory systems. Participants judged the intensity of longitudinal and lateral driving maneuvers in a dynamic driving simulator in passive driving conditions, with and without additional active somatosensory stimulation, as provided by an Active Seat (AS) and Active Belts (AB) integrated system (ASB). The results show that ASB enhances the perceived intensity of sustained decelerations, and increases the precision of acceleration perception overall. Our findings are consistent with models of perception, and indicate that active somatosensory stimulation can indeed be used to improve simulation fidelity.

@en

In this paper the potential of Motion Incongruence Rating (MIR) models for the optimization of Motion Cueing Algorithms (MCAs) is investigated. In a human-in-the-loop simulator experiment, two optimization-based MCAs are compared for a roundabout scenario simulated on a medium-stroke hexapod simulator. The first MCA uses standard cueing error weights from reference literature in its cost function, while for the second case these weights were based on a MIR model fitted to previous experiment data. Results show that such models provide a promising cueing error weight estimation method for optimization-based MCAs, but also highlight the limitations of these models due to, for example, their dependency on the richness of the datasets to which they are fitted. @en

Despite gaining popularity, the use of Motion Cueing Algorithms (MCAs) based on Model Predictive Control (MPC) remains challenging due to the required tuning of a large number of cost function parameters. This paper investigates the effects of two critical MPC cost function parameters, the lateral specific force and roll rate error weights (Way and Wp), on the motion cueing quality achieved with an MPC-based MCA for a curve driving scenario. An offline sensitivity analysis, which quantified the effects of varying Way and Wp on the Root Mean Square Error (RMSE) and Pearson Correlation Coefficient (PCC) of the resulting simulator motion outputs, shows that for the same percentage-wise variation, Way has a more pronounced effect on both cueing quality predictors than Wp. In addition, for both RMSE and PCC, the effects of Way and Wp are also found to be largely independent, i.e., without interaction effects. This was further tested in a passive human-in-the-loop experiment with 20 participants and with nine different Way and Wp parameter combinations as test conditions, performed in the hexapod moving-base simulator of the Max Planck Institute for Biological Cybernetics in T¨ubingen. The collected continuous rating data, which were found to be reliable for 18/20 participants, show a statistically significant variation across all experiment conditions, and especially a strong interaction effect of Way and Wp. Somewhat surprisingly, the overall lowest continuous ratings were given to the combination of both reference weight settings from earlier research (our baseline condition). In line with the interaction effect in the continuous data, an extended post-experiment correlation analysis shows that a weighted combination of lateral specific force RMSE and and roll rate RMSE above the roll rate perception threshold strongly correlates (_ = 0.98) with the variation in mean continuous ratings across all experiment conditions. This approach can potentially be used for straightforward prediction of perceived motion cueing quality and offline MCA optimization. @en

This paper describes a driving simulation experiment, executed on the Daimler Driving Simulator (DDS), in which a filter-based and an optimization-based motion cueing algorithm (MCA) were compared using a newly developed motion cueing quality rating method. The goal of the comparison was to investigate whether optimization-based MCAs have, compared to filter-based approaches, the potential to improve the quality of motion simulations. The paper describes the two algorithms, discusses their strengths and weaknesses and describes the experimental methods and results. The MCAs were compared in an experiment where 18 participants rated the perceived motion mismatch, i.e., the perceived mismatch between the motion felt in the simulator and the motion one would expect from a drive in a real car. The results show that the quality of the motion cueing was rated better for the optimization-based MCA than for the filter-based MCA, indicating that there exists a potential to improve the quality of the motion simulation with optimization-based methods. Furthermore, it was shown that the rating method provides reliable and repeatable results within and between participants, which further establishes the utility of the method.

@en

In motion simulation, motion input scaling is often applied to deal with the limited motion envelopes of motion simulators. In this research, the time-varying effects of scaling the lateral specific force up or down during passive curve driving in a car driving simulation are investigated through a simulator experiment. It is concluded that lateral specific force scaling has a time-varying effect on the perceived fidelity of a curve-driving simulation. In particular, motion scaling during a curve entry is found to be less detrimental than motion scaling during a curve's sustained part and during the curve exit.@en

Optimization-based motion cueing algorithms based on model predictive control have been recently implemented to reproduce the motion of a car within the limited workspace of a driving simulator. These algorithms require a reference of the future vehicle motion to compute a prediction of the system response. Assumptions regarding the future reference signals must be made in order to develop effective prediction strategies. However, it remains unclear how the prediction of future vehicle dynamics influences the quality of the motion cueing. In this study two prediction strategies are considered. Oracle: the ideal prediction strategy that knows exactly what the future reference is going to be. Constant: a prediction strategy that ignores every future change and keeps the current vehicle’s linear accelerations and angular velocities constant. The two prediction strategies are used to reproduce a sequence of maneuvers between 0 and 50 km/h. A comparative analysis is carried out to objectively evaluate the influence of the prediction strategies on motion cueing quality. Dedicated indicators of correlation, delay and absolute error are used to compare the effects of the adopted prediction on simulator motion. Also the motion cueing mechanisms adopted by the different conditions are analyzed, together with the usage of simulator workspace. While the constant strategy provided reasonable cueing quality, the results show that knowledge of the future vehicle trajectory reduces the delay and improves correlation with the reference trajectory, it allows the combined usage of different motion cueing mechanisms and increases the usage of workspace.

@en

Motion cueing algorithms are used in motion simulation to map the inertial vehicle motion onto the limited simulator motion space. This mapping causes mismatches between the unrestricted visual motion and the constrained inertial motion, which results in perceived motion incongruence (PMI). It is still largely unknown what exactly causes visual and inertial motion in a simulator to be perceived as incongruent. Current methods for measuring motion incongruence during motion simulation result in time-invariant measures of the overall incongruence, which makes it difficult to determine the relevance of the individual and short-duration mismatches between visual and inertial motion cues. In this paper, a novel method is presented to subjectively measure the time-varying PMI continuously throughout a simulation. The method is analyzed for reliability and validity of its measurements, as well as for its applicability in relating physical short-duration cueing errors to PMI. The analysis shows that the method is reliable and that the results can be used to obtain a deeper insight into the formation of motion incongruence during driving simulation.

@en

The human controller (HC) in manual control of a dynamical system often follows a visible and predictable reference path (target). The HC can adopt a control strategy combining closed-loop feedback and an open-loop feedforward response. The effects of the target signal waveform shape and the system dynamics on the human feedforward dynamics are still largely unknown, even for common, stable, vehicle-like dynamics. This paper studies the feedforward dynamics through computer model simulations and compares these to system identification results from human-in-the-loop experimental data. Two target waveform shapes are considered, constant velocity ramp segments and constant acceleration parabola segments. Furthermore, three representative vehicle-like system dynamics are considered: 1) a single integrator (SI); 2) a second-order system; and 3) a double integrator. The analyses show that the HC utilizes a combined feedforward/feedback control strategy for all dynamics with the parabola target, and for the SI and second-order system with the ramp target. The feedforward model parameters are, however, very different between the two target waveform shapes, illustrating the adaptability of the HC to task variables. Moreover, strong evidence of anticipatory control behavior in the HC is found for the parabola target signal. The HC anticipates the future course of the parabola target signal given extensive practice, reflected by negative feedforward time delay estimates.@en

Today, simulators are achieving levels of complexity and cost that are comparable to those of the aircraft they should replace. For this reason, questions have been raised, in both the technical and training communities, on the required level of simulation 1delity for effective pilot training. Computer Based Trainers (CBTs) are not currently considered in regulatory standards, because it has not been proven yet whether they can replace or complement actual 2ight training hours. The aim of this paper is to better understand to what extent the low-level hover skills developed on a CBT are effectively transferred to a more realistic simulation environment. To achieve this goal, a quasi-Transfer-of-Training (qToT) experiment with task-naïve participants was performed in the CyberMotion Simulator (CMS) at the Max Planck Institute for Biological Cybernetics. Twenty-four subjects, divided in two groups, were trained to perform the hover maneuver controlling an identi1ed model of a Robinson R44 civil light helicopter. The 1rst group (the “experimental” group) was trained in a CBT and then transferred to the realistic setting in the CMS. The second group (the “control” group) received the entire training in the CMS. At the end of the experiment, the two groups were found to show comparable performance. This suggests that, even for the training of low-level 2ying skills, CBTs may be a valid alternative to high 1delity simulators, if supported by a suitable training program. @en

Biodynamic feedthrough (BDFT) refers to the feedthrough of vehicle accelerations through the human body, leading to involuntary control device inputs. BDFT impairs control performance in a large range of vehicles under various circumstances. Research shows that BDFT strongly depends on adaptations in the neuromuscular admittance dynamics of the human body. This paper proposes a model-based approach of BDFT mitigation that accounts for these neuromuscular adaptations. The method was tested, as proof-of-concept, in an experiment where participants inside a motion simulator controlled a simulated vehicle through a virtual tunnel. Through evaluating tracking performance and control effort with and without motion disturbance active and with and without cancellation active, the effectiveness of the cancellation was evaluated. Results show that the cancellation approach is successful: the detrimental effects of BDFT were largely removed.@en

Realistic manual control tasks typically involve predictable target signals and random disturbances. The human controller (HC) is hypothesized to use a feedforward control strategy for target-following, in addition to feedback control for disturbance-rejection. Little is known about human feedforward control, partly because common system identification methods have difficulty in identifying whether, and (if so) how, the HC applies a feedforward strategy. In this paper, an identification procedure is presented that aims at an objective model selection for identifying the human feedforward response, using linear time-invariant autoregressive with exogenous input models. A new model selection criterion is proposed to decide on the model order (number of parameters) and the presence of feedforward in addition to feedback. For a range of typical control tasks, it is shown by means of Monte Carlo computer simulations that the classical Bayesian information criterion (BIC) leads to selecting models that contain a feedforward path from data generated by a pure feedback model: “false-positive” feedforward detection. To eliminate these false-positives, the modified BIC includes an additional penalty on model complexity. The appropriate weighting is found through computer simulations with a hypothesized HC model prior to performing a tracking experiment. Experimental human-in-the-loop data will be considered in future work. With appropriate weighting, the method correctly identifies the HC dynamics in a wide range of control tasks, without false-positive results.@en

The human controller (HC) can greatly improve target-tracking performance by utilizing a feedforward operation on the target signal, in addition to a feedback response. System identification methods are used to determine the correct HC model structure: purely feedback or a combined feedforward/feedback model. In this paper, we investigate three central issues that complicate this objective. First, the identification method should not require prior assumptions regarding the dynamics of the feedforward and feedback components. Second, severe biases might be introduced by high levels of noise in the data measured under closed-loop conditions. To address the first two issues, we will consider two identification methods that make use of linear ARX models: the classic direct method and the two-stage indirect method of van den Hof and Schrama (1993). Third, model complexity should be considered in the selection of the ‘best’ ARX model to prevent ‘false-positive’ feedforward identification. Various model selection criteria, that make an explicit trade-off between model quality and model complexity, are considered. Based on computer simulations with a HC model, we conclude that 1) the direct method provides more accurate estimates in the frequency range of interest, and 2) existing model selection criteria do not prevent false-positive feedforward identification.@en

In the manual control of a dynamic system, the human controller (HC) is often required to follow a visible and predictable reference path. Using the predictable aspect of a reference signal, through applying feedforward control, the HC can significantly improve performance as compared to a purely feedback control strategy. A proper definition of a signal’s predictability, however, is never given in literature. This paper investigates the predictability of a sum-of-sinusoids target signal, as a function of the number of sinusoid components and the fact whether the sinusoid frequencies are harmonic, or not. A human-in-the-loop experiment was done, with target signals varying for these two signal characteristics. A combined feedback-feedforward HC model was identified and parameters were estimated. It was found that for all experimental conditions, subjects used a feedforward strategy. Results further showed thatsubjects were able to perform better for harmonic signals as compared to non-harmonic signals, for signals with roughly the same frequency content.@en