Motion sickness on tilting trains

Abstract

Trains that tilt on curves can go faster, but passengers complain of motion sickness. We studied the control signals and tilts to determine why this occurs and how to maintain speed while eliminating motion sickness. Accelerometers and gyros monitored train and passenger yaw and roll, and a survey evaluated motion sickness. The experimental train had 3 control configurations: an untilted mode, a reactive mode that detected curves from sensors on the front wheel set, and a predictive mode that determined curves from the train's position on the tracks. No motion sickness was induced in the untilted mode, but the train ran 21% slower than when it tilted 8° in either the reactive or predictive modes (113 vs. 137 km/h). Roll velocities rose and fell faster in the predictive than the reactive mode when entering and leaving turns (0.4 vs. 0.8 s for a 4°/s roll tilt, P<0.001). Concurrently, motion sickness was greater (P<0.001) in the reactive mode. We conclude that the slower rise in roll velocity during yaw rotations on entering and leaving curves had induced the motion sickness. Adequate synchronization of roll tilt with yaw velocity on curves will reduce motion sickness and improve passenger comfort on tilting trains.—Cohen, B., Dai, M., Ogorodnikov, D., Laurens, J., Raphan, T., Müller, P., Athanasios, A., Edmaier, J., Grossenbacher, T., Stadtmüller, K., Brugger, U., Hauser, G., Straumann, D. Motion sickness on tilting trains.

Trains whose car bodies are actively tilted were first introduced >70 yr ago and are now in wide use over the world (1–11). The tilting compensates for the centripetal acceleration during turns by bringing the vertical axes of the cars closer to the gravitoinertial acceleration (GIA), which is the summation of the equivalent acceleration of gravity and the centripetal acceleration. As a result, the trains can run faster, and the thrusts of centrifugal force on the passengers during turns are reduced. Unexpectedly, passengers developed symptoms of motion sickness (nausea, sweating, hyperventilation, pallor, and drowsiness; refs. 12–14). A number of studies have considered which aspects of the tilting during turns contribute to the motion sickness, but, as yet, there is no definitive answer to this problem, and no effective countermeasures have been proposed (12–16).

Forstberg et al. (12–14) and Persson (15) showed that passengers had less motion sickness if the angle of the compensatory roll tilt and the lateral acceleration were reduced. The difficulty with this approach was that the trains would have to run slower on curves, negating the purpose of the tilt, which was to allow trains to go faster. Donohew and Griffin (17–19) and Golding et al. (20, 21) came to a somewhat different conclusion, namely, that it was the roll tilt and the lateral acceleration that were the most important factors. Vertical linear acceleration could also be a contributing factor, since, as shown by Golding et al. (22), oscillation along the body vertical (z) axis is more effective than oscillation along the body horizontal (x) axis in producing motion sickness. Neimer et al. (23) found that a tilted visual scene, viewed from the fore-aft side of the car, caused significant motion sickness if subjects did map reading and moved about during the tilts. Whether there was any statistical difference between motion sickness induced with and without these factors was not evaluated. A critical finding was presented by Joseph and Griffin (24), who showed that if subjects' heads were tilted during lateral acceleration, they had strong motion sickness, but if the head roll was initiated before the lateral acceleration, there was no motion sickness. This clearly implicated the relationship between roll tilt and linear acceleration in inducing motion sickness. They postulated that this was the cause of the motion sickness on tilting trains.

The preceding studies have concentrated on the general relationship between roll and lateral and/or vertical acceleration of the head. Many models of motion sickness have been proposed (see ref. 25 for review), but, as yet, the exact neural mechanisms that generate motion sickness have not been firmly established, especially for the motion sickness that occurs on tilting trains. In this regard, the centripetal acceleration generated during a turn is proportional to the product of the square of the angular velocity and the radius, making the angular velocity during the turn a dominant factor in the generation of the centrifugal force. Therefore, a critical question is whether the tilt during the significant yaw angular velocity could be responsible for the induction of motion sickness, which was not considered.

Miller and Graybiel (26, 27), working in the NASA space program, demonstrated that rolling the head while rotating is a potent stimulus for induction of motion sickness. Dai et al. (28) have further shown that this stimulus, which brings the vertical canal planes closer to the plane of rotation and maximally activates them in phase with head tilt, is particularly effective for generating motion sickness. On the basis of this work, we postulated that the motion sickness in tilting trains was caused by inadequate synchronization of roll and yaw angular velocities, which would compensate for the tilt while rotating, as the cars entered and left the turns. Because the turning of the train occurs at an eccentric radius of curvature, a lateral acceleration is induced. As a result, our postulate about motion sickness on the tilting trains could also encompass the effect of inappropriate phasing of lateral acceleration with the roll tilts, as described by Joseph and Griffin (24).

The present study was initiated by the Schweizerische Bundesbahnen (SBB) to investigate this problem, and if possible, to determine what causes motion sickness on tilting trains. We designed a wireless system to capture data from accelerometers and gyros, which were placed on the train and on the heads of passengers. We also conducted a survey of the incidence of motion sickness and passenger comfort. Preliminary results have been presented (29).

MATERIALS AND METHODS

General

The experiments were conducted from early September to late October, 2009. The same train, an SBB-RABDe 500 (Fig. 1A) that could tilt as it rounded curves (Fig. 1B) was used in the experiments. Of the 7 cars that composed the train, the first and the last cars had a cabin for the operator and seats for the 40 passengers, who were the subjects on each ride. Passengers always sat in the first car. This was chosen because motion sickness sensations that were reported from SBB files were greatest when passengers rode in the first car. To reverse direction at the end of each ride, the engineer and passengers walked to the other end of the train in the interval between runs. The distance from front to back was ∼150 m and took about 5 min to traverse. There were 2 round trips from Winterthur to Gossau in the morning and 2 in the afternoon, a total of 8 rides each day. The ride in each direction took ∼22–25 min and was followed by a 15–30 min rest before the return trip. The morning and afternoon sessions were separated by a 90-min lunch in Winterthur. Only one tilt control condition was tested on each round trip.

Figure 1.

A) Train (SBB-RABDe 500), similar to the train that was used in the experiments between Winterthur and Gossau, Switzerland. B) Typical tilt of the train on a curve. Note the cant of the tracks, which also helped reduce the tilt of the GIA vector relative to the car and passengers. (From Schweizerische Bundesbahnen) C) Tilt of a wagon car from an intercity tilting train by mechanisms on the bogies. (From Swiss Tilting Trains Modernisation Programme, Railway-Technology, June 20, 2010, p. 3; http://www.railway-technology.com/projects/sbb/).

Subjects

Two hundred subjects were recruited by SBB through e-mail from all of Switzerland. On each test day, 20 normal and 20 motion sickness-susceptible subjects came to Winterthur to participate in the train rides. They received 250 Swiss francs for participating. The motion sickness-susceptible subjects were drawn from the SBB files, based on their previous complaints. On each day, a new cohort of 40 passengers rode 4 abreast in the first car. Subjects were under no specific constraints and chatted freely with one another. A refreshment cart was wheeled through the car on each run to provide complimentary beverages. Rest rooms were available at the end of the car. Subjects rode forward or backward, on the inside (aisle) or outside (window) seats, with and without views of the passing visual scene. The passengers filled out a questionnaire every 10 min to establish their relative level of motion sickness and comfort.

Equipment and control signals

The tilt system consisted of angular velocity sensors (gyro: x, roll; z, yaw), and lateral acceleration sensors mounted on the front bogie of the first car, a processor control module, and electromechanical actuators. Each car had similar tilt apparatus mounted on its bogies. The control module on the first car produced the signals to tilt each subsequent car with a calculated delay, based on the measured yaw and roll angular velocities, lateral acceleration, and train velocity. The train had a specially designed suspension, in which the roll rotation of the car occurred about an axis that lay along the middle of the car at the level of the passengers seats, ∼143 cm from the track. Typical bogies with the car body in a tilted position are shown in Fig. 1C.

Cars were also subject to 2 angular movements. The turning movement on the curves caused the cars to rotate in yaw, i.e., in the horizontal plane. The cars also tilted inward toward the center of the turn, i.e., roll rotation, to partly compensate for the centrifugal acceleration. The inward tilting was due to a combination of the cant of the tracks and tilting of the car bodies on their bogies (Fig. 1B, C). This tilting was resisted by springs in the bogies that held the cars stable. In addition, the cars had oscillations in roll at ∼0.5 Hz and 1.67 Hz. Lateral oscillations were predominant at 0.5 and 1.67 Hz, and vertical oscillations were ∼1 Hz.

There were 3 control modes during the rides. In one, the train ran without active tilt. This was designated as the no-tilt mode. In the other two modes, the train actively tilted as it entered the curves, and the tilting was reversed as the train left the curves. In the reactive mode, which is now in wide use in Switzerland, the onset and end of the curves were sensed by accelerometers on the front wheel-set or bogie. This signal was used to initiate and terminate the tilting of the front car, and, with appropriate timing, the other cars in the train. Three angles of tilt (0.5, 1.24, and 8°) were studied in the reactive mode. In the second or predictive mode, in which the tilt was 8°, the onset and end of the curves were determined from a look-up table, based on the geographic position of the train on the tracks (6). The control signals were then relayed to the other cars in the train with appropriate delays.

Track properties and train behavior

The track between Winterthur and Gossau was 42.5 km long and had 58 turns. The curved tracks were 76% of the distance, and the tangential (straight) segments were 24%. The turns consisted of 3 parts: an initial transition curve, a circular arc, and a terminal transition curve. The circular part of the arc could be absent. During the turns, the track curvature changed smoothly from 0° to a constant value and back. The approximate average radius of curvature of the turns was >400 m. The change of heading direction during the turns (arc length) was 2–110°. The speed of the train as it rounded the turn was on average ∼120 km/h (72 mph), which was maintained constant around the curves. This resulted in an average tilt of the GIA of 10–12°. There was also a counterroll due to the spring action in the curves of up to 3°. The cant, i.e., the difference in height of the inside and outside rails (Fig. 1B), was up to 16 cm, which banked the tracks up to 6.4°. In both the reactive and predictive modes, the cars could be tilted up to 8°, which when coupled with the cant of the tracks in the curves, could align the yaw axis of the train with ∼75% of the GIA tilt.

RESULTS

The duration of the ride in each direction was ∼22 (22–25) min at an average velocity of 120 km/h (72 mph), with a maximum velocity of 155 km/h (93 mph). The velocity was dependent on the tilt of the cars. Velocities were slowest in the no-tilt condition and fastest when the cars were tilted to a maximum of 8°, either in the reactive or predictive mode (Table 1). During turns, the GIA vector was tilted by a combination of the centripetal linear acceleration produced during the turn and the vector of the upward equivalent linear acceleration of gravity. Both the reactive and predictive modes produced similar compensation, i.e., ∼75% of the tilt in roll of the GIA vector, based on standards developed in the United Kingdom in the 1970s, and widely accepted in Europe [passenger transport comfort (PTC)]. Since the tracks were also banked in the curves (Fig. 1B), this brought the total compensation for tilts closer to the tilt of the GIA.

Table 1.

Tilt, tilt rates, typical velocities, and lateral accelerations on curves

Train tilt control mode	Maximum active tilt (deg)	Maximum tilt rate (deg/s)	Typical velocity of train on curves [km/h (mph)]	Typical lateral acceleration (m/s2)
No tilt	0	0	113 (67.8)	1.8
Reactive	0.58	0.4	125 (75.0)	2.1
Reactive	1.24	2	130 (78.0)	2.4
Reactive	8	5	137 (82.2)	2.7
Predictive	8	5	137 (82.2)	2.7

See text for details.

On a typical portion of the track, the centripetal acceleration changed linearly during turns over 2.5–3.0 s. This was due to smooth changes in the yaw angular velocity from 0° to 2–4°/s and back (Fig. 2B). The approximately circular arc phase lasted from 0 to 22 s. The total car roll varied from 5 to 12° with a peak speed of 5°/s (Fig. 2A).

Figure 2.

Roll velocity (A) and yaw velocity (B) recorded in a typical portion of the track between Winterthur and Gossau from sensors fixed to the train. There were 10 turns, which produced 20 changes in roll velocity in 200 s. The train was in a reactive mode in these recordings. A) Dashed vertical lines and horizontal arrows show 2 typical rising and falling roll velocities that took 2.7 and 4.5 s. B) These roll velocities were associated with sustained increases in yaw velocity of 4°/s. Abscissa is 200 s; ordinates range from −10 to +10°/s.

The passengers made many head movements as they talked with each other and looked out of the window. These head movements, recorded with an accelerometer from a typical passenger, are shown as vertical spikes in the first (roll) and second (yaw) blue traces of Fig. 3. There were also shifts in the baseline in both roll and yaw velocities. When the car roll and yaw velocities (Fig. 3, third and fourth traces), were superimposed on the passengers' head roll and yaw velocities (Fig. 3, red traces), it was apparent that the passengers had experienced the same yaw and roll velocities on average as the car. The passengers also experienced rotations of the GIA from the spatial vertical in association with the shifts in yaw velocity (Fig. 3, bottom trace). This was also present in the power spectra of the head and car frequencies, which were similar (not shown). Overall, in the typical segment depicted in Fig. 2, there were 10 turns in 200 s (3.3 min; Fig. 2B), which were associated with 20 roll tilts of the car (Fig. 2A). Since there were 58 increments in yaw velocity and 116 in roll velocity on each of the 8 runs, every passenger experienced a total of 464 increases in yaw velocity and 928 changes in roll velocity on the day that they were subjects.

Figure 3.

Top and third blue traces: yaw and roll angular velocities from sensors on the head of a passenger. Train was making 8° tilts in the reactive mode. Second and fourth blue traces: roll and yaw velocities from sensors on the car. Red lines in top and third traces are train roll and pitch (second and fourth traces), drawn over the recordings of head roll and yaw. Despite many adventitious head movements, the passenger experienced the same yaw and roll velocities as the train. Bottom trace: uncompensated shifts in the GIA relative to the train vertical. Time base (abscissa) is 300 s; ordinates range from −10 to +10°/s or from −10 to +10° (bottom trace).

Comparison of the behavior of the reactive and predictive modes on the same curve (Fig. 4) revealed a number of differences between them. The control signals emanating from the system in the predictive mode (Fig. 4, second trace) were sharper and had less noise than in the reactive mode (Fig. 4, top trace). Because the predictive control system relied on stored information, based on the geographic position of the train on the rails, the time of initiation and the angle of the tilt could be calculated in advance. This allowed the train to roll-tilt faster as it entered and left the curve. Jitter on the reactive control signal was also reflected in oscillations in lateral acceleration (Fig. 4, third trace) and roll velocity (Fig. 4, fourth trace, upward arrow) at 0.5 and 1.7 Hz, the natural oscillation frequencies of the cars. Furthermore, roll velocity was less well maintained during the constant velocity portions of the curve in the reactive mode (Fig. 4, third and fourth traces). In contrast, because of the cleaner control signal in the predictive mode (Fig. 4, second trace), there was less jitter in lateral acceleration and roll velocity of the car (Fig. 4, fifth and bottom traces). Peak roll velocities were also held more constant during the onset and end of the turns in the predictive mode (Fig. 4, bottom trace).

Figure 4.

Control signals (top and second traces), lateral accelerations (third and fifth traces), and roll velocities (fourth and bottom traces) recorded with the train in the reactive mode (top, third, and fourth traces) and the predictive mode (second, fifth, and bottom traces). Upward arrow in the reactive roll velocity (fourth trace) has oscillations that were not present in the predictive roll velocity. Calibrations of the signals are at left and right. Time base is in seconds and is shown at bottom.

The reactive mode also did not approximate the changes in roll of the GIA as well as the predictive system during the turns (Fig. 4, vertical lines, fourth and bottom traces). This can be appreciated better when roll velocities at the onset and end of the same turn are viewed on a faster time base (Fig. 5, top and third traces). Changes in roll velocity began at 10 s into the traces in Fig. 5 and ended at 12 s. The times over which tilts to 4°/s occurred were, on average, 0.8 s for the reactive mode and 0.4 s for the predictive mode, a significant difference (P<0.001). Thus, the slopes of the rise and fall in car roll velocities in the predictive mode (Fig. 5, third trace) were twice as steep, i.e., twice as fast, as in the reactive mode (Fig. 5, top trace). Changes in the decrease of roll velocity were similar, i.e., twice as fast for the predictive as for the reactive mode.

Figure 5.

Car roll velocity (top and third traces) and car linear accelerations (second and fourth traces) for the reactive mode (top and second traces) and predictive mode (third and bottom traces) during the onset of 5 curves. Same curves are represented in the top and second as in the third and bottom traces. Calibrations are at right; time base is at bottom.

Some of the passengers wore accelerometers on their head during the rides, but they were not constrained to hold their head fixed. Therefore, there were many yaw and roll movements (Fig. 3, top and second traces) as they turned to look out of the window or to talk to their fellow travelers. One person, however, kept her head relatively stationary during both the 8° reactive and predictive rides, and average roll velocities of the passenger (Fig. 6A, C), and train (Fig. 6B, D) were determined over 30 turns on both the reactive and predictive rides. Most apparent was that the head roll velocities followed the train roll velocities during the tilts, but the rise in velocity occurred faster in the predictive (Fig. 6C, D) than the reactive mode (Fig. 6A, B). The average roll velocities were larger for the head roll (Fig. 6A, C; 10°/s) than for the train roll (Fig. 6B, D; 4°/s), which was probably due to a mixture of head movement velocities added to the train roll velocity. There was also an average delay in achieving the constant velocity phase in the reactive mode (Fig. 6A), but since the passengers were free to turn their heads, this may or may not have had a significant effect. Regardless, these data confirm that the rise in head and car velocity were slower at the beginning and end of the turns in the reactive mode. In addition, roll velocity was held more stably by the tilting cars in the predictive than the reactive mode.

Figure 6.

Comparison of head (A, C) and train (B, D) roll velocities during 30 turns on the rails between Winterthur and Gossau with the train in the reactive mode (A, B) and in the predictive mode (C, D). Data in A, B and C, D are from the same segments of track. Four thin vertical lines show the onset and end of train roll in the predictive mode in D. Note the faster rise and fall in train velocity in the predictive mode (D). Lines are projected vertically to demonstrate that roll velocity rose and fell more slowly in the reactive mode (B). Head velocity (A, C) mirrored the differences in train roll velocity in the two modes. Vertical bars at right are calibrations for the train and head velocities. Head roll velocities (A, C) were larger than train roll velocities (B, D). Horizontal bar in D is the time base calibration (1 s) for all of the traces.

Compensation for roll of the second car was also slower during the turns in the reactive mode, and roll velocity ended 3–4 s after the onset of yaw velocity and lateral acceleration of the first car (not shown). In contrast, compensatory tilts of the second car were concluded in <1 s after the first car at the onset and end of the turns when the control signals came from the predictive system.

The motion sickness questionnaire, shown in Fig. 7, was adapted from the Modified Pensacola Scale for clinical use and validated by Lyne (30), Young et al. (31), and Dai et al. (28, 32, 33).

Figure 7.

Passenger questionnaire (English translation of original questionnaire in German). Scale was adapted from the Modified Pensacola Scale (Lyne, ref. 30).

A 1-way ANOVA was performed on all passengers for nausea after 30 min of travel. There were 5 conditions (4 degrees of freedom). Half of the passengers had complained earlier of motion sickness in tilting trains. The analysis did not consider the fact that not all passengers were tested under the same conditions: 160 passengers were tested for reactive 8°, reactive 1.24°, reactive 0.58°, and no tilt; 40 passengers were tested for reactive 8° and predictive 8° tilt, each condition twice. The test was highly significant (Fig. 8A; P<0.001). A post hoc comparison with a Bonferroni correction, using a value of α < 0.05, showed that the reactive tilting was significantly different from either the predictive or the no-tilt condition (P<0.05).

Figure 8.

A) Increase in nausea scores after 30-min rides with the train in various configurations. B) Relative comfort of train rides in 10-min segments with the train in various configurations. Light shaded bars indicate scores after 10 min; medium shaded bars, after 20 min; solid bars, after 30 min.

The difference in comfort level of travel after 30 min was also highly significant (Fig. 8B; P<0.001). A post hoc comparison among the 5 conditions using a Bonferroni correction revealed that comfort in reactive mode using an 8° tilt was significantly lower from either no tilt or the predictive mode (P<0.05). The analysis also showed that no tilt was significantly different from any of the reactive modes and that the predictive mode was significantly different from the reactive mode using the same angle.

The same analysis was performed for the half of the passengers who had complained of motion sickness. Again, the 1-way ANOVA of nausea after 30 min was highly significant (P<0.001). Post hoc testing with Bonferroni correction showed similar results, as in all passengers. Nausea levels in the reactive mode, using the 8° tilt, was significantly different from the predictive mode and from no tilt (P<0.05). The untilted condition was significantly different from the reactive mode (P<0.05), but was not different from the reactive mode at 1.24° tilt (P<0.05). However, the predictive mode was significantly different from the reactive mode at 8° tilt.

The 1-way ANOVA on the comfort of travel after 30 min also indicated that it was highly significant in subjects prone to motion sickness (P<0.001). Post hoc testing with Bonferroni correction again showed similar results as for all passengers. The reactive mode at 8° tilt was significantly different (P<0.05) from the no-tilt and predictive mode at the 8° tilt conditions. Untilted was significantly different from the reactive modes at tilts of 8° and 1.24° but not different from the reactive mode tilts at 0.58°. The predictive mode at 8° tilt was significantly different from the reactive mode at the 8° tilt.

Despite findings of Neimer et al. (23), there was no significant difference in the comfort of passengers riding forward or backward, next to the aisles vs. next to the windows, or with an outside view vs. no outside view. Therefore, none of these factors had contributed measurably to the occurrence of motion sickness or to passenger comfort. There were significant differences in the effects on posture and locomotion between the two modes of control. It was much easier to maintain upright posture in the aisles in the predictive mode. Postural stability was tested standing with eyes closed, hands at side and feet together (Romberg position). Sharp lateral thrusts of centrifugal force at the onset and end of the curves disturbed upright posture during the untilted rides and during rides made with reactive control. In contrast, standing and walking in the aisles during rides with predictive control were much easier, and one was not likely to be thrown laterally in the direction of the centrifugal force as cars entered and left turns.

DISCUSSION

The major finding of this study is that motion sickness on tilting trains can be essentially eliminated by appropriate phasing of the tilts of the cars as they enter and leave the turns. As shown in Figs. 5 and 6, there were significant differences between the dynamics of the predictive and reactive modes. Roll velocities generated by the tilt of the car rose and fell twice as fast in the predictive as in the reactive mode at the beginning and end of the turn. Roll velocities were also maintained more stably during the constant velocity portions. As a result, the tilt of the cars followed the tilt of the GIA more precisely as the train rounded curves. Associated with this, more motion sickness and discomfort were produced by rides with the 8° reactive tilts than during rides without tilt or with 8° predictive tilts (Fig. 8A). Moreover, the amount of motion sickness increased steadily with increases in the angle of tilt in the reactive mode (Fig. 8A). Since a different cohort of 40 passengers rode each day, there was no day-to-day habituation to cloud the findings. Passenger comfort was also greater in the rides with the predictive mode, and there were similar levels of comfort in each 10-min period during the rides (Fig. 8B). This also indicated that there had been no habituation. Of interest, the motion sickness was equivalent in the passengers that had complained of it before as in the passengers that had had no previous complaints. This validates the power of the analysis.

Motion sickness scores were low, both for rides with no tilt and with the 8° predictive tilt, but the rides with predictive tilt were 21% faster, and the passengers were not subjected to the forceful thrusts of lateral acceleration on each turn when walking in the aisles. Consistent with findings from the test subjects, less motion sickness and discomfort were experienced by the 3 investigators who had had previous motion sickness susceptibility during the 8° predictive than the 8° reactive tilts. Standing and walking in the aisles was also maintained much better in the predictive than the reactive or untilted rides.

The critical questions, then, are the difference between the predictive and reactive modes that caused the motion sickness and how the motion sickness was produced. Since the rides with 8° tilts were done on the same train, riding on the same track at the same speeds, a number of factors can be eliminated. The difference between the predictive and reactive modes was not due to the angle of tilt of the cars nor the magnitude of the linear acceleration when rounding the curves, since they were the same in both conditions. Oscillations in vertical linear acceleration, which can be a potent cause of motion sickness (22) were also approximately the same during the reactive and predictive rides. It also made no difference whether the subjects were riding forward or backward, or whether they saw a tilted visual scene, although this has been considered as a source of motion sickness on tilting trains (23).

As the train goes into a curve, the radius gradually decreases while the train maintains the same linear velocity. This generates a buildup of centripetal acceleration, and as a consequence, an inward tilt of the GIA. On the basis of the roll velocity of the tilts shown in Figs. 5 and 6, it can be concluded that the roll tilt of the cars in the predictive mode was closer to the onset of the buildup of lateral acceleration in the predictive than in the reactive mode.

From this, we conclude that the major difference between the predictive and reactive modes was in the rise and fall of roll velocity, which reflects the inward shift of the GIA. It was approximately twice as fast in the predictive as in the reactive mode. We attribute the better performance of the predictive mode to its ability to estimate the buildup of centripetal acceleration better and the tilt of the GIA when going around curves.

This is relevant because Joseph and Griffin (24) have shown in a decisive experiment that if the head is rolled passively and then subjected to lateral acceleration, there is no motion sickness. If, however, the roll and lateral acceleration overlap, motion sickness is induced. The severity of the motion sickness increases as a function of the amount of the overlap. This is directly applicable to the current situation, i.e., if the roll occurred close to the onset of the lateral acceleration, as shown in Figs. 5 and 6 in the predictive situation, there was little or no motion sickness. If, however, the rise in roll was much slower, and overlapped the lateral acceleration, motion sickness was generated.

In addition, the situation in which there was roll of the head during yaw velocity recreated a condition known as roll while rotating. This paradigm has been studied experimentally (26–28), and it could also contribute to the induction of motion sickness in passengers on the tilting trains. As shown in Fig. 2, the passengers experienced 4°/s of yaw velocity during the turns that was coincident with train and head roll. The head roll during yaw rotation tends to bring the vertical semicircular canals closer to the plane of rotation, which produces a sense of pitch and nausea (28). It might be questioned whether a rotation of 4°/s is adequate to produce motion sickness in this paradigm, but it has been confirmed in both New York and Zurich laboratories (unpublished data). If passengers turned their head to look out of the windows or to talk to their neighbors, it would induce pitch while rotating, which would also activate the vertical canals, and this is also known to be a nauseating stimulus (34, 35). The mechanism occurs through a process in the vestibular system called velocity storage, which tends to maintain the axis of eye rotation close to the prevailing GIA (36, 37). A disparity between the yaw orientation axis of velocity storage and the angular velocity of the head triggers motion sickness. For small angles of disparity, the intensity of the motion sickness is linearly related to the displacement of the eye velocity vector from the spatial vertical (28, 32, 33). The original studies on roll while rotating and pitch while rotating were done at higher rates of rotation than those experienced on the tilting trains, which were in the range of 4°/s. However, much slower rates of rotation can also produce nausea in susceptible subjects (unpublished data). The GIA for the tilting train is along the sum of the equivalent acceleration of gravity and the centripetal acceleration. As a result, if the tilt of the train were synchronized with the changes in the GIA, the effects of tilt while rotating on velocity storage would be eliminated.

The data from the small tilt, reactive rides (0.8 and 1.24°) provide strong additional evidence that it was the delay in tilt that was responsible for producing the motion sickness. The finding that even the smallest reactive tilt (0.8°) caused more motion sickness than the 8° predictive tilt is particularly telling, as is the increase in motion sickness with the successive small increments in the tilt angles. These data confirm that it is not the angle of tilt that was responsible for producing the motion sickness, but the delay in having the vertical axis of the train move toward the GIA. Thus, regardless of the angle of compensation of the tilts, whether 0.8, 1.24, or 8°, they should be sharp in onset, fast in rise and fall times, and precise in maintaining whatever level of compensation in tilt they can provide. The control signals should also be free of oscillations in roll, regardless of the amplitude of the tilts.

Regardless of the relative contribution of the mechanisms related to lateral acceleration and/or yaw rotation while rolling the head, it is certain that the otolith system, which senses linear acceleration and head position relative to gravity, was a major source of the activity that produced the motion sickness. Velocity storage, which maintains its orientation vector close to the GIA or spatial vertical (see ref. 38 for review) was also probably heavily involved in producing the motion sickness though activation of the autonomic system (28, 34).

Finally, since there were both comfortable and uncomfortable rides with the same equipment, it should be possible to reconfigure the control signals on the present equipment by simply changing to a predictive mode that senses the geographic position of the train relative to the curves, and not by depending on a reactive system that obtains its information about the upcoming curves from sensors on the front bogies. In this regard, it was of interest that there was no difference between the motion sickness of the subjects who had complained of motion sickness previously on the tilting trains and the so-called normal passengers who did not. The motion sickness was the same in both groups. This suggests that if the motion sickness is abated by appropriate control of the speed of the tilts, that it will have an equal effect in reducing the motion sickness in both groups, thereby generalizing the results of this investigation.