- Email: [email protected]
Haptic shared steering control with an adaptive level of authority based on time-to-line crossing⁎
14th IFAC Symposium on 14th IFAC Symposium on 14th IFACDesign Symposium on Analysis and Evaluation of Human Machine Systems 14th IFACDesign Symposium on Available online at www.sciencedirect.com Analysis and Evaluation of Machine Systems Analysis Design and Evaluation of Human Human Machine Systems 14th IFAC Symposium on Tallinn, Estonia, Sept. 16-19, 2019 Analysis Design and Evaluation of Human Machine Systems Tallinn, Estonia, Sept. 16-19, 2019 Tallinn, Estonia, 16-19, 2019 Analysis Design Sept. and Evaluation of Human Machine Systems Tallinn, Estonia, Sept. 16-19, 2019 Tallinn, Estonia, Sept. 16-19, 2019
ScienceDirect
IFAC PapersOnLine 52-19 (2019) 49–54
Haptic shared steering control with an Haptic shared steering control with an Haptic shared steering control with an Haptic shared steering control with adaptive level of authority based on adaptive level of authority based onan adaptive level based on adaptive level of of authority authority based on time-to-line crossing time-to-line crossing time-to-line crossing time-to-line crossing ∗ ∗ Hugo Hugo Hugo Hugo Hugo
M. M. M. M. M.
Zwaan Petermeijer ∗ ∗ Sebastiaan Zwaan Sebastiaan M. M. Petermeijer ∗∗∗∗ Zwaan M. ∗ ∗ Sebastiaan Zwaan M.∗∗∗ Petermeijer Petermeijer David A. Abbink ∗ Sebastiaan David A. Abbink Zwaan Sebastiaan M.∗∗ Petermeijer ∗ David A. Abbink David A. Abbink David A. Abbink ∗ ∗ of Cognitive Robotics, Faculty of Mechanical ∗ ∗ Department Department of Cognitive Faculty of Mechanical Robotics, Faculty of the Mechanical ∗ Department of Cognitive Robotics, ∗ Engineering, Delft University of Technology, Delft, Netherlands Department of Cognitive Robotics, Faculty of Mechanical ∗ Engineering, Delft University of Technology, Delft, the Netherlands Department of Cognitive Robotics, Faculty of Mechanical Engineering, Delft University of Technology, Delft, the Netherlands (email: [email protected]; [email protected]; Engineering, Delft University of Technology, Delft, the Netherlands (email: [email protected]; [email protected]; Engineering, Delft University of Technology, Delft, the Netherlands (email: [email protected]; [email protected]; (email: [email protected]; [email protected]; [email protected]) [email protected]) (email: [email protected]; [email protected]; [email protected]) [email protected]) [email protected]) Abstract: Traditional Traditional driver-automation driver-automation interaction interaction trades trades control control over over the vehicle vehicle back back and Abstract: Abstract: Traditional driver-automation interaction trades control over the the vehicle back and and forth between driver and automation. Haptic shared control offers an alternative by continuously Abstract: Traditional driver-automation interaction trades control over the vehicle back and forth between driver and automation. Haptic shared control offers an alternative by continuously Abstract: Traditional driver-automation interaction trades control over the vehicle back and forth between driver and automation. Haptic shared control offers an alternative by continuously sharing the control through torques on the steering wheel and pedals. When designing additional forth between driver and automation. Haptic shared control offers an alternative by continuously sharing the control control through torques on onHaptic the steering steering wheel andoffers pedals. When designing designing additional forth between driver and automation. shared control an alternative by continuously sharing the through torques the wheel and pedals. When additional feedbackthe torques, part of the thetorques design on choice lies in in the the stiffness around When the neutral neutral steering point: sharing control through the steering wheel and pedals. designing additional feedback torques, part of design choice lies stiffness around the steering point: sharing the control through on the(LoHA), steering wheel and pedals. When designing additional feedback torques, part the design choice lies stiffness around the steering point: feedback torques, part ofHaptic thetorques design choice lies in in the the stiffness aroundstatic the neutral neutral steering point: also called called the Level ofof Haptic Authority which is usually usually static and tuned to balance also the Level of Authority (LoHA), which is and tuned to balance feedback torques, partofat ofHaptic the design choice lies in the stiffness around the neutral steering point: also called the (better Level Authority which isinusually andintentions tuned to between balance safety benefits high LoHA) with (LoHA), conflicts torques case of static different also called the Level of Haptic Authority (LoHA), which is usually static and tuned to balance safety benefits (better at high LoHA) with conflicts torques in case of different intentions between also called the Level of Haptic Authority (LoHA), which is usually static and tuned to balance safety benefits (better at high LoHA) with conflicts torques in case of different intentions between automation and driverat(higher (higher conflict torques with increased LoHA).In thisintentions paper we webetween explore safety benefits (better high with conflicts torques in case of automation and driver conflict torques with increased LoHA).In this paper explore safety benefits (better high LoHA) LoHA) with conflicts torques in case of different different intentions between automation and driver conflict with increased LoHA).In paper explore automation and driverat(higher (higher conflict torques with increased LoHA).In this paper we we explore the idea idea of situation-adaptive situation-adaptive LoHA: intorques this case during lane-keeping bythis changing the LoHA the of LoHA: in this case during lane-keeping by changing the LoHA automation and (higher LoHA: conflictin torques withduring increased LoHA).In paper we the idea situation-adaptive this by changing the LoHA the idea of situation-adaptive LoHA: in this case case during lane-keeping bythis changing theexplore LoHA based on of time todriver lane crossing crossing (TLC). Consequently, whenlane-keeping safety margins are high (e.g., (e.g., when based on time to lane (TLC). Consequently, when safety margins are high when the idea situation-adaptive LoHA: in thisthe case during lane-keeping by when changing themargins LoHA based ononof time to road) lane crossing (TLC). Consequently, when safety margins are high (e.g., when driving a wide the LoHA is low, but LoHA would only increase safety based on time to lane crossing (TLC). Consequently, when safety margins are high (e.g., when driving a the LoHA is but the would only increase when safety margins based onon to road) lane two crossing (TLC). Consequently, safety margins are high (e.g., when driving ontime a wide wide road) thealternative LoHA is low, low, but approaches the LoHA LoHAwhen would onlythe increase safety margins decrease. We propose design to apply apply LoHA:when symmetrically and driving on a wide road) the LoHA is low, but the LoHA would only increase when safety margins decrease. We propose two alternative design approaches to the LoHA: symmetrically and driving on a wide road) the LoHA is low, but the LoHA would only increase when safety margins decrease. We propose two alternative design approaches to apply the LoHA: symmetrically and decrease. We propose two alternative design approaches to apply the LoHA: symmetrically and asymmetrically (i.e., (i.e., only only increase increase of of LoHA LoHA in in the the direction direction of of the low low TLC). TLC). We We compared compared asymmetrically decrease. Wein propose two alternative to apply LoHA: symmetrically and asymmetrically (i.e., only increase of LoHA in the of low TLC). compared asymmetrically (i.e., only increase ofdesign LoHAapproaches in study the direction direction of the low with TLC). We compared these design design an explorative driving simulator study (n=14) to the driving with twoWe static LoHA these in an explorative driving simulator (n=14) to driving two static LoHA asymmetrically (i.e., only increase of LoHA in the direction of the low TLC). We compared these design in an explorative driving simulator study (n=14) to driving with two static LoHA designs (low in and high).We found thatsimulator comparedstudy to the the high to LoHA controller, both adaptive these design an explorative driving (n=14) driving with static LoHA designs (low and found that compared to high LoHA controller, both adaptive these design anhigh).We explorative driving simulator study (n=14) to driving with two two static LoHA designs (low in and high).We found that compared tomargins, the highbut LoHA controller, both adaptive LoHA controllers designs resulted in similar safety at decreased conflict torques. designs (low and high).We found that compared to the high LoHA controller, both adaptive LoHA controllers controllers designs resulted resulted in similar similar safetytomargins, conflict torques. but at decreased designs (low and high).We found that compared the high LoHA controller, both adaptive LoHA designs in safety margins, but at decreased conflict torques. LoHA designs resulted in similar safety margins, at decreased conflict torques. Hence, controllers a TLC-based TLC-based adaptive LoHA controller seems to be be but an effective effective approach to mitigate Hence, a adaptive LoHA controller seems to an approach to LoHA controllers designs resulted in similar safety margins, at decreased conflict torques. Hence, a while TLC-based adaptive LoHA controller seems to with be but anHSC. effective approach to mitigate mitigate conflicts maintaining the safety benefits associated Hence, a TLC-based adaptive LoHA controller seems to be an effective approach to mitigate conflicts maintaining the safety benefits Hence, TLC-based adaptive seems to with be anHSC. effective approach to mitigate conflictsa while while maintaining the LoHA safety controller benefits associated associated with HSC. conflicts while maintaining the safety associated with conflicts maintaining thePublished safety benefits benefits associated with HSC. HSC. Copyrightwhile © 2019. The Authors. by Elsevier Ltd. All rights reserved. Keywords: Haptic Shared Control, Adaptive, Level of Haptic Authority, Human-Machine Human-Machine Keywords: Haptic Shared Control, Adaptive, Level of Haptic Authority, Keywords: Haptic Shared Control, Adaptive, Level of Haptic Authority, Human-Machine interaction, driving simulator. Keywords: Haptic Shared Control, Adaptive, Level of Haptic Authority, Human-Machine interaction, driving simulator. Keywords: Control, Adaptive, Level of Haptic Authority, Human-Machine interaction,Haptic drivingShared simulator. interaction, interaction, driving driving simulator. simulator. 1. INTRODUCTION driver and controller can negotiate, by adjusting their 1. driver and controller negotiate, by their 1. INTRODUCTION INTRODUCTION driver and controller can can negotiate, by adjusting adjusting their stiffness (co-contraction and LoHA, resp.), whose desired 1. INTRODUCTION driver and controller can negotiate, by adjusting their stiffness (co-contraction and LoHA, resp.), whose desired 1. INTRODUCTION driver and controller can negotiate, by adjusting their stiffness (co-contraction and LoHA, resp.), whose desired steering wheel angle is realized (Abbink et al., 2012). Haptic Shared Control (HSC ) implements the control acstiffness (co-contraction and LoHA, resp.), whose desired steering angle (Abbink et 2012). Haptic Shared Shared Control Control (HSC (HSC )) implements implements the the control control acac- stiffness (co-contraction and LoHA, resp.), whose desired steering wheel wheel angle is is realized realized (Abbink et al., al., 2012). Haptic tion of of an an autonomous driving through the steering steering wheel angle is (Abbink et al., 2012). Haptic Shared Control (HSC )) system implements the control action autonomous driving system through the steering Generally, higher LoHA controllers yield wheel angle is realized realized (Abbink etthe al., better 2012). perHaptic Shared Control (HSCand implements the control ac- steering tion of an autonomous driving system through the steering Generally, higher LoHA controllers yield the better perdriving system through the steering wheel, in which the driver the vehicle communicate Generally, higher LoHA controllers yield the better pertion of an autonomous wheel, in autonomous which the the driver driver and the vehicle vehicle communicate formance improvements in terms of lateral position, safety tion of an driving system through the steering Generally, higher LoHA controllers yield the better perwheel, in which and the communicate formance improvements in terms of lateral position, safety wheel, in which the driver and the vehicle communicate continuously through forces on the steering wheel (Abbink Generally, higher LoHA controllers yield the better performance improvements in terms of lateral position, safety continuously through forces on the steering wheel (Abbink margins and variability (Mars et al., 2014; Petermeijer wheel, in which the driver and the vehicle communicate formance improvements in terms of lateral position, safety continuously through forces on the steering wheel (Abbink margins and variability (Mars et al., 2014; Petermeijer et al., 2012). By keeping the driver actively engaged in formance improvements in terms of lateral position, safety margins and variability (Mars et al., 2014; Petermeijer continuously through forces on the steering wheel (Abbink et al., al., 2012). 2012).through By keeping keeping the driver actively engaged in et margins and variability (Marsin etforceful al., 2014; Petermeijer al., 2014), but can result corrections, as continuously on driver the steering wheel (Abbink et By the actively engaged in et al., but can in corrections, as theal., control loop, the forces interactions mitigates the “pitfalls margins and variability (Mars al., 2014; Petermeijer et al., 2014), 2014), butexhibit can result result inetforceful forceful corrections, as et 2012). By keeping the driver actively engaged in the control loop, the interactions mitigates the “pitfalls drivers typically satisficing instead of optimizing et al., 2012). By keeping the driver actively engaged in et al., 2014), but can result in forceful corrections, as the control loop, the interactions mitigates the “pitfalls drivers typically exhibit satisficing instead of optimizing of automation” automation” related to supervised supervised autonomous driving et al., 2014), but can result in forceful corrections, as drivers typically exhibit satisficing instead of optimizing the control loop, the interactions mitigates the “pitfalls of related to autonomous driving behaviour (Goodrich et al., 2000; Boer, 2016). During the control loop, the interactions mitigates the “pitfalls drivers typically exhibit satisficing instead of optimizing of automation” related to supervised autonomous driving behaviour (Goodrich et al., 2000; Boer, 2016). During of automation” to supervised autonomous driving drivers (Flemisch et al., al.,related 2008) while while benefiting from an an increased increased typically exhibit satisficing instead of optimizing behaviour (Goodrich et al., 2000; Boer, 2016). During (Flemisch et 2008) benefiting from such conflicts, conflicts, driversetneed need to ‘fight’ the2016). controller to of automation” to supervised autonomous driving (Goodrich al., 2000; Boer, During benefiting from increased (Flemisch et al.,related 2008)(Forsyth while such drivers ‘fight’ the controller to benefiting from an an2006) increased driving performance and MacLean, MacLean, 2006) and behaviour behaviour (Goodrich al., to 2000; Boer, During such conflicts, drivers to ‘fight’ the controller to (Flemisch et al., 2008) while driving performance (Forsyth and and such conflicts, driversetneed need to ‘fight’ the2016). controller to overcome the generated steering wheel torques, resulting (Flemisch et al., 2008) while benefiting from an increased driving performance (Forsyth and MacLean, 2006) and overcome the generated steering wheel torques, resulting reduced workload. such conflicts, drivers need to ‘fight’ the controller to overcome the generated steering wheel torques, resulting driving performance (Forsyth and MacLean, 2006) and reduced performance workload. in lower user acceptance and an increased chance of disuse driving (Forsyth and MacLean, 2006) and overcome the generated steering wheel torques, resulting reduced workload. in lower user acceptance and an increased chance of disuse reduced workload. overcome the generated steering wheel torques, resulting in lower user acceptance and an increased chance of disuse (De Winter and Dodou, 2011). This is probably the reason One approach approach to achieve achieve HSC HSC is is to to combine combine force force and and in reduced workload. lower user acceptance and an increased chance of disuse (De Winter and Dodou, 2011). is probably the One to lower user acceptance and anThis increased chance of reason disuse (De Winter and Dodou, 2011). This is probably the reason One approach to achieve HSC is toacombine forceturns and in (De Winter and Dodou, 2011). This is probably the reason that low LoHA controllers are preferred by drivers. stiffness feedback. More specifically, torque that One approach to achieve HSC is to combine force and that low LoHA controllers are preferred by drivers. stiffness feedback. More specifically, specifically, acombine torque that that turns (De Winter and Dodou, 2011). This is probably the reason that low LoHA controllers are preferred by drivers. One approach to achieve HSC is to force and stiffness feedback. More a torque turns stiffness feedback. More specifically, torque wheel that turns the steering steering wheel towards towards the desired desired asteering steering wheel angle that low LoHA controllers are preferred by drivers. the wheel the angle Tuning LoHA is thus trade-off between performance that lowthe LoHA controllers preferred by drivers. stiffness feedback. More through specifically, asteering torque thataround turns the steering wheel towards the desired wheel angle Tuning the LoHA is thus aaa are trade-off between performance the steering wheel towards the desired steering wheel angle and an added stiffness a virtual spring Tuning is between andsteering an added added stiffness through a virtual virtual spring around Tuning the LoHA is thus thus a trade-off trade-off between performance benefitsthe andLoHA driver acceptance. Though, anperformance alternative the wheel towards the desired steering wheel angle around and an stiffness through a spring benefits and driver acceptance. Though, an alternative that steering wheel angle (Abbink and Mulder, 2009). This Tuning the LoHA is thus a trade-off between performance benefits and driver acceptance. Though, an alternative and an added stiffness through a virtual spring around that steering wheel angle (Abbink and Mulder, 2009). This benefits and driver acceptance. Though, an alternative solution is to implement an adaptive LoHA, which assists and an added stiffness through a virtual spring around that steering wheel angle (Abbink and Mulder, 2009). This solution is to implement an adaptive LoHA, which assists artificial stiffness around the controller’s desired steering benefits and driver acceptance. Though, an alternative solution is to implement an adaptive LoHA, which assists that steering wheel angle (Abbink and Mulder, 2009). This artificial stiffness around the controller’s controller’s desired steering the driver when needed. As such, adaptive LoHA based that steering wheel angle (Abbink and Mulder, 2009). This solution is to implement an adaptive LoHA, which assists artificial stiffness around the desired steering the driver when needed. As such, adaptive LoHA based artificial stiffness around the controller’s desired steering angle and is called the Level of Haptic Authority (LoHA). solution is to implement an adaptive LoHA, which assists the driver when needed. As such, adaptive LoHA based angle and is called the Level of Haptic Authority (LoHA). on lateral error, with respect to a centerline reference artificial stiffness around the controller’s desired steering the driver when needed. As such, adaptive LoHA based angle and is called the Level of Haptic Authority (LoHA). on lateral error, with respect to aa centerline reference angle and is called the Level of Haptic Authority (LoHA). The stiffer the LoHA is, the harder it becomes to deviate the driver when needed. As such, adaptive LoHA based on lateral error, with respect to centerline reference The stiffer the LoHA is, the harder it becomes to deviate on lateral error, with respect to a centerline reference trajectory, has been implemented by Abbink and Mulder angle and is called the Level of Haptic Authority (LoHA). The stiffer the LoHA is, the harder it becomes to deviate trajectory, has been implemented by Abbink and Mulder from the controller’s actions and vice versa. Consequently, on lateral error, with respect to a centerline reference trajectory, has been implemented by Abbink and Mulder The stiffer the LoHA is, the harder it becomes to deviate from stiffer the controller’s controller’s actions and viceitversa. versa. Consequently, (2009). They found aa negative effect on control effort, The the LoHAactions is, the and harder becomes to deviate trajectory, has been implemented by and from the vice Consequently, (2009). They effect on control effort, trajectory, has found been implemented by Abbink Abbink and Mulder Mulder (2009). They found a negative negative effect on between control effort, from controller’s actions and vice versa. Consequently, Thethe work presented in this article was made possible by the Dutch as the controller inflated conflict torques driver from the controller’s actions and vice versa. Consequently, (2009). They found a negative effect on control effort, The work presented in this article was made possible by the Dutch as the controller inflated conflict torques between driver The work presented in this article was made possible by the Dutch (2009). They found a negative effect on control effort, as the controller inflated conflict torques between driver Technology Foundation STW (VIDI project 14127), which is part of and reference and did not allow for satisficing The work Foundation presented inSTW this article was made possible by the Dutch as the controller inflated conflict torques between driver Technology (VIDI project 14127), which is part of driver and reference and did not allow for satisficing Technology Foundation STW (VIDI project 14127), which is part of The work presented in this article was made possible by the Dutch as the controller inflated conflict torques between driver and reference and did not allow for satisficing the Dutch Organization for Scientific Research (NWO). Technology Foundation STW (VIDI project 14127), which is part of and reference and did not allow for satisficing driver the Dutch Organization for Scientific Research (NWO). the Dutch Organization for Scientific Research (NWO). Technology Foundation STW (VIDI project 14127), which is part of and reference and did not allow for satisficing driver the Dutch Organization for Scientific Research (NWO).
the Dutch Copyright Organization for The Scientific Research (NWO). 2405-8963 © 2019. Authors. Published by Elsevier Ltd. All rights reserved. Copyright © 2019 IFAC 49 Copyright © 2019 IFAC 49 Peer review under responsibility of International Federation of Automatic Copyright © 2019 IFAC 49 Control. Copyright © 2019 IFAC 49 10.1016/j.ifacol.2019.12.085 Copyright © 2019 IFAC 49
2019 IFAC HMS 50 Tallinn, Estonia, Sept. 16-19, 2019
Hugo M. Zwaan et al. / IFAC PapersOnLine 52-19 (2019) 49–54
is improper, a second order Butterworth filter is added (Eq. 3). The filter’s cut-off frequency (ωc ) is set at 3 Hz, the natural cut-off frequency of the human response (Van der Helm et al., 2002). 1 Hbutw (s) = s 2 (3) ( ωc ) − (λ1 + λ2 ) ωsc + λ1 λ2
behaviour. Instead, an adaptive controller should reduce the size of trivial conflicts, when the driver is safe, by reducing the LoHA; whilst providing strict support, in critical situations by increasing the LoHA. Satisficing behaviour in driving is based on the assumption that drivers want do not want to exceed a certain safety margin and perception of risk (Gibson and Crooks, 1938); an intuitive way to interpret safety margins is through the Time to Lane Crossing (TLC; Godthelp, 1988), which captures a combination of risk related factors in a single value, namely velocity, road width, curvature, lateral position, and heading.
−1 Hsw (s) =
Isw · s2 + Bsw · s + Ksw s 2 ( ωc ) − (λ1 + λ2 ) ωsc + λ1 λ2 iπ 2n (2k−1+n)
λk(1,2) = e
In this study we present a novel haptic shared controller, that adapts the LoHA based on TLC. First, the mathematical implementation of LoHA in a Haptic Shared Controller will be presented. Second, the LoHA’s effect on the perceived steering wheel torques in case of a disagreement between driver and controller will be analysed. Third, design of the adaptive LoHA algorithms, based on the TLC, will be explained. Finally, the preliminary results of a human-in-the-loop simulator study will be presented to evaluate if the controller operated as expected. 1.1 Four Design Choice Architecture The Four Design Choice Architecture (FDCA; Van Paassen et al., 2017), was used to implement a LoHA based on TLC. Compared to previously used haptic shared controllers (Mulder et al., 2008), the FDCA separates the feedforward and feedback torques, which allows them to be independently tuned. The FDCA controller consists of the following four components (see Figure 1): • Human Compatible Reference (HCR): The predetermined reference trajectory for the road. • Level of Haptic Support (LoHS ): The feedforward percentage of torques required to follow the HCR. • Strength of Haptic Feedback (SoHF ): The feedback gains that correct deviations from the HCR. • Level of Haptic Authority (LoHA): A virtual spring around the controller’s desired steering wheel angle.
(4) (5)
Level of Haptic Support The LoHS is essentially the percentage of the feedforward steering action that is being performed by the controller (Van Paassen et al., 2017). At 0% the controller has no feedforward component and acts only on the feedback controller, at 100% the controller performs all feedforward steering actions which means that without disturbances it will follow the HCR and basically act like a fully autonomous car. The feedforward torque (τF F ) required to generate the steering wheel angles is given in Eq. 6, with steering wheel angle (θHCR ), percentage of LoHS (λLoHS ), and inverse steering dynamics −1 (Hsw ). τF F = λLoHS · τhcr (6) −1 = λLoHS · Hsw · θHCR (7) Strength of Haptic Feedback The Strength of Haptic Feedback (SoHF) is the gain with which the controller corrects deviations from the vehcile with respect to the HCR. The lateral and heading devation (∆y and ∆ψ , resp.) with their respective gains (Ky and Kψ ), divided by the inherent steering wheel stiffness (Ksw ), determine the feedback steering wheel angle (θF B ). In order to obtain the required feedback steering wheel torque, the steering angle is multiplied by the inverse steering wheel stiffness as in Eq. 9. K y · ∆ y + Kψ · ∆ψ (8) θF B = Ksw −1 · θF B (9) τF B = Hsw Level of Haptic Authority The LoHA determines how forceful the controller will exert the feedback and feedforward torque. Mechanically it acts like a virtual spring on the steering wheel around the desired steering wheel angle of the FDCA controller. The LoHA (KLoHA ) is decomposed into the inherent steering wheel stiffness (Ksw ) and an added stiffness (Kadded ). Torque calculation is given in Eq. 11, as a function of the inverse steering wheel −1 dynamics (Hsw ),the added stiffness (Kadded ), the secondorder butterworth filter (Hf ilt ), the current steering wheel angle (θsw ), and the combined desired feedback and HCR steering angle (i.e., θco = θF B + θHCR ). KLoHA = Ksw + Kadded (10) K added −1 + τLoHA = (θco − θsw ) · Hsw (11) Hf ilt Consequently, the torque the driver needs to exert on the steering wheel to follow the HCR is independent of the chosen LoHA.
Human Compatible Reference The Human Compatible Reference (HCR) is the trajectory that the automatic controller tries to follow. The HCR was generated a priori, as in Scholtens et al. (2018), and consists of a trajectory (i.e., lateral position and heading on the road) and the steering wheel angles required to follow that trajectory. Steering Wheel Dynamics The steering wheel is modelled as a simple mass-spring damper system, with steering wheel angle (θ), steering wheel torque (τ ), inertia (Isw ), damping (Bsw ), and stiffness (Ksw ). 1 τ − Bsw θ˙ − Ksw θ (1) θ¨ = Isw A Laplace transformation is applied to obtain the transfer function (Eq. 2) of the steering wheel dynamics. 1 (2) Hsw (s) = 2 Isw · s + Bsw · s + Ksw
The total haptic share control torque (τHSC ) is obtained by adding up the feedforward torque (τF F : Eq. 7), feedback torque (τF B ; Eq. 9), and (τLoHA ; Eq. 11) as in Eq. 13.
Through inversion a transfer function of the inverse steering wheel dynamics is obtained. Because this equation 50
2019 IFAC HMS Tallinn, Estonia, Sept. 16-19, 2019
Hugo M. Zwaan et al. / IFAC PapersOnLine 52-19 (2019) 49–54
51
Car position & heading + Reference (HCR)
Feedback (SoHF)
θFB
HCR position & heading
+
θSW, HCR
Controller Driver Environment
θSteering Wheel
TLC
Feed Forward (LoHS)
+
Level of Haptic Authority (LoHA)
+
τHSC
Inverse Steering Dynamics
+
+
Vehicle/ Road
Steering Wheel
Inverse Steering Dynamics Cognitive Control
θSW, DRIVER
+-
Neuromuscular Stiffness (NMS)
+ τ Driver θSteering Wheel Car position & heading
Fig. 1. A block scheme of the FDCA-controller (top, green) and the driver (bottom, blue) in a haptic shared steering control system. Note the four design components, namely the Human Compatible Reference (HCR), Level of Haptic Support (LoHS), Strength of Haptic Feedback (SoHF), and Level of Haptic Authority (LoHA). τHSC = τF F + τF B + τLoHA −1 −1 · θHCR + Hsw · θF B ... τHSC = λLoHS · Hsw Kadded −1 ... + (θHCR + θF B − θsw ) · Hsw + Hf ilt
that this simulation the desired steering angles of driver and controller are a-priori defined, hence it essentially simulates feedforward torques and does not take feedback into account.
(12)
(13)
To evaluate the effect of LoHA and LoHS on the steering wheel angle, controller torques, and driver torques a simulation of their behaviour is performed.In the simulation both the controller and driver make the exact same steering movement but the human command is delayed by five seconds, this results in consecutive sections of conflict and agreement between the controller and the driver. Note
The driver’s dynamics are simulated by a separate massspring-damper system. In order to evaluate the contribution of each actor (i.e., driver and controller) fairly, the driver’s inertia and dampening are chosen the same as those of the steering wheel. The controller has no estimates of the human inertia, dampening or stiffness while the human model does anticipate the steering wheel’s inertia and dampening because of system identification naturally performed by humans (Kawato, 1999). The resulting steering wheel angle and torque responses can be observed in Fig. 2 and Fig. 3
Fig. 2. Steering angle response during a conflict for different ratios of LoHS, LoHA and driver stiffness. The black dashed line represents the driver’s steering angle, the dash-dot line represents the controller’s angle.
Fig. 3. Driver and controller torque responses during a conflict for different ratios of LoHS, LoHA and driver stiffness. The solid lines represent the controller torque. The dashed lines represent the driver torque.
1.2 Simulation of LoHA controller
51
2019 IFAC HMS 52 Tallinn, Estonia, Sept. 16-19, 2019
Hugo M. Zwaan et al. / IFAC PapersOnLine 52-19 (2019) 49–54
TLC-straight
TLC-right TLC-left
Fig. 4. Illustration of three different TLC calculations: Yellow - extrapolation of vehicle heading ; Blue extrapolation of yaw rate ; White - yaw uncertainty swath. The latter is used in this research.
Fig. 5. LoHA profile as a function of the TLC. Note the difference between the asymmetric controller steering left or right.
The following observations are made during conflicting steering angles: LoHA is an effective way to shift the resulting steering wheel angle towards the controller’s desired angle (LoHA = 5; Fig. 2, right plots). However, if the driver compensates his own stiffness (Human stiffness: Khu ) this effect on the steering wheel angle is mitigated (Khu = 5; Fig. 2, bottom plots) while the conflicting torques increase (dashed lines; bottom Fig. 3).
Upper and lower limits TLC A two second threshold for the adaptive algorithm was based on research by Godthelp (1988) who reported that in straight road driving corrective steering movements are made, on average, at a TLC of approximately 1.3 seconds. Moreover, a two second threshold was also found as the time by which more than 95% of drivers have responded to static roadside stimuli (Triggs and Harris, 1982).
When the LoHS is one and the LoHA is zero, the controller’s desired steering angle is followed (i.e. blue and dashed line are identical). Vice versa, when the LoHS is zero the implemented steering angle is identical to that of the desired steering angle of the driver.
A partial cosine is used to define the relationship between TLC and LoHA to prevent sudden torque pulses on the steering wheel. When (tT LC ≤ 0), the LoHA reaches its maximum value and does not increase further (i.e., upper limit).
When driver and controller agree, the LoHA has no effect on either the steering angle or the torques; in this case the LoHS fully determines how the torques are divided between the controller and driver. Alternatively, when controller and driver are in conflict the ratio between controller LoHA and driver stiffness (Khu ) determine who’s desired steering angle will be followed most closely.
K = ... LoHA K + , tT LC ≤ 0 sw Kadd π , 2 ≥ tT LC > 0 Ksw + Kadd 1 + cos(tT LC + 2) 4 K , tT LC > 2 sw (14)
These results confirm that the LoHA can be used to shift the steering wheel response between controller and driver during conflicts without affecting torques when they agree. Moreover, the driver is still capable to overrule the controller by adapting their own admittance.
Symmetric and asymmetric LoHA A symmetric implementation of the LoHA (i.e, stiffness around the desired steering wheel angle), shifts authority towards the controller regardless of driver intent and consequently limits the driver to steer into the desired direction. Hence, an asymmetric LoHA was implemented that applies a high stiffness in the direction of the low TLC, but does not apply additional stiffness towards low TLC. Essentially, the asymmetric LoHA restricts the driver to steer towards danger and it allows steering movements to safety.
2. ADAPTIVE LEVEL OF HAPTIC AUTHORITY 2.1 TLC calculation The TLC was calculated as in Boer (2016), which takes into account an uncertainty range of yaw rates and determines a cone-like swath rather than a single TLC value. Consequently, the TLC swath provides directional information about the safety margins around the current trajectory.
The same adaptive LoHA profile is used as with adaptive symmetric LoHA but rather than being based on the lowest TLC, the independent value is based on the lowest TLC in the direction of the human torque. 3. SIMULATOR EXPERIMENT
2.2 Adaptive LoHA based on TLC 3.1 Method
Upper and lower limits LoHA The lower LoHA limit is chosen to 0.0085 (Nm/deg), which is equal to the standard steering wheel stiffness of the simulator’s vehicle model (i.e., self-aligning torque during the manual driving). The upper limit at 0.0255 (Nm/deg) was heuristically determined as the highest multiple of the lower limit LoHA before drivers would report dissatisfaction.
Participants Fourteen participants voluntarily participated in the driving simulator experiment (5 female, 9 male). The average age was 25.8 years old (SD = 1.8), and participants had their driving license for an average of 6.8 years (SD = 1.7). 52
2019 IFAC HMS Tallinn, Estonia, Sept. 16-19, 2019
Hugo M. Zwaan et al. / IFAC PapersOnLine 52-19 (2019) 49–54
53
3.2 Results Fig. 7 shows the time traces of one participant for the symmetric and assymetric controller, to visualize the controller design. In the wide section of the road (2000-2500 m) the TLC-values is mostly above the 2 second activation threshold, whereas in the narrow section (2500-3000 m) the TLC rarely reaches above 2 seconds. Consequently, in the wide section the LoHA of the controller is mostly equal to the lower limit , whereas in the narrow section it varies between the upper and lower limit. Note the short torque pulses in the narrow section when drivers exceeded the lane boundaries (i.e., TLC = 0). Moreover, in the narrow section the controller torques sometimes change abruptly, due to the erratic nature of the TLC-metric. Subsequent analysis should identify if such abrupt torque changes are disruptive to the driver’s steering behaviour and subjective experience of the haptic shared controller.
Fig. 6. Visualisation of the steering wheel stiffness of the different LoHA regimes in different road positions. The springs indicate a high or low stiffness for each algorithm in that steering direction, the coloured cones indicate the corresponding TLC swaths.
Fig. 8 shows the mean minimum 10% of the TLC (i.e., indicating of how hazardous the most safety critical situation were) and the mean conflict torques (i.e., opposing torque vectors) between driver and controller. A repeated measures ANOVA and a pairwise comparison with Bonferroni correction was performed to identify statistical differences. It can be seen that there are no significant differences in terms of TLC-distributions between the controller in the wide nor the narrow sections. Moreover, the High LoHA controller has significantly higher conflict torques compared to the adaptive controllers in the narrow sections.
Table 1. FDCA parameter values used in the experiment for the SoHF, LoHS, and LoHA Ky−SoHF Tψ−SoHF λLoHS KLoHA
0.3 [Nm/m] 2.0 [Nm/rad] 80 [%] 0.0085 - 0.0255 [Nm/deg]
Apparatus A fixed-based driving simulator at the department of Control and Simulation of the Delft University of Technology was used, which has a horizontal field of view of approximately 180 deg and a vertical view of 40 deg. The steering wheel actuation was done with a MOOG FCS Ecol8000S actuator at 2500 Hz; steering wheel damping: 2 (Nms/rad) and inertia: 0.3 (Nms2 /rad). All data were recorded at a frequency of 100 Hz. Experimental design & procedure Five different conditions were driven, namely: Manual (M), Low LoHA (L), High LoHA (H), Symmetrical adaptive LoHA (S), and Asymmetrical adaptive LoHA (A). The road consisted of 16 curves (8 left, 8 right) with radii of 500 m, separated by 150 m straight section to prevent interference between subsequent curve exits and entries. Two road widths (3.6 and 2.2m) were used as an independent factor to manipulate the TLC. Road widths were alternated twice to provide both width transitions: from wide to narrow and narrow to wide. The road had a total length of 7.3 kilometers. The vehicle’s velocity was fixed at 24 (m/s). If the vehicle was outside of the lane boundaries a high frequency disturbance, modelled by a 100 Hz sinewave, was applied on the steering wheel to alert the driver of the lane departure. The controller parameters are listed in Table 1. Participants were instructed to drive as they normally would and stay within the lane boundaries. Each participant drove two training routes, one manual and one with a high stiffness controller, so they experienced both ends of the assistance spectrum prior to the experimental trials. Next, the experimental trials were driven in a randomised order to mitigate any learning effects
Fig. 7. From top to bottom: Lateral position w.r.t. the centerline, absolute Time to Line Crossing (dashed line: 2 s threshold), Level of Haptic Authority as determined by the controller (dashed lines: upper and lower limit), and the Controller Torque on the steering wheel, of one participant in a wide (2000-2500 m) and narrow (3000-3500 m) road section. 53
2019 IFAC HMS 54 Tallinn, Estonia, Sept. 16-19, 2019
Hugo M. Zwaan et al. / IFAC PapersOnLine 52-19 (2019) 49–54
Boer, E.R. (2016). Satisficing Curve Negotiation: Explaining Drivers’ Situated Lateral Position Variability. IFAC-PapersOnLine, 49(19), 183–188. doi: 10.1016/j.ifacol.2016.10.483. De Winter, J.C. and Dodou, D. (2011). Preparing drivers for dangerous situations: A critical reflection on continuous shared control. In Conference Proceedings - IEEE International Conference on Systems, Man and Cybernetics, 1050–1056. doi:10.1109/ICSMC.2011.6083813. Flemisch, F.O., Kelsch, J., L¨oper, C., Schieben, A., and Schindler, J. (2008). Automation spectrum , inner/outer compatibility and other potentially useful human factors concepts for assistance and automation. Human Factors for Assistance and Automation, (2008), 1–16. Forsyth, B.A.C. and MacLean, K.E. (2006). Predictive Haptic Guidance: Intelligent User Assistance for the Control of Dynamic Tasks. IEEE Transactions on Visualization and Computer Graphics, 12(1), 103–113. doi:10.1109/TVCG.2006.11. Gibson, J.J. and Crooks, L.E. (1938). A Theoretical FieldAnalysis of Automobile-Driving. The American Journal of Psychology, 51(3), 453–471. Godthelp, H. (1988). The limits of path error-neglecting in straight lane driving. Ergonomics, 31(4), 609–619. doi:10.1080/00140138808966703. Goodrich, M.A., Stirling, W.C., and Bolr, L.R. (2000). Satisficing revisited. Minds and Machines, 10(1), 79– 110. doi:10.1023/A:1008325423033. Kawato, M. (1999). Internal Models for Motor Control and Trajectory Planning. doi:10.1016/S0959-4388(99)000288. Mars, F., Deroo, M., and Hoc, J.M. (2014). Analysis of human-machine cooperation when driving with different degrees of haptic shared control. IEEE Transactions on Haptics, 7(3), 324–333. doi:10.1109/TOH.2013.2295095. Mulder, M., Abbink, D.A., and Boer, E.R. (2008). The effect of haptic guidance on curve negotiation behavior of young, experienced drivers. In IEEE International Conference on Systems, Man and Cybernetics, 804–809. doi:10.1109/ICSMC.2008.4811377. Petermeijer, S.M., Abbink, D.A., and de Winter, J.C. (2014). Should Drivers Be Operating Within an Automation-Free Bandwidth? Evaluating Haptic Steering Support Systems With Different Levels of Authority. Human Factors: The Journal of the Human Factors and Ergonomics Society, 57, 5–20. doi: 10.1177/0018720814563602. Scholtens, W.M., Barendswaard, S., and Abbink, D.A. (2018). Reducing conflicts in Haptic Shared Control during curve negotiation (MSc thesis). Technical report. Triggs, T.J. and Harris, W.G. (1982). Reaction Time of Drivers to Road Stimuli. Medicinski Pregled, 62(June 1982), 114–9. Van der Helm, F.C.T., Schouten, A.C., De Vlugt, E., and Brouwn, G.G. (2002). Identification of intrinsic and reflexive components of human arm dynamics during postural control. Journal of Neuroscience Methods, 119(1), 1–14. doi:10.1016/S0165-0270(02)00147-4. Van Paassen, M.M., Boink, R., Abbink, D.A., Mulder, M., and Mulder, M. (2017). Four design choices in Haptic shared control. In Advances in Aviation Psychology, Volume 2: Using Scientific Methods to Address Practical Human Factors Needs, chapter 12, 237 – 254.
Fig. 8. Boxplot of the mean min 10% TLC (left) and the mean conflict torques (right) across participants per controller (Low, High, Symmetric, Asymmetric) and road width (wide and narrow). p<.05, p<.01 4. DISCUSSION & CONCLUSION The goal of this study was to develop an adaptive LoHA algorithm based on TLC that can achieve an increase in safety margins similar to a static high stiffness controller without the drawbacks of high conflicts. An symmetric and asymmetric adaptive Level of Haptic Authority (LoHA) were implemented in the Four Design Choice Architecture, based on a time-to-lane-crossing (TLC) swath. The preliminary results human-in-the-loop simulator experiment revealed that a TLC-based adaptive controller is a promising approach to shift authority between the driver and controller based on the criticality of the driving situation. The symmetric and asymmetric adaptive controllers yielded decreased conflict torques between driver and controller. Subsequent analysis, should investigate the effect of the controller in terms of steering behaviour, effectiveness, and torque conflicts in more detail. ACKNOWLEDGEMENTS The authors would like to thank Sarvesh Kolekar for his assistance in setting up the driving simulator experiment. REFERENCES Abbink, D.A. and Mulder, M. (2009). Exploring the Dimensions of Haptic Feedback Support in Manual Control. Journal of Computing and Information Science in Engineering, 9, 011006. doi:10.1115/1.3072902. Abbink, D.A., Mulder, M., and Boer, E.R. (2012). Haptic shared control: Smoothly shifting control authority? Cognition, Technology and Work, 14(1), 19–28. doi: 10.1007/s10111-011-0192-5. 54
ScienceDirect
IFAC PapersOnLine 52-19 (2019) 49–54
Haptic shared steering control with an Haptic shared steering control with an Haptic shared steering control with an Haptic shared steering control with adaptive level of authority based on adaptive level of authority based onan adaptive level based on adaptive level of of authority authority based on time-to-line crossing time-to-line crossing time-to-line crossing time-to-line crossing ∗ ∗ Hugo Hugo Hugo Hugo Hugo
M. M. M. M. M.
Zwaan Petermeijer ∗ ∗ Sebastiaan Zwaan Sebastiaan M. M. Petermeijer ∗∗∗∗ Zwaan M. ∗ ∗ Sebastiaan Zwaan M.∗∗∗ Petermeijer Petermeijer David A. Abbink ∗ Sebastiaan David A. Abbink Zwaan Sebastiaan M.∗∗ Petermeijer ∗ David A. Abbink David A. Abbink David A. Abbink ∗ ∗ of Cognitive Robotics, Faculty of Mechanical ∗ ∗ Department Department of Cognitive Faculty of Mechanical Robotics, Faculty of the Mechanical ∗ Department of Cognitive Robotics, ∗ Engineering, Delft University of Technology, Delft, Netherlands Department of Cognitive Robotics, Faculty of Mechanical ∗ Engineering, Delft University of Technology, Delft, the Netherlands Department of Cognitive Robotics, Faculty of Mechanical Engineering, Delft University of Technology, Delft, the Netherlands (email: [email protected]; [email protected]; Engineering, Delft University of Technology, Delft, the Netherlands (email: [email protected]; [email protected]; Engineering, Delft University of Technology, Delft, the Netherlands (email: [email protected]; [email protected]; (email: [email protected]; [email protected]; [email protected]) [email protected]) (email: [email protected]; [email protected]; [email protected]) [email protected]) [email protected]) Abstract: Traditional Traditional driver-automation driver-automation interaction interaction trades trades control control over over the vehicle vehicle back back and Abstract: Abstract: Traditional driver-automation interaction trades control over the the vehicle back and and forth between driver and automation. Haptic shared control offers an alternative by continuously Abstract: Traditional driver-automation interaction trades control over the vehicle back and forth between driver and automation. Haptic shared control offers an alternative by continuously Abstract: Traditional driver-automation interaction trades control over the vehicle back and forth between driver and automation. Haptic shared control offers an alternative by continuously sharing the control through torques on the steering wheel and pedals. When designing additional forth between driver and automation. Haptic shared control offers an alternative by continuously sharing the control control through torques on onHaptic the steering steering wheel andoffers pedals. When designing designing additional forth between driver and automation. shared control an alternative by continuously sharing the through torques the wheel and pedals. When additional feedbackthe torques, part of the thetorques design on choice lies in in the the stiffness around When the neutral neutral steering point: sharing control through the steering wheel and pedals. designing additional feedback torques, part of design choice lies stiffness around the steering point: sharing the control through on the(LoHA), steering wheel and pedals. When designing additional feedback torques, part the design choice lies stiffness around the steering point: feedback torques, part ofHaptic thetorques design choice lies in in the the stiffness aroundstatic the neutral neutral steering point: also called called the Level ofof Haptic Authority which is usually usually static and tuned to balance also the Level of Authority (LoHA), which is and tuned to balance feedback torques, partofat ofHaptic the design choice lies in the stiffness around the neutral steering point: also called the (better Level Authority which isinusually andintentions tuned to between balance safety benefits high LoHA) with (LoHA), conflicts torques case of static different also called the Level of Haptic Authority (LoHA), which is usually static and tuned to balance safety benefits (better at high LoHA) with conflicts torques in case of different intentions between also called the Level of Haptic Authority (LoHA), which is usually static and tuned to balance safety benefits (better at high LoHA) with conflicts torques in case of different intentions between automation and driverat(higher (higher conflict torques with increased LoHA).In thisintentions paper we webetween explore safety benefits (better high with conflicts torques in case of automation and driver conflict torques with increased LoHA).In this paper explore safety benefits (better high LoHA) LoHA) with conflicts torques in case of different different intentions between automation and driver conflict with increased LoHA).In paper explore automation and driverat(higher (higher conflict torques with increased LoHA).In this paper we we explore the idea idea of situation-adaptive situation-adaptive LoHA: intorques this case during lane-keeping bythis changing the LoHA the of LoHA: in this case during lane-keeping by changing the LoHA automation and (higher LoHA: conflictin torques withduring increased LoHA).In paper we the idea situation-adaptive this by changing the LoHA the idea of situation-adaptive LoHA: in this case case during lane-keeping bythis changing theexplore LoHA based on of time todriver lane crossing crossing (TLC). Consequently, whenlane-keeping safety margins are high (e.g., (e.g., when based on time to lane (TLC). Consequently, when safety margins are high when the idea situation-adaptive LoHA: in thisthe case during lane-keeping by when changing themargins LoHA based ononof time to road) lane crossing (TLC). Consequently, when safety margins are high (e.g., when driving a wide the LoHA is low, but LoHA would only increase safety based on time to lane crossing (TLC). Consequently, when safety margins are high (e.g., when driving a the LoHA is but the would only increase when safety margins based onon to road) lane two crossing (TLC). Consequently, safety margins are high (e.g., when driving ontime a wide wide road) thealternative LoHA is low, low, but approaches the LoHA LoHAwhen would onlythe increase safety margins decrease. We propose design to apply apply LoHA:when symmetrically and driving on a wide road) the LoHA is low, but the LoHA would only increase when safety margins decrease. We propose two alternative design approaches to the LoHA: symmetrically and driving on a wide road) the LoHA is low, but the LoHA would only increase when safety margins decrease. We propose two alternative design approaches to apply the LoHA: symmetrically and decrease. We propose two alternative design approaches to apply the LoHA: symmetrically and asymmetrically (i.e., (i.e., only only increase increase of of LoHA LoHA in in the the direction direction of of the low low TLC). TLC). We We compared compared asymmetrically decrease. Wein propose two alternative to apply LoHA: symmetrically and asymmetrically (i.e., only increase of LoHA in the of low TLC). compared asymmetrically (i.e., only increase ofdesign LoHAapproaches in study the direction direction of the low with TLC). We compared these design design an explorative driving simulator study (n=14) to the driving with twoWe static LoHA these in an explorative driving simulator (n=14) to driving two static LoHA asymmetrically (i.e., only increase of LoHA in the direction of the low TLC). We compared these design in an explorative driving simulator study (n=14) to driving with two static LoHA designs (low in and high).We found thatsimulator comparedstudy to the the high to LoHA controller, both adaptive these design an explorative driving (n=14) driving with static LoHA designs (low and found that compared to high LoHA controller, both adaptive these design anhigh).We explorative driving simulator study (n=14) to driving with two two static LoHA designs (low in and high).We found that compared tomargins, the highbut LoHA controller, both adaptive LoHA controllers designs resulted in similar safety at decreased conflict torques. designs (low and high).We found that compared to the high LoHA controller, both adaptive LoHA controllers controllers designs resulted resulted in similar similar safetytomargins, conflict torques. but at decreased designs (low and high).We found that compared the high LoHA controller, both adaptive LoHA designs in safety margins, but at decreased conflict torques. LoHA designs resulted in similar safety margins, at decreased conflict torques. Hence, controllers a TLC-based TLC-based adaptive LoHA controller seems to be be but an effective effective approach to mitigate Hence, a adaptive LoHA controller seems to an approach to LoHA controllers designs resulted in similar safety margins, at decreased conflict torques. Hence, a while TLC-based adaptive LoHA controller seems to with be but anHSC. effective approach to mitigate mitigate conflicts maintaining the safety benefits associated Hence, a TLC-based adaptive LoHA controller seems to be an effective approach to mitigate conflicts maintaining the safety benefits Hence, TLC-based adaptive seems to with be anHSC. effective approach to mitigate conflictsa while while maintaining the LoHA safety controller benefits associated associated with HSC. conflicts while maintaining the safety associated with conflicts maintaining thePublished safety benefits benefits associated with HSC. HSC. Copyrightwhile © 2019. The Authors. by Elsevier Ltd. All rights reserved. Keywords: Haptic Shared Control, Adaptive, Level of Haptic Authority, Human-Machine Human-Machine Keywords: Haptic Shared Control, Adaptive, Level of Haptic Authority, Keywords: Haptic Shared Control, Adaptive, Level of Haptic Authority, Human-Machine interaction, driving simulator. Keywords: Haptic Shared Control, Adaptive, Level of Haptic Authority, Human-Machine interaction, driving simulator. Keywords: Control, Adaptive, Level of Haptic Authority, Human-Machine interaction,Haptic drivingShared simulator. interaction, interaction, driving driving simulator. simulator. 1. INTRODUCTION driver and controller can negotiate, by adjusting their 1. driver and controller negotiate, by their 1. INTRODUCTION INTRODUCTION driver and controller can can negotiate, by adjusting adjusting their stiffness (co-contraction and LoHA, resp.), whose desired 1. INTRODUCTION driver and controller can negotiate, by adjusting their stiffness (co-contraction and LoHA, resp.), whose desired 1. INTRODUCTION driver and controller can negotiate, by adjusting their stiffness (co-contraction and LoHA, resp.), whose desired steering wheel angle is realized (Abbink et al., 2012). Haptic Shared Control (HSC ) implements the control acstiffness (co-contraction and LoHA, resp.), whose desired steering angle (Abbink et 2012). Haptic Shared Shared Control Control (HSC (HSC )) implements implements the the control control acac- stiffness (co-contraction and LoHA, resp.), whose desired steering wheel wheel angle is is realized realized (Abbink et al., al., 2012). Haptic tion of of an an autonomous driving through the steering steering wheel angle is (Abbink et al., 2012). Haptic Shared Control (HSC )) system implements the control action autonomous driving system through the steering Generally, higher LoHA controllers yield wheel angle is realized realized (Abbink etthe al., better 2012). perHaptic Shared Control (HSCand implements the control ac- steering tion of an autonomous driving system through the steering Generally, higher LoHA controllers yield the better perdriving system through the steering wheel, in which the driver the vehicle communicate Generally, higher LoHA controllers yield the better pertion of an autonomous wheel, in autonomous which the the driver driver and the vehicle vehicle communicate formance improvements in terms of lateral position, safety tion of an driving system through the steering Generally, higher LoHA controllers yield the better perwheel, in which and the communicate formance improvements in terms of lateral position, safety wheel, in which the driver and the vehicle communicate continuously through forces on the steering wheel (Abbink Generally, higher LoHA controllers yield the better performance improvements in terms of lateral position, safety continuously through forces on the steering wheel (Abbink margins and variability (Mars et al., 2014; Petermeijer wheel, in which the driver and the vehicle communicate formance improvements in terms of lateral position, safety continuously through forces on the steering wheel (Abbink margins and variability (Mars et al., 2014; Petermeijer et al., 2012). By keeping the driver actively engaged in formance improvements in terms of lateral position, safety margins and variability (Mars et al., 2014; Petermeijer continuously through forces on the steering wheel (Abbink et al., al., 2012). 2012).through By keeping keeping the driver actively engaged in et margins and variability (Marsin etforceful al., 2014; Petermeijer al., 2014), but can result corrections, as continuously on driver the steering wheel (Abbink et By the actively engaged in et al., but can in corrections, as theal., control loop, the forces interactions mitigates the “pitfalls margins and variability (Mars al., 2014; Petermeijer et al., 2014), 2014), butexhibit can result result inetforceful forceful corrections, as et 2012). By keeping the driver actively engaged in the control loop, the interactions mitigates the “pitfalls drivers typically satisficing instead of optimizing et al., 2012). By keeping the driver actively engaged in et al., 2014), but can result in forceful corrections, as the control loop, the interactions mitigates the “pitfalls drivers typically exhibit satisficing instead of optimizing of automation” automation” related to supervised supervised autonomous driving et al., 2014), but can result in forceful corrections, as drivers typically exhibit satisficing instead of optimizing the control loop, the interactions mitigates the “pitfalls of related to autonomous driving behaviour (Goodrich et al., 2000; Boer, 2016). During the control loop, the interactions mitigates the “pitfalls drivers typically exhibit satisficing instead of optimizing of automation” related to supervised autonomous driving behaviour (Goodrich et al., 2000; Boer, 2016). During of automation” to supervised autonomous driving drivers (Flemisch et al., al.,related 2008) while while benefiting from an an increased increased typically exhibit satisficing instead of optimizing behaviour (Goodrich et al., 2000; Boer, 2016). During (Flemisch et 2008) benefiting from such conflicts, conflicts, driversetneed need to ‘fight’ the2016). controller to of automation” to supervised autonomous driving (Goodrich al., 2000; Boer, During benefiting from increased (Flemisch et al.,related 2008)(Forsyth while such drivers ‘fight’ the controller to benefiting from an an2006) increased driving performance and MacLean, MacLean, 2006) and behaviour behaviour (Goodrich al., to 2000; Boer, During such conflicts, drivers to ‘fight’ the controller to (Flemisch et al., 2008) while driving performance (Forsyth and and such conflicts, driversetneed need to ‘fight’ the2016). controller to overcome the generated steering wheel torques, resulting (Flemisch et al., 2008) while benefiting from an increased driving performance (Forsyth and MacLean, 2006) and overcome the generated steering wheel torques, resulting reduced workload. such conflicts, drivers need to ‘fight’ the controller to overcome the generated steering wheel torques, resulting driving performance (Forsyth and MacLean, 2006) and reduced performance workload. in lower user acceptance and an increased chance of disuse driving (Forsyth and MacLean, 2006) and overcome the generated steering wheel torques, resulting reduced workload. in lower user acceptance and an increased chance of disuse reduced workload. overcome the generated steering wheel torques, resulting in lower user acceptance and an increased chance of disuse (De Winter and Dodou, 2011). This is probably the reason One approach approach to achieve achieve HSC HSC is is to to combine combine force force and and in reduced workload. lower user acceptance and an increased chance of disuse (De Winter and Dodou, 2011). is probably the One to lower user acceptance and anThis increased chance of reason disuse (De Winter and Dodou, 2011). This is probably the reason One approach to achieve HSC is toacombine forceturns and in (De Winter and Dodou, 2011). This is probably the reason that low LoHA controllers are preferred by drivers. stiffness feedback. More specifically, torque that One approach to achieve HSC is to combine force and that low LoHA controllers are preferred by drivers. stiffness feedback. More specifically, specifically, acombine torque that that turns (De Winter and Dodou, 2011). This is probably the reason that low LoHA controllers are preferred by drivers. One approach to achieve HSC is to force and stiffness feedback. More a torque turns stiffness feedback. More specifically, torque wheel that turns the steering steering wheel towards towards the desired desired asteering steering wheel angle that low LoHA controllers are preferred by drivers. the wheel the angle Tuning LoHA is thus trade-off between performance that lowthe LoHA controllers preferred by drivers. stiffness feedback. More through specifically, asteering torque thataround turns the steering wheel towards the desired wheel angle Tuning the LoHA is thus aaa are trade-off between performance the steering wheel towards the desired steering wheel angle and an added stiffness a virtual spring Tuning is between andsteering an added added stiffness through a virtual virtual spring around Tuning the LoHA is thus thus a trade-off trade-off between performance benefitsthe andLoHA driver acceptance. Though, anperformance alternative the wheel towards the desired steering wheel angle around and an stiffness through a spring benefits and driver acceptance. Though, an alternative that steering wheel angle (Abbink and Mulder, 2009). This Tuning the LoHA is thus a trade-off between performance benefits and driver acceptance. Though, an alternative and an added stiffness through a virtual spring around that steering wheel angle (Abbink and Mulder, 2009). This benefits and driver acceptance. Though, an alternative solution is to implement an adaptive LoHA, which assists and an added stiffness through a virtual spring around that steering wheel angle (Abbink and Mulder, 2009). This solution is to implement an adaptive LoHA, which assists artificial stiffness around the controller’s desired steering benefits and driver acceptance. Though, an alternative solution is to implement an adaptive LoHA, which assists that steering wheel angle (Abbink and Mulder, 2009). This artificial stiffness around the controller’s controller’s desired steering the driver when needed. As such, adaptive LoHA based that steering wheel angle (Abbink and Mulder, 2009). This solution is to implement an adaptive LoHA, which assists artificial stiffness around the desired steering the driver when needed. As such, adaptive LoHA based artificial stiffness around the controller’s desired steering angle and is called the Level of Haptic Authority (LoHA). solution is to implement an adaptive LoHA, which assists the driver when needed. As such, adaptive LoHA based angle and is called the Level of Haptic Authority (LoHA). on lateral error, with respect to a centerline reference artificial stiffness around the controller’s desired steering the driver when needed. As such, adaptive LoHA based angle and is called the Level of Haptic Authority (LoHA). on lateral error, with respect to aa centerline reference angle and is called the Level of Haptic Authority (LoHA). The stiffer the LoHA is, the harder it becomes to deviate the driver when needed. As such, adaptive LoHA based on lateral error, with respect to centerline reference The stiffer the LoHA is, the harder it becomes to deviate on lateral error, with respect to a centerline reference trajectory, has been implemented by Abbink and Mulder angle and is called the Level of Haptic Authority (LoHA). The stiffer the LoHA is, the harder it becomes to deviate trajectory, has been implemented by Abbink and Mulder from the controller’s actions and vice versa. Consequently, on lateral error, with respect to a centerline reference trajectory, has been implemented by Abbink and Mulder The stiffer the LoHA is, the harder it becomes to deviate from stiffer the controller’s controller’s actions and viceitversa. versa. Consequently, (2009). They found aa negative effect on control effort, The the LoHAactions is, the and harder becomes to deviate trajectory, has been implemented by and from the vice Consequently, (2009). They effect on control effort, trajectory, has found been implemented by Abbink Abbink and Mulder Mulder (2009). They found a negative negative effect on between control effort, from controller’s actions and vice versa. Consequently, Thethe work presented in this article was made possible by the Dutch as the controller inflated conflict torques driver from the controller’s actions and vice versa. Consequently, (2009). They found a negative effect on control effort, The work presented in this article was made possible by the Dutch as the controller inflated conflict torques between driver The work presented in this article was made possible by the Dutch (2009). They found a negative effect on control effort, as the controller inflated conflict torques between driver Technology Foundation STW (VIDI project 14127), which is part of and reference and did not allow for satisficing The work Foundation presented inSTW this article was made possible by the Dutch as the controller inflated conflict torques between driver Technology (VIDI project 14127), which is part of driver and reference and did not allow for satisficing Technology Foundation STW (VIDI project 14127), which is part of The work presented in this article was made possible by the Dutch as the controller inflated conflict torques between driver and reference and did not allow for satisficing the Dutch Organization for Scientific Research (NWO). Technology Foundation STW (VIDI project 14127), which is part of and reference and did not allow for satisficing driver the Dutch Organization for Scientific Research (NWO). the Dutch Organization for Scientific Research (NWO). Technology Foundation STW (VIDI project 14127), which is part of and reference and did not allow for satisficing driver the Dutch Organization for Scientific Research (NWO).
the Dutch Copyright Organization for The Scientific Research (NWO). 2405-8963 © 2019. Authors. Published by Elsevier Ltd. All rights reserved. Copyright © 2019 IFAC 49 Copyright © 2019 IFAC 49 Peer review under responsibility of International Federation of Automatic Copyright © 2019 IFAC 49 Control. Copyright © 2019 IFAC 49 10.1016/j.ifacol.2019.12.085 Copyright © 2019 IFAC 49
2019 IFAC HMS 50 Tallinn, Estonia, Sept. 16-19, 2019
Hugo M. Zwaan et al. / IFAC PapersOnLine 52-19 (2019) 49–54
is improper, a second order Butterworth filter is added (Eq. 3). The filter’s cut-off frequency (ωc ) is set at 3 Hz, the natural cut-off frequency of the human response (Van der Helm et al., 2002). 1 Hbutw (s) = s 2 (3) ( ωc ) − (λ1 + λ2 ) ωsc + λ1 λ2
behaviour. Instead, an adaptive controller should reduce the size of trivial conflicts, when the driver is safe, by reducing the LoHA; whilst providing strict support, in critical situations by increasing the LoHA. Satisficing behaviour in driving is based on the assumption that drivers want do not want to exceed a certain safety margin and perception of risk (Gibson and Crooks, 1938); an intuitive way to interpret safety margins is through the Time to Lane Crossing (TLC; Godthelp, 1988), which captures a combination of risk related factors in a single value, namely velocity, road width, curvature, lateral position, and heading.
−1 Hsw (s) =
Isw · s2 + Bsw · s + Ksw s 2 ( ωc ) − (λ1 + λ2 ) ωsc + λ1 λ2 iπ 2n (2k−1+n)
λk(1,2) = e
In this study we present a novel haptic shared controller, that adapts the LoHA based on TLC. First, the mathematical implementation of LoHA in a Haptic Shared Controller will be presented. Second, the LoHA’s effect on the perceived steering wheel torques in case of a disagreement between driver and controller will be analysed. Third, design of the adaptive LoHA algorithms, based on the TLC, will be explained. Finally, the preliminary results of a human-in-the-loop simulator study will be presented to evaluate if the controller operated as expected. 1.1 Four Design Choice Architecture The Four Design Choice Architecture (FDCA; Van Paassen et al., 2017), was used to implement a LoHA based on TLC. Compared to previously used haptic shared controllers (Mulder et al., 2008), the FDCA separates the feedforward and feedback torques, which allows them to be independently tuned. The FDCA controller consists of the following four components (see Figure 1): • Human Compatible Reference (HCR): The predetermined reference trajectory for the road. • Level of Haptic Support (LoHS ): The feedforward percentage of torques required to follow the HCR. • Strength of Haptic Feedback (SoHF ): The feedback gains that correct deviations from the HCR. • Level of Haptic Authority (LoHA): A virtual spring around the controller’s desired steering wheel angle.
(4) (5)
Level of Haptic Support The LoHS is essentially the percentage of the feedforward steering action that is being performed by the controller (Van Paassen et al., 2017). At 0% the controller has no feedforward component and acts only on the feedback controller, at 100% the controller performs all feedforward steering actions which means that without disturbances it will follow the HCR and basically act like a fully autonomous car. The feedforward torque (τF F ) required to generate the steering wheel angles is given in Eq. 6, with steering wheel angle (θHCR ), percentage of LoHS (λLoHS ), and inverse steering dynamics −1 (Hsw ). τF F = λLoHS · τhcr (6) −1 = λLoHS · Hsw · θHCR (7) Strength of Haptic Feedback The Strength of Haptic Feedback (SoHF) is the gain with which the controller corrects deviations from the vehcile with respect to the HCR. The lateral and heading devation (∆y and ∆ψ , resp.) with their respective gains (Ky and Kψ ), divided by the inherent steering wheel stiffness (Ksw ), determine the feedback steering wheel angle (θF B ). In order to obtain the required feedback steering wheel torque, the steering angle is multiplied by the inverse steering wheel stiffness as in Eq. 9. K y · ∆ y + Kψ · ∆ψ (8) θF B = Ksw −1 · θF B (9) τF B = Hsw Level of Haptic Authority The LoHA determines how forceful the controller will exert the feedback and feedforward torque. Mechanically it acts like a virtual spring on the steering wheel around the desired steering wheel angle of the FDCA controller. The LoHA (KLoHA ) is decomposed into the inherent steering wheel stiffness (Ksw ) and an added stiffness (Kadded ). Torque calculation is given in Eq. 11, as a function of the inverse steering wheel −1 dynamics (Hsw ),the added stiffness (Kadded ), the secondorder butterworth filter (Hf ilt ), the current steering wheel angle (θsw ), and the combined desired feedback and HCR steering angle (i.e., θco = θF B + θHCR ). KLoHA = Ksw + Kadded (10) K added −1 + τLoHA = (θco − θsw ) · Hsw (11) Hf ilt Consequently, the torque the driver needs to exert on the steering wheel to follow the HCR is independent of the chosen LoHA.
Human Compatible Reference The Human Compatible Reference (HCR) is the trajectory that the automatic controller tries to follow. The HCR was generated a priori, as in Scholtens et al. (2018), and consists of a trajectory (i.e., lateral position and heading on the road) and the steering wheel angles required to follow that trajectory. Steering Wheel Dynamics The steering wheel is modelled as a simple mass-spring damper system, with steering wheel angle (θ), steering wheel torque (τ ), inertia (Isw ), damping (Bsw ), and stiffness (Ksw ). 1 τ − Bsw θ˙ − Ksw θ (1) θ¨ = Isw A Laplace transformation is applied to obtain the transfer function (Eq. 2) of the steering wheel dynamics. 1 (2) Hsw (s) = 2 Isw · s + Bsw · s + Ksw
The total haptic share control torque (τHSC ) is obtained by adding up the feedforward torque (τF F : Eq. 7), feedback torque (τF B ; Eq. 9), and (τLoHA ; Eq. 11) as in Eq. 13.
Through inversion a transfer function of the inverse steering wheel dynamics is obtained. Because this equation 50
2019 IFAC HMS Tallinn, Estonia, Sept. 16-19, 2019
Hugo M. Zwaan et al. / IFAC PapersOnLine 52-19 (2019) 49–54
51
Car position & heading + Reference (HCR)
Feedback (SoHF)
θFB
HCR position & heading
+
θSW, HCR
Controller Driver Environment
θSteering Wheel
TLC
Feed Forward (LoHS)
+
Level of Haptic Authority (LoHA)
+
τHSC
Inverse Steering Dynamics
+
+
Vehicle/ Road
Steering Wheel
Inverse Steering Dynamics Cognitive Control
θSW, DRIVER
+-
Neuromuscular Stiffness (NMS)
+ τ Driver θSteering Wheel Car position & heading
Fig. 1. A block scheme of the FDCA-controller (top, green) and the driver (bottom, blue) in a haptic shared steering control system. Note the four design components, namely the Human Compatible Reference (HCR), Level of Haptic Support (LoHS), Strength of Haptic Feedback (SoHF), and Level of Haptic Authority (LoHA). τHSC = τF F + τF B + τLoHA −1 −1 · θHCR + Hsw · θF B ... τHSC = λLoHS · Hsw Kadded −1 ... + (θHCR + θF B − θsw ) · Hsw + Hf ilt
that this simulation the desired steering angles of driver and controller are a-priori defined, hence it essentially simulates feedforward torques and does not take feedback into account.
(12)
(13)
To evaluate the effect of LoHA and LoHS on the steering wheel angle, controller torques, and driver torques a simulation of their behaviour is performed.In the simulation both the controller and driver make the exact same steering movement but the human command is delayed by five seconds, this results in consecutive sections of conflict and agreement between the controller and the driver. Note
The driver’s dynamics are simulated by a separate massspring-damper system. In order to evaluate the contribution of each actor (i.e., driver and controller) fairly, the driver’s inertia and dampening are chosen the same as those of the steering wheel. The controller has no estimates of the human inertia, dampening or stiffness while the human model does anticipate the steering wheel’s inertia and dampening because of system identification naturally performed by humans (Kawato, 1999). The resulting steering wheel angle and torque responses can be observed in Fig. 2 and Fig. 3
Fig. 2. Steering angle response during a conflict for different ratios of LoHS, LoHA and driver stiffness. The black dashed line represents the driver’s steering angle, the dash-dot line represents the controller’s angle.
Fig. 3. Driver and controller torque responses during a conflict for different ratios of LoHS, LoHA and driver stiffness. The solid lines represent the controller torque. The dashed lines represent the driver torque.
1.2 Simulation of LoHA controller
51
2019 IFAC HMS 52 Tallinn, Estonia, Sept. 16-19, 2019
Hugo M. Zwaan et al. / IFAC PapersOnLine 52-19 (2019) 49–54
TLC-straight
TLC-right TLC-left
Fig. 4. Illustration of three different TLC calculations: Yellow - extrapolation of vehicle heading ; Blue extrapolation of yaw rate ; White - yaw uncertainty swath. The latter is used in this research.
Fig. 5. LoHA profile as a function of the TLC. Note the difference between the asymmetric controller steering left or right.
The following observations are made during conflicting steering angles: LoHA is an effective way to shift the resulting steering wheel angle towards the controller’s desired angle (LoHA = 5; Fig. 2, right plots). However, if the driver compensates his own stiffness (Human stiffness: Khu ) this effect on the steering wheel angle is mitigated (Khu = 5; Fig. 2, bottom plots) while the conflicting torques increase (dashed lines; bottom Fig. 3).
Upper and lower limits TLC A two second threshold for the adaptive algorithm was based on research by Godthelp (1988) who reported that in straight road driving corrective steering movements are made, on average, at a TLC of approximately 1.3 seconds. Moreover, a two second threshold was also found as the time by which more than 95% of drivers have responded to static roadside stimuli (Triggs and Harris, 1982).
When the LoHS is one and the LoHA is zero, the controller’s desired steering angle is followed (i.e. blue and dashed line are identical). Vice versa, when the LoHS is zero the implemented steering angle is identical to that of the desired steering angle of the driver.
A partial cosine is used to define the relationship between TLC and LoHA to prevent sudden torque pulses on the steering wheel. When (tT LC ≤ 0), the LoHA reaches its maximum value and does not increase further (i.e., upper limit).
When driver and controller agree, the LoHA has no effect on either the steering angle or the torques; in this case the LoHS fully determines how the torques are divided between the controller and driver. Alternatively, when controller and driver are in conflict the ratio between controller LoHA and driver stiffness (Khu ) determine who’s desired steering angle will be followed most closely.
K = ... LoHA K + , tT LC ≤ 0 sw Kadd π , 2 ≥ tT LC > 0 Ksw + Kadd 1 + cos(tT LC + 2) 4 K , tT LC > 2 sw (14)
These results confirm that the LoHA can be used to shift the steering wheel response between controller and driver during conflicts without affecting torques when they agree. Moreover, the driver is still capable to overrule the controller by adapting their own admittance.
Symmetric and asymmetric LoHA A symmetric implementation of the LoHA (i.e, stiffness around the desired steering wheel angle), shifts authority towards the controller regardless of driver intent and consequently limits the driver to steer into the desired direction. Hence, an asymmetric LoHA was implemented that applies a high stiffness in the direction of the low TLC, but does not apply additional stiffness towards low TLC. Essentially, the asymmetric LoHA restricts the driver to steer towards danger and it allows steering movements to safety.
2. ADAPTIVE LEVEL OF HAPTIC AUTHORITY 2.1 TLC calculation The TLC was calculated as in Boer (2016), which takes into account an uncertainty range of yaw rates and determines a cone-like swath rather than a single TLC value. Consequently, the TLC swath provides directional information about the safety margins around the current trajectory.
The same adaptive LoHA profile is used as with adaptive symmetric LoHA but rather than being based on the lowest TLC, the independent value is based on the lowest TLC in the direction of the human torque. 3. SIMULATOR EXPERIMENT
2.2 Adaptive LoHA based on TLC 3.1 Method
Upper and lower limits LoHA The lower LoHA limit is chosen to 0.0085 (Nm/deg), which is equal to the standard steering wheel stiffness of the simulator’s vehicle model (i.e., self-aligning torque during the manual driving). The upper limit at 0.0255 (Nm/deg) was heuristically determined as the highest multiple of the lower limit LoHA before drivers would report dissatisfaction.
Participants Fourteen participants voluntarily participated in the driving simulator experiment (5 female, 9 male). The average age was 25.8 years old (SD = 1.8), and participants had their driving license for an average of 6.8 years (SD = 1.7). 52
2019 IFAC HMS Tallinn, Estonia, Sept. 16-19, 2019
Hugo M. Zwaan et al. / IFAC PapersOnLine 52-19 (2019) 49–54
53
3.2 Results Fig. 7 shows the time traces of one participant for the symmetric and assymetric controller, to visualize the controller design. In the wide section of the road (2000-2500 m) the TLC-values is mostly above the 2 second activation threshold, whereas in the narrow section (2500-3000 m) the TLC rarely reaches above 2 seconds. Consequently, in the wide section the LoHA of the controller is mostly equal to the lower limit , whereas in the narrow section it varies between the upper and lower limit. Note the short torque pulses in the narrow section when drivers exceeded the lane boundaries (i.e., TLC = 0). Moreover, in the narrow section the controller torques sometimes change abruptly, due to the erratic nature of the TLC-metric. Subsequent analysis should identify if such abrupt torque changes are disruptive to the driver’s steering behaviour and subjective experience of the haptic shared controller.
Fig. 6. Visualisation of the steering wheel stiffness of the different LoHA regimes in different road positions. The springs indicate a high or low stiffness for each algorithm in that steering direction, the coloured cones indicate the corresponding TLC swaths.
Fig. 8 shows the mean minimum 10% of the TLC (i.e., indicating of how hazardous the most safety critical situation were) and the mean conflict torques (i.e., opposing torque vectors) between driver and controller. A repeated measures ANOVA and a pairwise comparison with Bonferroni correction was performed to identify statistical differences. It can be seen that there are no significant differences in terms of TLC-distributions between the controller in the wide nor the narrow sections. Moreover, the High LoHA controller has significantly higher conflict torques compared to the adaptive controllers in the narrow sections.
Table 1. FDCA parameter values used in the experiment for the SoHF, LoHS, and LoHA Ky−SoHF Tψ−SoHF λLoHS KLoHA
0.3 [Nm/m] 2.0 [Nm/rad] 80 [%] 0.0085 - 0.0255 [Nm/deg]
Apparatus A fixed-based driving simulator at the department of Control and Simulation of the Delft University of Technology was used, which has a horizontal field of view of approximately 180 deg and a vertical view of 40 deg. The steering wheel actuation was done with a MOOG FCS Ecol8000S actuator at 2500 Hz; steering wheel damping: 2 (Nms/rad) and inertia: 0.3 (Nms2 /rad). All data were recorded at a frequency of 100 Hz. Experimental design & procedure Five different conditions were driven, namely: Manual (M), Low LoHA (L), High LoHA (H), Symmetrical adaptive LoHA (S), and Asymmetrical adaptive LoHA (A). The road consisted of 16 curves (8 left, 8 right) with radii of 500 m, separated by 150 m straight section to prevent interference between subsequent curve exits and entries. Two road widths (3.6 and 2.2m) were used as an independent factor to manipulate the TLC. Road widths were alternated twice to provide both width transitions: from wide to narrow and narrow to wide. The road had a total length of 7.3 kilometers. The vehicle’s velocity was fixed at 24 (m/s). If the vehicle was outside of the lane boundaries a high frequency disturbance, modelled by a 100 Hz sinewave, was applied on the steering wheel to alert the driver of the lane departure. The controller parameters are listed in Table 1. Participants were instructed to drive as they normally would and stay within the lane boundaries. Each participant drove two training routes, one manual and one with a high stiffness controller, so they experienced both ends of the assistance spectrum prior to the experimental trials. Next, the experimental trials were driven in a randomised order to mitigate any learning effects
Fig. 7. From top to bottom: Lateral position w.r.t. the centerline, absolute Time to Line Crossing (dashed line: 2 s threshold), Level of Haptic Authority as determined by the controller (dashed lines: upper and lower limit), and the Controller Torque on the steering wheel, of one participant in a wide (2000-2500 m) and narrow (3000-3500 m) road section. 53
2019 IFAC HMS 54 Tallinn, Estonia, Sept. 16-19, 2019
Hugo M. Zwaan et al. / IFAC PapersOnLine 52-19 (2019) 49–54
Boer, E.R. (2016). Satisficing Curve Negotiation: Explaining Drivers’ Situated Lateral Position Variability. IFAC-PapersOnLine, 49(19), 183–188. doi: 10.1016/j.ifacol.2016.10.483. De Winter, J.C. and Dodou, D. (2011). Preparing drivers for dangerous situations: A critical reflection on continuous shared control. In Conference Proceedings - IEEE International Conference on Systems, Man and Cybernetics, 1050–1056. doi:10.1109/ICSMC.2011.6083813. Flemisch, F.O., Kelsch, J., L¨oper, C., Schieben, A., and Schindler, J. (2008). Automation spectrum , inner/outer compatibility and other potentially useful human factors concepts for assistance and automation. Human Factors for Assistance and Automation, (2008), 1–16. Forsyth, B.A.C. and MacLean, K.E. (2006). Predictive Haptic Guidance: Intelligent User Assistance for the Control of Dynamic Tasks. IEEE Transactions on Visualization and Computer Graphics, 12(1), 103–113. doi:10.1109/TVCG.2006.11. Gibson, J.J. and Crooks, L.E. (1938). A Theoretical FieldAnalysis of Automobile-Driving. The American Journal of Psychology, 51(3), 453–471. Godthelp, H. (1988). The limits of path error-neglecting in straight lane driving. Ergonomics, 31(4), 609–619. doi:10.1080/00140138808966703. Goodrich, M.A., Stirling, W.C., and Bolr, L.R. (2000). Satisficing revisited. Minds and Machines, 10(1), 79– 110. doi:10.1023/A:1008325423033. Kawato, M. (1999). Internal Models for Motor Control and Trajectory Planning. doi:10.1016/S0959-4388(99)000288. Mars, F., Deroo, M., and Hoc, J.M. (2014). Analysis of human-machine cooperation when driving with different degrees of haptic shared control. IEEE Transactions on Haptics, 7(3), 324–333. doi:10.1109/TOH.2013.2295095. Mulder, M., Abbink, D.A., and Boer, E.R. (2008). The effect of haptic guidance on curve negotiation behavior of young, experienced drivers. In IEEE International Conference on Systems, Man and Cybernetics, 804–809. doi:10.1109/ICSMC.2008.4811377. Petermeijer, S.M., Abbink, D.A., and de Winter, J.C. (2014). Should Drivers Be Operating Within an Automation-Free Bandwidth? Evaluating Haptic Steering Support Systems With Different Levels of Authority. Human Factors: The Journal of the Human Factors and Ergonomics Society, 57, 5–20. doi: 10.1177/0018720814563602. Scholtens, W.M., Barendswaard, S., and Abbink, D.A. (2018). Reducing conflicts in Haptic Shared Control during curve negotiation (MSc thesis). Technical report. Triggs, T.J. and Harris, W.G. (1982). Reaction Time of Drivers to Road Stimuli. Medicinski Pregled, 62(June 1982), 114–9. Van der Helm, F.C.T., Schouten, A.C., De Vlugt, E., and Brouwn, G.G. (2002). Identification of intrinsic and reflexive components of human arm dynamics during postural control. Journal of Neuroscience Methods, 119(1), 1–14. doi:10.1016/S0165-0270(02)00147-4. Van Paassen, M.M., Boink, R., Abbink, D.A., Mulder, M., and Mulder, M. (2017). Four design choices in Haptic shared control. In Advances in Aviation Psychology, Volume 2: Using Scientific Methods to Address Practical Human Factors Needs, chapter 12, 237 – 254.
Fig. 8. Boxplot of the mean min 10% TLC (left) and the mean conflict torques (right) across participants per controller (Low, High, Symmetric, Asymmetric) and road width (wide and narrow). p<.05, p<.01 4. DISCUSSION & CONCLUSION The goal of this study was to develop an adaptive LoHA algorithm based on TLC that can achieve an increase in safety margins similar to a static high stiffness controller without the drawbacks of high conflicts. An symmetric and asymmetric adaptive Level of Haptic Authority (LoHA) were implemented in the Four Design Choice Architecture, based on a time-to-lane-crossing (TLC) swath. The preliminary results human-in-the-loop simulator experiment revealed that a TLC-based adaptive controller is a promising approach to shift authority between the driver and controller based on the criticality of the driving situation. The symmetric and asymmetric adaptive controllers yielded decreased conflict torques between driver and controller. Subsequent analysis, should investigate the effect of the controller in terms of steering behaviour, effectiveness, and torque conflicts in more detail. ACKNOWLEDGEMENTS The authors would like to thank Sarvesh Kolekar for his assistance in setting up the driving simulator experiment. REFERENCES Abbink, D.A. and Mulder, M. (2009). Exploring the Dimensions of Haptic Feedback Support in Manual Control. Journal of Computing and Information Science in Engineering, 9, 011006. doi:10.1115/1.3072902. Abbink, D.A., Mulder, M., and Boer, E.R. (2012). Haptic shared control: Smoothly shifting control authority? Cognition, Technology and Work, 14(1), 19–28. doi: 10.1007/s10111-011-0192-5. 54