Evaluation of Lane Change Maneuvers on a Dynamic Motorcycle Riding Simulator Utilizing a Rider Leaning Input

Type of the Paper: Conference Paper

Evaluation of Lane Change Maneuvers on a Dynamic Motorcycle Riding Simulator Utilizing a Rider Leaning Input

Raphael Pleß^{1, *}, Sebastian Will¹, Nora L. Merkel¹, Alexandra Neukum¹

¹ Würzburger Institut für Verkehrswissenschaften GmbH, Germany; pless@wivw.de, ORCID 0009-0006-5984-6723; will@wivw.de, ORCID
0000-0003-0098-6212; merkel@wivw.de, ORCID 0000-0002-4865-368X; neukum@wivw.de

*corresponding author

Name of Editor: Jason Moore

Submitted: 15/09/2023

Accepted: 21/09/2023

Published: 03/10/2023

Citation: Pleß, R., Will, S., Merkel, N. & Neukum, A. (2023). Evaluation of Lane Change Maneuvers on a Dynamic Motorcycle Riding Simulator Utilizing a Rider Leaning Input. The Evolving Scholar - BMD 2023, 5th Edition. This work is licensed under a Creative Commons Attribution License (CC-BY).

Abstract:

Up until today, high fidelity dynamic motorcycle riding simulators (DMRS) lack behind the rideability and accessibility of real motorcycles. This is a limiting factor, when it comes to the applicability of such simulators in the development processes of motorcycle manufacturers, suppliers and research institutes. Extensive training of the study participants enables valid studies, but decreases the test efficiency and weakens the trust of managers and decision makers into the results gained on the simulator.
One approach to increase the rideability of DMRS is to introduce technology, that allows utilizing rider motion as an input to the simulation, instead of only implementing a steering input. This approach is called “Dual Loop Rider Control” (DLRC) and is realized on the DESMORI simulator by means of a roll torque measurement that takes any coupling torque between the rider and the motorcycle frame around the vehicle’s longitudinal axis into account (Pleß, 2016).

The objective of the paper at hand is to discuss, if and how the applicability and performance of DLRC in dynamic riding maneuvers can be rated. Scales and ratings known from literature, that are for example applied for the analysis of motorcycle handling, are not sufficient for this purpose. For instance, the Lane-Change-Roll Index will decrease when implementing DLRC and utilizing lean-ing (vs. riding with steering input only). Typically, such lower steer torque efforts would indicate improved handling ratings. But ultimately, they have no relevance in terms of rideability, accessibility and realism of the simulator, as these qualities cannot be boiled down to lower steering efforts. Thus, there is the need for new objective performance measures.

It is hypothesized, that an increased rideability of the simulator is observable in a higher precision and repeatability when performing a lane change. A set of characteristic values describing this maneuver is presented, that aims at objectively rating the performance of the simulator. The values result from a curve fitting of the vehicle trajectory to a hyperbolic tangent function. In order to investigate the effects of DLRC on these characteristic values, the lane change maneuver is tested at velocities between 30 km/h and 100 km/h in three different configurations: pure steering control, pure leaning control and DLRC. The collected data highlights the effectiveness of the added leaning input and indicates slight improvements in rideability of the lane change maneuver. However, the objective performance ratings still don’t suffice to draw a precise picture of the gain in rideability through DLRC.

Keywords: Motorcycle, Simulator, Lane-change

Dual Loop Rider Control

One key difference between motorcycling and driving cars is the vehicle’s response to rider/driver motion. The car driver may move around in the seat without any subjectively perceivable effect and without significant reactions in vehicle dynamics. The steering wheel can typically be considered as the only driver input to the vehicle’s lateral dynamics. In contrast, motorcycles will not only react to a rider’s steering inputs, but to lateral motion of the rider’s center of gravity (CoG) as well. The effects of an eccentric rider CoG on the steady state behavior of a motorcycle are well known (Cossalter, 2006): While maintaining a constant velocity and curvature of the motorcycle trajectory, leaning the rider’s body towards the center of a curve will for example decrease the motorcycle’s geometric roll angle and steering torque. Furthermore, the rider motion generates dynamic effects on the motorcycle that can even suffice to follow a slightly curved trajectory without applying any torques to the handlebar at all. The relevant dynamic equations have been derived e.g. by (Åström, 2005) and result in the simplified transfer function shown in Equation (1):

Where \(b_{1,2}\) describe the rider’s inertial properties and \(a_{1,2,3}\) contain inertial and geometric properties of the motorcycle, with the latter two being subject to the current vehicle speed \(v\). Figure 1 exemplarily shows the response of the vehicle roll angle \(\varphi_{mcy}\) to a sigmoid shaped rider lean angle excitation \(\varphi_{rid} = 10{^\circ}\left( 1 - \cos\left( \frac{\pi t}{0.5\ s} \right) \right);0 \leq t \leq 0.5\ s\)

Figure 1. Exemplary response of the Lean-to-Roll transfer function (Pleß, 2023) derived from the model of (Åström, 2005)

It can be seen, how a rider leaning sideways (blue line) will initially cause the motorcycle to lean in the opposite direction (red lines, magnified representation in the middle image of the figure) with decreasing amplitudes, as the velocity increases. After a few oscillations, a new equilibrium results, that is as well subject to the vehicle speed.

While the effects of rider motion on the lateral dynamics are well known, they have not been considered much in the development of motorcycle riding simulators. Many static and dynamic motorcycle riding simulators ever only relied on the steering as their only input to the lateral dynamics (“Single Loop Rider Control”) and until today, the design of steering systems for motorcycle riding simulators is an open field of research, see e.g. (Werle, 2022). Of the better-known motorcycle riding simulators, the DIMEG Simulator at the University of Padua was the first to consider rider motion as an input by means of a vertical load measurement at the foot pegs (Cossalter, 2011). In the following, such systems will be referenced to as “Dual Loop Rider Control” (DLRC), as they not only utilize steering, but rider motion as a rider input to the vehicle dynamics model. The motorcycle riding simulators of BMW Motorrad (Guth, 2017) and Cruden (Westerhof, 2018) implemented optical measurement methods to track the rider’s upper body. All abovementioned systems have in common, that they rely on rather specific input cues. Pressure sensitive foot pegs will for example not react to body motion at all, if the rider will support their weight through the simulator frame and handlebar instead of deliberately pressing down on a foot peg. Optical measurements on the other hand are limited to the motion of the observed body-parts and will neglect all other motions. An approach for a holistic rider motion determination was presented with the DESMORI simulator (Pleß, 2016). It consists of a longitudinally mounted mechanical axis that allows for a rotation of the motorcycle frame relative to the motion platform. This rotation is however supported against a load cell and will not result in a perceptible movement.

Figure 2. Measuring concept of the rider induced roll torque.

The concept is sketched in Figure 2. The measured torque \(T_{LC,meas}\) that is acting around the mechanical axis (yellow) contains effects from both rider motion and platform motion. Assuming the rider being a stiff point-mass, the latter can be estimated, as the inertial properties and motion cues of the system are known (green). The difference between the measured signal \(T_{LC,meas}\) and the estimated signal \(T_{LC,est}\) will therefore contain the torque that the rider applies on the motorcycle frame both actively (leaning) and passively (being an elastic and inert mass). This so called “rider induced roll torque” \(T_{rid}\ \)(red) is then used as input to the multi body simulation model of the motorcycle, which is here built in the commercially available software VI-grade Bike RealTime.

The described system will react to any rider body excitation, may it result from leaning to one side, shaking the head or stretching a leg. The precision of the system depends on the quality of the torque estimator as well as mechanical properties like the component’s stiffnesses or bearing friction. The estimator was tested in various realistic motion scenarios, where it showed root mean square errors below \(3\ Nm\) while the rider induced torques range up to around \(\pm 150\ Nm\) (Pleß, 2023).

Scenario Description

The lane change experiment discussed in this paper is performed on a virtual test track that is based on a planar, straight, three-lane road. Cones are used to define the areas where the rider is supposed to enter, perform and exit the lane change maneuver. Figure 3 shows the track segments by means of a map view (top) and screenshots of the simulation environment SILAB^® (bottom). The maneuver is repeated for ten times consecutively, before changing the target velocity or control configuration.

Firstly, the length \(l_{pre}\) allows the rider to control and adjust the target velocity as well as smoothly entering the segment \(l_{entry}\), where the rider is forced into a 50 m long corridor with a limited lane width of 1.8 m. (leftmost picture). When passing the corridor, the rider will see the upcoming blocked middle lane. The traffic sign will show either a left or right arrow at the instance the rider exits the entry corridor. From this moment on, the rider has about 2.5 s to react and change the lane (second left picture). The exit segment is offset by 3 m and uses a 75 m long corridor of cones with a width of 1.8 m as before (second right picture). This is intended to urge the rider to come to a straight riding state at a rather well constrained lateral offset, rather than only passing and “corner cutting” alongside the obstacle on the middle lane. Lastly, in the segment \(l_{post}\), the rider can smoothly transition back towards the next entry. The “pre” and “post” lengths are chosen such that the rider may take up to 7 seconds at the indicated target velocity to reach the entry.

Figure 3. Map view of the lane change scenario (top) and sequence of events in the rider perspective (bottom).

The maneuver was performed in different control configurations. Each rider was either allowed to use the handlebar (H1) or not (H0), while activating (L1) or deactivating (L0) the leaning input. In the following discussions and figures, the lane change maneuver is indicated as “DL”. This results in the configuration descriptors \(DL_{L0}^{H0},\ DL_{L0}^{H1},\ DL_{L1}^{H0}\ \&\ DL_{L1}^{H1}.\ \)Obviously, the first variant is not possible to perform, as no steering or leaning input exist. Therefore, three configurations have been tested. Both H1 configurations were tested at 30, 50, 70, 90 and 110 km/h. For the H0/L1 configuration – i.e. riding without hands – the test was performed at 50 and 80 km/h (set by a cruise control) while using the track geometries from the 70 and 110 km/h versions of the H1 configurations. This results in a total of twelve scenario configurations with ten repetitions. The test parameters are listed in Table 1:

Table 1. Lane Change Scenario Configurations.

The table highlights the \(l_{mid}\) section in blue. It can be seen, that the lateral offset \(\Delta y_{mid}\) is kept constant in all variations and the length is adjusted such that a duration about 2.5 seconds is achieved for the variation with steering control and a duration of 3.6 seconds without steering control. Only participants that were well trained on the motorcycle simulator performed the experiment. Each variation was repeated ten times before lowering the target velocity and eventually changing the control configuration.

Measurements

The data shown in the following sections contain data from the same rider. The data collected with two other riders showed no relevant differences and is therefore not further depicted or discussed here. For the following detailed discussions on the timeseries of dynamic quantities, the middle scenario of Table 1 was selected such that all plots are based on the same track geometry.

Firstly, Figure 4 shows the front tire trajectories for the two configurations with active steering. Each blue line represents one of the ten repetitions that the rider performed. The left plot shows the configuration with a disabled roll torque input, the right plot shows the configuration with DLRC, i.e. both steering and leaning inputs being active. The \({DL}_{L1}^{H1}\) condition arguably shows smaller deviations in the entry and exit corridors as well as less amount of overshooting at the end of the lane change maneuver.

Figure 4. Trajectories of the front tire contact point during lane change without (left) and with (right) activated DLRC.

All correlation coefficients between each measurement per condition are greater than 99%. In the following it is therefore assumed, that left- and right-hand lane changes can be evaluated as equal by mirroring the left-hand lane change data onto the right-hand data. The following plots show the relevant lateral-dynamic quantities of the same configurations as above. Figure 5 represents the left-hand data of Figure 4, where the rider could only use the steering as input, while Figure 6 represents the right-hand data, where DLRC was active and both steering and leaning of the rider affected the vehicle dynamics. Lastly, Figure 7 shows the remaining configuration, where only the leaning input was utilized. The dynamic quantities are discriminated by color, blue and green indicating the roll and steer angle of the virtual motorcycle respectively, red and yellow indicating the rider induced roll torque and the steering torque measured on the simulator. The thick lines depict the mean values of the ten repetitions, while the shaded areas span over one standard deviation. The angles are referenced on the left ordinate, torques on the right.

The \({DL}_{L0}^{H1}\) condition can be seen as the baseline condition, as it represents the state of the art of controlling a motorcycle simulator only by means of a steering input. Figure 5 shows the typical behavior of a motorcycle during a lane change maneuver. At a distance of approximately 60 m a rapid increase in the steer torque in the opposite direction of the lane change (negative sign) causes a counter steering motion. In consequence, a roll angle towards the direction of the lane change is building up, reaching it’s maximum at around 75 m. At this point, the rider will apply a positively signed steer torque in order to quickly build up a roll angle that allows the rider to return to a straight direction of travelling again. The amplitudes of the roll and steer angles reach similar values both for entering and exiting the lane change. At the end of the plot, the increasing roll angles show, how the rider is not able to find a straight running equilibrium but rather initiates the return to the center lane for the experiment’s next repetition early within the exit corridor. As the speed is easily kept constant, and the target trajectory is rather constrained by the given cones, there is not really any room for choosing alternative lines in the mid-section of the scenario. Furthermore, any variation can only result from steering inputs – no other disturbances exist. Therefore, the data shows rather small standard deviations.

In Figure 6, the rider induced roll torque is added as an input (red line and shaded area). While the standard deviation of the roll angle increases slightly, the opposite is true for the steer torque and angle. When initiating the lane change, the rider induced roll torque points towards the direction of the lane change. This lean-in approach requires less steer torque and generates smaller counter steer angles. When exiting the lane change, the leaning motion results in a much smoother up-righting of the motorcycle, allowing to ride through the final corridor in a straighter line as before. The prominent peak in the roll torque signal at around 85 m can be traced back to a malfunction of the motion platform during one repetition and can be neglected here.

Figure 5: Average values and standard deviations of dynamic quantities in \(\mathbf{DL}_{\mathbf{L0}}^{\mathbf{H1}}\) condition.

Figure 6. Average values and standard deviations of dynamic quantities in \(\mathbf{DL}_{\mathbf{L1}}^{\mathbf{H1}}\) condition.

In the last condition \({DL}_{L1}^{H0}\) the rider does not grip the handlebar and the velocity is set by a cruise control at 50 km/h while maintaining the same track geometry as before. Figure 7 shows an overall similar behavior of the roll and steer angle as before. However, it must be noted that in this case the counter steering angle is not the result of a steer torque. The initial rider leaning motion towards the direction of the lane change causes an opposing support torque in the motorcycle frame. (i.e. as the rider moves right, the motorcycle frame wants to move to the left). This is observable in a small roll angle peak at about 62 m in the opposite direction of the lane change. This will in consequence cause a small steering angle in the same direction as well (see magnified representation in the top right corner). Only then, as the tilted front wheel causes sideslip and camber side forces, the roll motion towards the direction of the lane change will begin. After that, a similar process leads to returning the motorcycle to a straight riding.

Figure 7. Average values and standard deviations of dynamic quantities in \(\mathbf{DL}_{\mathbf{L1}}^{\mathbf{H0}}\) condition.

The previous timeseries have shown, that adding a control cue sensitive to rider motion can cause plausible vehicle reactions of the virtual motorcycle and even allow a rider to maneuver it without hands. The data shows, that the rider is applying a lean-in strategy to maneuver through the scenario. As known from literature, it is expected that lean-in should have a decreasing effect on the steer torque. To gather an overview about this effect, Figure 8 shows a statistical analysis of the Lane-Change-Roll Index \(\Upsilon_{LC}\) (Cossalter, 2006) that relates the peak to peak steering torque \(T_{hb,p - p}\) to the product of the mean velocity  \(\overline{v}\) and the peak to peak roll rate \({\dot{\varphi}}_{th,p - p}\). When comparing different vehicles or vehicle configurations, smaller values of the Lane-Change-Roll Index are considered superior to larger values, as the rider needs less effort to control the vehicle through the lane change, which arguably results in better “handling”. As observed in Figure 8, all experiments at velocities above 50 km/h result in smaller \(\Upsilon_{LC}\) for the \({DL}_{L1}^{H1}\) condition. Only at 30 km/h, the index increases when adding the leaning input, which can be explained by higher steering efforts needed due to a decreasing vehicle stability at such speeds and the tendency to utilize lean-out. As there is no absolute rating possible for the Lane-Change-Roll Index, it is not sufficient to rating the performance of the DLRC.

(Cheli, 2011) investigates the rider’s body influence in double lane change maneuvers. The results indicate, that the utilization of rider motion will result in a delay in the peak steer torques and steer angles with respect to the peak roll angles. Figure 9 shows the absolute locations (0 m indicating the start of \(l_{mid}\)) of peak values of multiple quantities on the left side and the relative distances between the peaks of different quantities on the right side. The data shows that the configuration with active DLRC results in little to no delay of the absolute locations of the roll angle peak \(s({\widehat{\varphi}}_{BRT})\), steer torque peak \(s({\widehat{T}}_{\delta,BRT})\) and steer angle peak \(s({\widehat{\delta}}_{BRT})\). The relative delay from the roll angle peak to the steer angle peak \(\Delta s_{\delta,\varphi}\) – if anything – decreases slightly.

Figure 8. Lane-Change-Roll-Indexes for two configurations at different velocities.

The behavior described by (Cheli, 2011) could therefore not be reproduced. However, to the knowledge of the authors, there exists no definition of an ideal behavior in that regard. The delay or advance of the peak locations may vastly differ when utilizing lean-in or lean-out. But again, as there is no absolute reference available, this measure is not well suited to rate the performance of DLRC.

Figure 9. Locations of peak values and delays between the peaks of different quantities at lane change with 70 km/h.

Due to the described deficiencies of the known indexes, a new approach was developed and firstly described in (Hammer, 2021), that aims to compare the performance of a lane change by fitting a hyperbolic tangent function to the motorcycle trajectory by minimizing the loss function shown in Equation (2).

Therein, the hyperbolic tangent function is offset by \({\widetilde{x}}_{mid}\) and \({\widetilde{y}}_{0}\) and scaled by the maximum offset rate \(\max(\frac{\partial\widetilde{y}}{\partial\widetilde{x}})\) and amplitude \(\Delta y\tilde{}\). These four fitting parameters as well as the RMS value allow statements about the precision of the lane change maneuver. The initial offset \({\widetilde{y}}_{0}\) should ideally be zero, and the amplitude \(\Delta y\tilde{}\) should ideally match the width of the lane change maneuver. The maximum offset rate \(\max\ (\frac{\partial\widetilde{y}}{\partial\widetilde{x}}\ )\) shows, how quickly a rider performs the lateral transition and the RMS value rates the quality of the tanh-fitting. Small values will indicate a smooth, undistorted trajectory. Figure 10 (left) shows the trajectory of the motorcycle’s CoG (blue) and the hyperbolic tangent that is fitted on this trajectory (green). The abovementioned parameters of this fitted trajectory define the green bubble in the righthand plot. The RMS value is depicted by the size of the bubble, the ordinate shows the lateral offset and the abscissa shows the maximum offset rate. Both latter variables have been normalized by the target offset.

Figure 10. Development of the Bubble-Cluster Plot.

In this representation, an optimally performed lane change would result in a single point (rather than a large bubble) that is placed exactly at the 100% line, while the maximum lateral change rate depends on the length that is available to perform the lane change. During 10 repetitions, the resulting points should show a minimal scattering, indicating a good repeatability and therefore good controllability and ease of riding.

In the experiment, we see differently sized bubbles that are scattered around a center. The red and blue color indicate the configuration of the simulator – red showing the DLRC condition. The diamonds represent the mean location and size of the bubbles of all ten repetitions. It can be seen from this example, that the maximum lateral change rate decreases, when DLRC is activated and the RMS error is slightly smaller compared to the configuration without the leaning input. However, no effect is visible for the reached lateral displacement.

Figure 11 shows the bubble clusters for all tested velocities with and without the use of DLRC. For a better comparability, the abscissa was additionally normalized by the length \(l_{mid}\). Starting at high velocities (rightmost plot) it can be seen, that the activation of DLRC causes a slight decrease of the RMS error and a shift of the cluster towards lower maximum lateral change rates. This indicates a smoother transition from the center lane to the outside lane. While this supports the previously shown results collected at 70 km/h, the picture is less clear when looking at the 90 km/h and 50 km/h data that show a less pronounced separation of the two clusters. At 30 km/h, the size of the bubbles increases clearly, indicating much larger deviations of the trajectory from the hyperbolic tangent function, which is plausible for the lesser amounts of self-stabilization given at this speed. With activated DLRC, the overshooting of the target lateral offset decreases slightly.

Figure 11. Comparison of bubble clusters in two configurations for different velocities.

Conclusion and Outlook

The described implementation of DLRC aims to increase the rideability of a dynamic motorcycle riding simulator, as it allows the rider to utilize behavioral patterns known from real life riding. The data shows, that the presented determination of the rider induced roll torque is an effective method to design a DLRC. The system is capable of provoking plausible vehicle reactions known from real life riding. In order to objectively rate the effectiveness of this method, known indices like the Lane-Change-Roll Index or the peak locations have been investigated. The results however show, that they are not really suited to rate the effect of DLRC on the rideability of the simulator. Therefore, another set of characteristic values was developed to try and gain insight on the lane change behavior. First results show small but noticeable differences in the measurements that indicate smoother lane transitions with less steer torque efforts and less scattering, when DLRC is activated. However, the amount of available data is yet too small to provide final results.

Therefore, in the next steps, more data of various participants should be collected to improve the statistical evidence. Furthermore, the newly defined characteristic values might need to be revised, by e.g., adjusting the fitting function from a hyperbolic tangent to a different sigmoid, or by investigating e.g., the tire trajectories rather than the CoG trajectory. In general, the method provides absolute values that could be used for comparison of different real and simulated vehicles in various lane change parametrizations. The shown representation of bubble clusters provides a quick overview of the data and allows for data clustering, such that it promises to be feasible for analytics of larger data bases as well.

References

Åström, K. J., Klein, R. E., & Lennartsson, A. (2005). Bicycle Dynamics and Control. Control Systems Magazine, 25(4), 26-47.

Cheli, F.; Pezzola, M.; Taroni, N.; Mazzoleni, P.; Zappa, E. (2011) Driver’s movements influence on the lateral dynamic of a sport motorbike, 19th Mediterranean Conference on Control and Automation

Cossalter V., (2006). Motorcycle Dynamics. Lulu, Raileigh

Cossalter, V.; Lot, R.; Massaro, M.; Sartori, R.: Development and validation of an advanced motorcycle riding simulator. Proceedings of the Institution of Mechanical Engineers, Part D: Journal of Automobile Engineering (6), Issues 225, pp. 705–720, 2011

Guth, S. (2017) Absicherungsmethode von Anzeigekonzepten zur Darstellung fahrfremder Informationen mittels eines Motorrad-Fahrsimulators, Dissertation, TU Darmstadt

Hammer, T.; Pleß, R.; Will, S., et al. (Eds.) (2021). Anwendungsmöglichkeiten von Motorradsimulatoren, Berichte der Bundesanstalt für Straßenwesen, Mensch und Sicherheit Heft 323, Fachverlag NW in der Carl Schünemann Verlag GmbH, Bremen, Germany

Pleß, R. (2016). Approach to a holistic input determination for a motorcycle riding simulator. Proceedings of the Bicycle & Motorcycle Dynamics Symposium, Milwaukee, 2016

Pleß, R. (2023). Dual Loop Rider Control of a Dynamic Motorcycle Riding Simulator, Dissertation, TU Darmstadt

Werle, A., Diermeyer, F. (2022). An Approach for Steering Advancement in Motorcycle Riding Simulation. Human Interaction & Emerging Technologies (IHIET-AI 2022)

Westerhof, B. E. (2018). Evaluation of the Cruden Motorcycle Simulator, Master-Thesis, Delft University of Technology