- Email: [email protected]
A Cybernetic Approach to Assess Flight Simulator Motion Fidelity
A Cybernetic Approach to Assess Flight Simulator Motion Fidelity ⋆ D.M. Pool, P.M.T. Zaal, H.J. Damveld, M.M. van Paassen and M. Mulder ∗ Control and Simulation Division, Aerospace Engineering, Delft University of Technology, 2629 HS, Delft, The Netherlands (e-mail: [email protected], [email protected], [email protected], [email protected], [email protected]). ∗
Abstract: Due to a limited understanding of human multimodal motion perception during aircraft control, the definition of requirements for flight simulator motion fidelity is a problem the flight simulation community has struggled with for years. The development of adequate tuning procedures for motion washout algorithms has remained problematic for the same reason. This paper proposes a method for assessment of simulator motion fidelity that makes use of a cybernetic, model-based approach to measuring multimodal pilot control behavior. As illustrated here with data from a recent human-in-the-loop experiment, this approach allows for objective detection of changes in pilot control that result from degraded simulator motion fidelity. Keywords: manual control, pilot-vehicle systems, flight simulation, motion cueing 1. INTRODUCTION Currently, the aviation industry relies heavily on flight simulators for research into aircraft operating procedures, flight control, cockpit interface design and aircraft handling qualities. In addition, pilots conduct a major part of their training of critical maneuvers in flight simulators. For all these applications of flight simulator devices, a certain level of fidelity of the simulation is required to justify the simulator’s use. The most notable flight simulator subsystems that are important for achieving adequate overall fidelity are flight instruments, control manipulators, an outof-the-window visual view, and the motion cues generated with the simulator motion system (Baarspul, 1986). Especially for the generation of flight simulator motion cues, still much controversy and uncertainty exists with respect to the required level of fidelity (Hosman et al., 2001). Two main factors are accepted to influence the level of motion fidelity: the quality of the simulator motion system hardware and the implementation and tuning of the simulator motion washout algorithm. The current industry standard for evaluating simulator motion fidelity (Anon., 2009) mainly consists of regulations on the quality of the motion system hardware, using metrics such as smoothness, latency and bandwidth of the simulator motion system. Despite the fact that recent additions to (Anon., 2009) also include quantitative requirements on the combined motion system and cueing dynamics, it is still common practice that an evaluation-pilot is enlisted to tune the washout algorithm coefficients to achieve motion cueing of adequate fidelity (Grant and Reid, 1997). ⋆ This research was supported under grant number 07058 by the Dutch Technology Foundation STW, the applied science division of The Netherlands Organization for Scientific Research (NWO), and the technology program of the Ministry of Economic Affairs.
As for instance described by Grant and Reid (1997), results of such a subjective tuning process for achieving adequate simulator motion fidelity are less deterministic than desired, as the success of subjective motion tuning is heavily dependent on the communication between the evaluation pilot and the simulator engineer. In addition, perhaps due to the fact that human motion perception is a largely subconscious process, subjective indications of the level of motion fidelity have been found to yield unexpected results at a number of occasions (Parrish and Martin Jr., 1976; Beukers et al., 2009). This paper proposes a method for objective optimization of the level of flight simulator motion fidelity. The cybernetic approach described here uses objective measurements of pilot control behavior, quantified using multimodal pilot models, to signal when pilots’ control strategy is affected by degraded simulator motion fidelity. In addition to explaining the method itself, this paper will also apply the cybernetic approach to data from a preliminary human-in-the-loop experiment performed in the SIMONA Research Simulator at Delft University of Technology. In this experiment, the extent to which the settings of the motion washout algorithm influence pilot control behavior in an aircraft roll-attitude control task was evaluated. 2. SIMULATOR MOTION FIDELITY Due to severe limitations on the motion capabilities of flight simulators, motion washout algorithms are required for attenuating and limiting the simulated aircraft motion. A large diversity in washout algorithms has been developed over the years (Schmidt and Conrad, 1970; Reid and Nahon, 1985; Telban et al., 2000). A systematic method for evaluating how choices in washout filter design affect the resulting simulator motion is, however, still not available.
flight simulator desired state, ft
turbulence, fd
pilot
remnant, n
tracking error, e
+ −
+ instruments
visual response, Hpv (jω)
washout + motion system
vestibular response, Hpm (jω)
control action, u
+ −
aircraft state, φ
aircraft model
Fig. 1. Closed-loop representation of skill-based aircraft control performed in a flight simulator environment. 2.1 Simulator Washout Filters
disturbance rejection, the pilot’s objective is essentially to minimize the tracking error e.
The most elementary strategy for washing out simulator motion, which is applied in most simulator washout algorithms is high-pass filtering of the aircraft linear and rotational accelerations. High-pass filters attenuate the low-frequency motions that tend to drive simulator motion systems to the limits of their workspaces, but leaves the high-frequency motion largely unaffected (Schmidt and Conrad, 1970). As a simple example of this strategy, this paper will investigate the effects of a first-order highpass washout filter in the rotational roll degree-of-freedom of the simulator on pilot control behavior. The transfer function of this washout filter is given by Eq. (1).
Fig. 1 further indicates how degraded simulator fidelity may affect the pilot-aircraft system. The inherent mechanical limitations of simulator visual and motion systems – for example, possible time delays associated with the generation of these cues – and the motion attenuation caused by the washout algorithm may significantly affect the fidelity of the the sensations provided by the simulator. In extreme cases, it is feared that this might cause pilots to learn to fly the simulator, but not the aircraft.
Hwφ (s) =
φw (s) s = Kφ φ(s) s + ωφ
(1)
The first-order washout filter given by Eq. (1) has two parameters: the filter gain Kφ and the filter break frequency ωφ . The former controls the scaling between simulator and aircraft motion, while the latter dictates the frequency below which motion is attenuated by the filter. The lower Kφ and the higher ωφ , the more the washed-out roll attitude φw will be attenuated compared to the true roll, φ. 2.2 Motion Cueing for Pilot Training It can be argued that high-fidelity motion cueing in flight simulators is not needed for all aspects of pilot training. For instance, during large portions of regular airliner operation, the aircraft is maneuvered relatively little, yielding almost no relevant perceivable aircraft motion. Still much controversy exists, however, with respect to the question of how a certain level of simulator motion fidelity affects the training of low-level manual flying skills and the development of pilots’ internal models of an aircraft’s inherent handling qualities (Hosman et al., 2001). A schematic representation of a skill-based manual aircraft control task, where the pilot essentially closes a loop around the controlled aircraft, is given in Fig. 1. Note from Fig. 1 that pilot control inputs u in such a control task are believed to result from individual responses to information perceived using the visual and vestibular systems, supplemented with uncorrelated control inputs accounted for by the remnant n (McRuer and Jex, 1967). Many skill-based vehicle control tasks can be represented as either the following of a desired reference state, the compensation for a disturbance (turbulence) signal, or a combination of both. In Fig. 1 these driving or forcing function signals are depicted using the symbols ft and fd , respectively. Note that for both target following and
2.3 Evaluating Simulator Motion Cueing Fidelity Evaluating simulator motion cueing fidelity is a difficult problem that the flight simulation community has struggled with for years. As indicated in Fig. 2, there are different levels at which the performance of a flight simulator could be compared to what is provided in a real aircraft (indicated with ∆s in Fig. 2). Perhaps the most workable definition of fidelity is physical fidelity, which is determined directly by the quality of the simulated aircraft dynamics and the simulator cueing systems. In fact, the few available flight simulator fidelity requirements (Anon., 2009) are largely stated in terms of physical fidelity. However, as clearly pointed out by Brown et al. (1989), human motion perception has its limitations, which can (and are) frequently exploited in simulator motion cueing. Therefore, perfect fidelity at the physical level may not be required for achieving a motion sensation that is indistinguishable from the aircraft. When evaluating the motion performance of a simulator it would therefore be more prudent to evaluate its perceptual fidelity. aircraft pilot aircraft dynamics
aircraft systems
simulated aircraft dynamics
simulator cueing systems
aircraft pilot perception
∆
aircraft pilot control
∆ simulator pilot perception
∆ simulator pilot control
flight simulator physical fidelity
perceptual fidelity
behavioral fidelity
Fig. 2. Evaluating flight simulator fidelity at physical, perceptual and behavioral levels.
Problems, however, arise when attempting to evaluate perceptual fidelity. Human motion perception is a result of a number of subconscious perception and integration processes that take place in the brain, which makes measuring perceptual fidelity problematic. The current industry standard for evaluating simulator motion fidelity is to judge it using subjective evaluation-pilot comments (Grant and Reid, 1997). Such subjective evaluation of simulator motion systems has, however, been found to yield surprising and inconclusive results at a number of different occasions (Parrish and Martin Jr., 1976; Beukers et al., 2009). In addition, without knowledge of how the perceived motion affects pilot behavior, it is impossible to weigh an observed discrepancy in perceptual motion fidelity against another. This paper therefore proposes a method for evaluating simulator motion at the behavioral fidelity level depicted in Fig. 2. The advantage of determining fidelity at the behavioral level is that pilot control behavior can be measured objectively, thereby allowing for objective assessment of simulator motion cueing performance. 3. THE CYBERNETIC APPROACH Obtaining meaningful objective measurements of pilot control for evaluating simulator fidelity at the behavioral level is by no means straightforward. The approach taken in this project relies heavily on the modeling of pilot control behavior using quasi-linear pilot models, as for instance proposed by McRuer and Jex (1967). As also illustrated by Fig. 1, McRuer and Jex noted the similarities between closed-loop automatic control systems and skillbased manual vehicle control. In addition, they described one of the key properties of human manual control: its adaptability to a myriad of factors. In order to yield optimal dynamical characteristics of the combined pilotvehicle system, human controllers are seen to adapt their control strategy to defining features of the control task. Low simulator motion fidelity, resulting in degraded motion feedback during manual control, is therefore highly likely to cause changes in pilot control strategy. This allows for optimization of simulator motion cueing using a cybernetic approach: by using mathematical identification techniques and behavioral pilot models, objective measurements of pilot control behavior can be used as the metric for evaluating the level of behavioral simulator motion fidelity. As illustrated in Fig. 3, the cybernetic approach proposed in this paper involves measuring pilot control behavior during real flight, and using these measurements as a baseline for evaluating behavioral discrepancies (“∆”) sensor dynamics
Hpv (jω)
1 multimodal pilot models & identification techniques
2
3 identify pilot model in
identify pilot model in real flight
the simulator
∆
4
5
improve the simulator motion cueing
framework for human-centered motion fidelity metric
Fig. 3. Framework for optimization of flight simulator motion cueing using a cybernetic approach. that occur in a flight simulator. These behavioral discrepancies are then used to improve simulator motion cueing and are expected to yield a framework for the definition of a human-centered motion fidelity metric. This model-based cybernetic approach requires models of pilot control behavior that incorporate the contributions of the different perceptual modalities. An example of such a multimodal pilot model, typical for manual control of aircraft roll attitude (Hosman, 1996), is depicted in Fig. 4. Note that the dominant visual and vestibular motion perception processes are included as separate channels. For each of these separate channels, the perceptual sensor dynamics, the adopted pilot equalization and relevant physical limitations of the human controller are included in the model. Note from Fig. 4 that the dynamics of the visual sensor (eyes) are assumed to be unity. The dynamics corresponding to the perception of angular motion using the semicircular canals of the vestibular system, Hsc (jω), are taken from previous research (Hosman, 1996). The main free parameters of the pilot model depicted in Fig. 4 are the parameters of the pilot equalization model (the visual gain Kv , visual lead constant TL and the motion gain Km ), and the visual and motion perception delays τv and τm . The dynamics of the neuromuscular actuation required for giving control inputs are described by the transfer function Hnm (jω), which is typically modeled as a second-order mass-spring-damper system with two additional parameters: the neuromuscular system frequency ωnm and the neuromuscular damping factor ζnm .
equalization
limitations
n +
e
Kv (1 + jωTL)
−
2 ωnm 2 2 (jω) + 2ζnmωnm + ωnm
φw
(jω)2 Hpm (jω)
1 + jω 0.11 (1 + jω 5.9)(1 + jω 0.005) |
{z
}
Km
u
+
e−jωτv
e−jωτm |
Hsc(jω)
Fig. 4. A typical multimodal pilot model for manual control of aircraft roll attitude.
{z
Hnm(jω)
}
Especially the values of the parameters associated with the pilot equalization (Kv , TL and Km ) reveal how pilots internally weigh visual and vestibular information to achieve a certain control action. Degraded motion fidelity is therefore expected to yield a shift in control strategy, which can then be measured objectively by evaluating changes in these multimodal pilot model parameters. 4. SIMULATOR EXPERIMENT This model-based approach to investigating changes in pilot control behavior has been successfully applied in many previous experiments (Zaal et al., 2009b; Pool et al., 2010). Here, the same method will be utilized to analyze data from a recent experiment performed in the SIMONA Research Simulator (SRS) at Delft University of Technology. The objective of this experiment was to evaluate the effects of roll motion washout on pilot behavior. 4.1 Control Task
disturbance, fd ωd Ad rad/s deg 0.383 0.672 0.844 0.508 1.764 0.253 2.838 0.129 3.912 0.078 5.446 0.048 7.747 0.030 10.508 0.022 13.116 0.018 17.334 0.015
φd rad -0.269 4.016 -0.806 4.938 5.442 2.274 1.636 2.973 3.429 3.486
target, ft ωt At rad/s deg 0.460 0.698 0.997 0.489 2.071 0.220 3.145 0.119 4.065 0.080 5.599 0.049 7.900 0.031 10.661 0.023 14.880 0.018 17.564 0.016
nt − 6 13 27 41 53 73 103 139 194 229
interest (0.1–20 rad/s) and the amplitude distributions Ad,t (k) of both forcing function signals were defined using the same low-pass filter described by Zaal et al. (2009b) to yield signals with reduced power at the higher frequencies. The sinusoid frequencies were defined to be integer multiples of the measurement time base frequency ωm = 2π/Tm : ωd,t (k) = nd,t (k)ωm . The numerical values of all multisine forcing function signal parameters defined in Eq. (3) are listed in Table 1. 4.3 Simulator Motion
Fig. 5. Compensatory visual display.
The controlled aircraft dynamics were the aileron-to-roll attitude dynamics of a Cessna Citation II business jet, linearized at an altitude of 10,000 ft and an airspeed of 160 kts with the yaw damper active. The transfer function of this controlled element is given by: 4.627(s + 2.038) s(s2 + 4.646s + 7.937)
(2)
Note that the system defined by Eq. (2) is approximately equal to K/s for frequencies below 2.8 rad/s, and to K/s2 above this frequency. 4.2 Forcing Functions The disturbance and target forcing functions signals fd and ft were constructed as quasi-random multisine signals according to:
The experiment was performed in the SRS at Delft University of Technology. The SRS has a hydraulic six degreeof-freedom hexapod motion system, which was used to supply participants with roll motion cues. Four different roll motion cueing conditions of different levels of motion fidelity were evaluated in the experiment, defined by a variation in motion filter gain Kφ and break frequency ωφ (see Eq. (1)). Fig. 6 depicts an evaluation of the well-known (physical) motion fidelity criterion proposed by Schroeder (1999) for the four different motion conditions, which are depicted by the square markers. Note from Fig. 6 that this criterion states that increased gain and phase attenuation caused by the motion filter result in a degradation in motion fidelity. The motion filter parameters corresponding to the different conditions are depicted between brackets alongside each marker in the following format: (Kφ ,ωφ ). Note from Fig. 6 that the four selected motion filter parameter sets range from high to medium/low fidelity according to the Schroeder criterion. Furthermore, note that the conditions labeled (0,∞) and (1,0) correspond to no simulator motion and 1-to-1 roll motion cues, respectively. 70
(0,∞)
60
low
50
fidelity medium
40
fidelity
30 (0.5,0.5)
20
(1,0.5) high
10
fidelity
0
X
0
Nd,t
fd,t (t) =
φt rad 1.288 6.089 5.507 1.734 2.019 0.441 5.175 3.415 1.066 3.479
e
An electrical sidestick was used to give control inputs u to the controlled dynamics. The sidestick was tuned to function as a force-stick, where ±36 N stick force was required for attaining maximum positive and negative control inputs to the controlled element.
Hφ,u (s) =
nd − 5 11 23 37 51 71 101 137 171 226
phase error at 1 rad/s
A combined target-following and disturbance-rejection task as depicted in Fig. 1 was performed in this experiment. Each experiment run lasted 90 seconds, of which the last 81.92 seconds were used as the measurement interval, Tm . The roll tracking error e, which participants were to minimize during the control task, was depicted on a compensatory visual display, see Fig. 5.
Table 1. Multisine forcing function properties.
Ad,t (k) sin [ωd,t (k)t + φd,t (k)]
(3)
k=1
Both signals were sums of 10 sinusoids. Sinusoid frequencies ωd,t (k) were distributed over the frequency range of
0.2
0.4
0.6
0.8
1 (1,0)
rotational gain at 1 rad/s
Fig. 6. Evaluation of the rotational simulator motion fidelity criterion proposed by Schroeder (1999) for the different experimental conditions.
4.4 Participants and Experimental Procedures Six subjects performed the roll-tracking task for the four experimental conditions depicted in Fig. 6. All participants were active Cessna Citation II pilots employed by the Faculty of Engineering and most had experience with similar control tasks from previous simulator experiments. Participants performed a significant number of training runs – typically 5-6 repetitions of each experimental condition – until their proficiency in performing the task had clearly reached an asymptote. Then five more repetitions of each experimental condition were collected as the measurement data. The different roll motion conditions were presented in random order (Latin square) throughout both the training and measurement phases of the experiment. Participants were instructed to minimize the roll-tracking error e presented on the visual display, by counteracting the rotation of the target line with respect to the aircraft symbol (see Fig. 5). After each run participants were informed of their tracking score, defined as the root mean square of the error signal e, to motivate them to constantly control at their maximum level of performance. 5. EXPERIMENT RESULTS To investigate changes in pilot control behavior due to the differences in roll motion cueing illustrated by Fig. 6, the multimodal pilot model depicted in Fig. 4 was fit to (a) Visual gain
(c) Motion gain
0.6
0.3 0.2 within-subject data mean data 0.0
1.0
0.5
motion gain Km , −
0.4
visual lead constant TL , s
visual gain Kv , −
Fig. 7(a) and (b) clearly show that the presence of roll motion cues affects pilots’ responses to visual information compared to the no-motion condition (0,∞). Higher values for the visual gain Kv and markedly reduced values for the visual lead constant TL are found for the conditions where roll motion was present. The reduction in visual lead results from the parallel vestibular perception of roll motion, that is the nonzero value of Km (Fig. 7(c)), which yields a more efficient (compare the values of τv and τm in Fig. 7(d) and (e)) source of pilot lead (Hosman, 1996). This more efficient lead generation allows for control with a higher gain (increased Kv ) and crossover frequency, which typically yields an increase in task performance. Highly similar effects of vestibular motion feedback are reported in many previous investigations (Zaal et al., 2009b; Pool et al., 2010).
(b) Visual lead constant
0.5
0.1
measured time traces of e, φw and u. The average recorded time traces for each subject were used to estimate the seven free pilot model parameters using a maximum likelihood estimation algorithm (Zaal et al., 2009a). Fig. 7(a) to (e) show the estimates obtained for the five most defining parameters of the pilot model: the visual gain Kv and lead constant TL , the motion gain Km and the visual and motion delays τv and τm , respectively. Note that the motion conditions (horizontal axis) are ordered from low to high motion fidelity and that parameter estimates for the individual subjects are depicted in addition to the means and 95% confidence intervals.
0.4 0.3 0.2 0.1 0.0
(0,∞) (0.5,0.5) (1,0.5)
(1,0)
0.8 0.6 0.4 0.2 0.0
(0,∞) (0.5,0.5) (1,0.5)
(1,0)
(0,∞) (0.5,0.5) (1,0.5)
motion condition (Kφ ,ωφ )
motion condition (Kφ ,ωφ ) (d) Visual delay
(e) Motion delay
(f ) Variance accounted for
0.4
0.35
(1,0)
motion condition (Kφ ,ωφ )
100
0.25 0.20 0.15
80
0.3
70
VAF, %
motion delay τm , s
visual delay τv , s
90 0.30
0.2
60 50 40 30
0.1
20
total visual motion
10 0.10
0.0 (0,∞) (0.5,0.5) (1,0.5)
(1,0)
motion condition (Kφ ,ωφ )
0 (0,∞) (0.5,0.5) (1,0.5)
(1,0)
motion condition (Kφ ,ωφ )
(0,∞) (0.5,0.5) (1,0.5)
(1,0)
motion condition (Kφ ,ωφ )
Fig. 7. Effects of roll motion cueing on pilot control behavior in a roll-attitude control task evaluated using a multimodal pilot model. Figs. (a) to (e) show changes in the most important pilot model parameters as a function of roll motion cueing setting (Kφ ,ωφ ). Fig. (f) shows the contributions of the visual and motion channels of the pilot model to the overall response, expressed in terms of model variance accounted for (VAF).
Furthermore, note that the differences in the level of simulator motion fidelity for the different experimental conditions also yield appreciable differences in pilot control behavior. Even though the motion gain Km and the visual delay τv are not found to change over the different motion conditions, the other pilot model parameters show clear trends with increasing motion fidelity. If roll motion is attenuated less, the pilot visual gain Kv is clearly increased, while less lead is generated visually, as indicated by the marked reduction in TL . More specifically, note that the measured values for these two parameters are highly similar for the (1,0) and (1,0.5) conditions, indicating behavioral motion fidelity is hardly affected when the filter break frequency ωφ is increased from 0 to 0.5 rad/s. When the motion is also scaled down for the (0.5,0.5) condition, Kv and TL are seen to move toward their values for the no-motion condition (0,∞), indicating a reduction in behavioral motion fidelity for the (0.5,0.5) condition compared to those where Kφ was unity. Finally, Fig. 7(f) shows the average pilot model variance accounted for (VAF) achieved for the different motion conditions. The VAF indicates the percentage of the measured model output (u) that is explained by the model (Zaal et al., 2009a). In addition to the total pilot model VAF (black markers), also the VAFs of only the visual and motion channels of the pilot model are indicated with the white and gray markers, respectively. Note from Fig. 7(f) that the total model VAF is more or less constant, at approximately 80%. In addition, note that the VAF of the different perceptual channels of the pilot model follows the same trends observed in Kv and TL for the different motion conditions. This is further indication that the contribution of roll motion feedback to pilot control is highly similar for the (1,0) and (1,0.5) conditions, but that pilots make significantly less use of motion feedback for the (0.5,0.5) condition, due to the reduction in motion fidelity. 6. CONCLUSIONS AND FUTURE WORK This paper proposed an objective method for evaluation of flight simulator motion fidelity based on measurements of multimodal pilot control behavior. This method relies on pilot models and mathematical identification techniques to quantify changes in pilot control behavior due to degraded motion fidelity. Using data from a recent simulator experiment, it was shown that degraded roll motion fidelity clearly affected pilot roll attitude control behavior. In addition, it was shown that no discrepancies in pilot control behavior occurred compared to the 1-to-1 motion case when roll motion was filtered by a unity-gain highpass filter with a break frequency of 0.5 rad/s. The findings described in this paper will be refined in future experiments by evaluating other washout filters and combinations of filter parameters. In addition, in-flight measurements of pilot control behavior will be collected to obtain a true baseline for simulator motion fidelity evaluations as described in this paper, allowing for optimization of behavioral flight simulator motion fidelity. REFERENCES Anon. (2009). ICAO 9625: Manual of Criteria for the Qualification of Flight Simulation Training Devices.
Volume 1 – Aeroplanes. ICAO. 3rd edition. Baarspul, M. (1986). Flight Simulation Techniques with Emphasis on the Generation of High Fidelity 6 DOF Motion Cues. Memorandum M-553, Delft University of Technology, Department of Aerospace Engineering. Beukers, J.T., Stroosma, O., Pool, D.M., Mulder, M., and Van Paassen, M.M. (2009). Investigation into Pilot Perception and Control During Decrab Maneuvers in Simulated Flight. In Proceedings of the AIAA Modeling and Simulation Technologies Conference and Exhibit, Chicago (IL), AIAA-2009-6030. Brown, Y.J., Cardullo, F.M., and Sinacori, J.B. (1989). Need-Based Evaluation of Simulator Force and Motion Cuing Devices. In AIAA Flight Simulation Technologies Conference and Exhibit, Boston (MA), Technical Papers (A89-48376), AIAA-1989-3272. Grant, P.R. and Reid, L.D. (1997). Motion Washout Filter Tuning: Rules and Requirements. Journal of Aircraft, 34(2), 145–151. Hosman, R.J.A.W. (1996). Pilot’s Perception and Control of Aircraft Motions. Ph.D. thesis, Delft University of Technology, Faculty of Aerospace Engineering. Hosman, R., Hamman, B., Lehman, C., Pelchat, Y., and Schroeder, J. (2001). Summary of the Panel Discussion on Motion Cueing Requirements. In Proceedings of the AIAA Modeling and Simulation Technologies Conference & Exhibit, Montreal, Canada, AIAA-2001-4253. McRuer, D.T. and Jex, H.R. (1967). A Review of QuasiLinear Pilot Models. IEEE Transactions on Human Factors in Electronics, HFE-8(3), 231–249. Parrish, R.V. and Martin Jr., D.J. (1976). Comparison of a Linear and a Nonlinear Washout For Motion Simulators Utilizing Objective and Subjective Data from CTOL Transport Landing Approaches. Technical Report NASA TN D-8157, Langley Research Center. Pool, D.M., Zaal, P.M.T., Mulder, M., and van Paassen, M.M. (2010). Effects of Heave Washout Settings in Aircraft Pitch Disturbance Rejection. Journal of Guidance, Control, and Dynamics, 33(1), 29–41. Reid, L.D. and Nahon, M.A. (1985). Flight Simulation Motion-Base Drive Algorithms: Part 1 – Developing and Testing the Equations. Technical Report UTIAS 296, Institute for Aerospace Studies, University of Toronto. Schmidt, S.F. and Conrad, B. (1970). Motion Drive Signals for Piloted Flight Simulators. Technical Report NASA CR-1601, NASA Ames Research Center. Schroeder, J.A. (1999). Helicopter Flight Simulation Motion Platform Requirements. Technical Report NASA TP-1999-208766, NASA Ames Research Center. Telban, R.J., Wu, W., and Cardullo, F.M. (2000). Motion Cueing Algorithm Development: Initial Investigation and Redesign of the Algorithms. Technical Report NASA CR-2000-209863, Dept. of MechanicaI Engineering, State University of New York, Binghamton (NY). Zaal, P.M.T., Pool, D.M., Chu, Q.P., Van Paassen, M.M., Mulder, M., and Mulder, J.A. (2009a). Modeling Human Multimodal Perception and Control Using Genetic Maximum Likelihood Estimation. Journal of Guidance, Control and Dynamics, 32(4), 1089–1099. Zaal, P.M.T., Pool, D.M., De Bruin, J., Mulder, M., and Van Paassen, M.M. (2009b). Use of Pitch and Heave Motion Cues in a Pitch Control Task. Journal of Guidance, Control and Dynamics, 32(2), 366–377.
Abstract: Due to a limited understanding of human multimodal motion perception during aircraft control, the definition of requirements for flight simulator motion fidelity is a problem the flight simulation community has struggled with for years. The development of adequate tuning procedures for motion washout algorithms has remained problematic for the same reason. This paper proposes a method for assessment of simulator motion fidelity that makes use of a cybernetic, model-based approach to measuring multimodal pilot control behavior. As illustrated here with data from a recent human-in-the-loop experiment, this approach allows for objective detection of changes in pilot control that result from degraded simulator motion fidelity. Keywords: manual control, pilot-vehicle systems, flight simulation, motion cueing 1. INTRODUCTION Currently, the aviation industry relies heavily on flight simulators for research into aircraft operating procedures, flight control, cockpit interface design and aircraft handling qualities. In addition, pilots conduct a major part of their training of critical maneuvers in flight simulators. For all these applications of flight simulator devices, a certain level of fidelity of the simulation is required to justify the simulator’s use. The most notable flight simulator subsystems that are important for achieving adequate overall fidelity are flight instruments, control manipulators, an outof-the-window visual view, and the motion cues generated with the simulator motion system (Baarspul, 1986). Especially for the generation of flight simulator motion cues, still much controversy and uncertainty exists with respect to the required level of fidelity (Hosman et al., 2001). Two main factors are accepted to influence the level of motion fidelity: the quality of the simulator motion system hardware and the implementation and tuning of the simulator motion washout algorithm. The current industry standard for evaluating simulator motion fidelity (Anon., 2009) mainly consists of regulations on the quality of the motion system hardware, using metrics such as smoothness, latency and bandwidth of the simulator motion system. Despite the fact that recent additions to (Anon., 2009) also include quantitative requirements on the combined motion system and cueing dynamics, it is still common practice that an evaluation-pilot is enlisted to tune the washout algorithm coefficients to achieve motion cueing of adequate fidelity (Grant and Reid, 1997). ⋆ This research was supported under grant number 07058 by the Dutch Technology Foundation STW, the applied science division of The Netherlands Organization for Scientific Research (NWO), and the technology program of the Ministry of Economic Affairs.
As for instance described by Grant and Reid (1997), results of such a subjective tuning process for achieving adequate simulator motion fidelity are less deterministic than desired, as the success of subjective motion tuning is heavily dependent on the communication between the evaluation pilot and the simulator engineer. In addition, perhaps due to the fact that human motion perception is a largely subconscious process, subjective indications of the level of motion fidelity have been found to yield unexpected results at a number of occasions (Parrish and Martin Jr., 1976; Beukers et al., 2009). This paper proposes a method for objective optimization of the level of flight simulator motion fidelity. The cybernetic approach described here uses objective measurements of pilot control behavior, quantified using multimodal pilot models, to signal when pilots’ control strategy is affected by degraded simulator motion fidelity. In addition to explaining the method itself, this paper will also apply the cybernetic approach to data from a preliminary human-in-the-loop experiment performed in the SIMONA Research Simulator at Delft University of Technology. In this experiment, the extent to which the settings of the motion washout algorithm influence pilot control behavior in an aircraft roll-attitude control task was evaluated. 2. SIMULATOR MOTION FIDELITY Due to severe limitations on the motion capabilities of flight simulators, motion washout algorithms are required for attenuating and limiting the simulated aircraft motion. A large diversity in washout algorithms has been developed over the years (Schmidt and Conrad, 1970; Reid and Nahon, 1985; Telban et al., 2000). A systematic method for evaluating how choices in washout filter design affect the resulting simulator motion is, however, still not available.
flight simulator desired state, ft
turbulence, fd
pilot
remnant, n
tracking error, e
+ −
+ instruments
visual response, Hpv (jω)
washout + motion system
vestibular response, Hpm (jω)
control action, u
+ −
aircraft state, φ
aircraft model
Fig. 1. Closed-loop representation of skill-based aircraft control performed in a flight simulator environment. 2.1 Simulator Washout Filters
disturbance rejection, the pilot’s objective is essentially to minimize the tracking error e.
The most elementary strategy for washing out simulator motion, which is applied in most simulator washout algorithms is high-pass filtering of the aircraft linear and rotational accelerations. High-pass filters attenuate the low-frequency motions that tend to drive simulator motion systems to the limits of their workspaces, but leaves the high-frequency motion largely unaffected (Schmidt and Conrad, 1970). As a simple example of this strategy, this paper will investigate the effects of a first-order highpass washout filter in the rotational roll degree-of-freedom of the simulator on pilot control behavior. The transfer function of this washout filter is given by Eq. (1).
Fig. 1 further indicates how degraded simulator fidelity may affect the pilot-aircraft system. The inherent mechanical limitations of simulator visual and motion systems – for example, possible time delays associated with the generation of these cues – and the motion attenuation caused by the washout algorithm may significantly affect the fidelity of the the sensations provided by the simulator. In extreme cases, it is feared that this might cause pilots to learn to fly the simulator, but not the aircraft.
Hwφ (s) =
φw (s) s = Kφ φ(s) s + ωφ
(1)
The first-order washout filter given by Eq. (1) has two parameters: the filter gain Kφ and the filter break frequency ωφ . The former controls the scaling between simulator and aircraft motion, while the latter dictates the frequency below which motion is attenuated by the filter. The lower Kφ and the higher ωφ , the more the washed-out roll attitude φw will be attenuated compared to the true roll, φ. 2.2 Motion Cueing for Pilot Training It can be argued that high-fidelity motion cueing in flight simulators is not needed for all aspects of pilot training. For instance, during large portions of regular airliner operation, the aircraft is maneuvered relatively little, yielding almost no relevant perceivable aircraft motion. Still much controversy exists, however, with respect to the question of how a certain level of simulator motion fidelity affects the training of low-level manual flying skills and the development of pilots’ internal models of an aircraft’s inherent handling qualities (Hosman et al., 2001). A schematic representation of a skill-based manual aircraft control task, where the pilot essentially closes a loop around the controlled aircraft, is given in Fig. 1. Note from Fig. 1 that pilot control inputs u in such a control task are believed to result from individual responses to information perceived using the visual and vestibular systems, supplemented with uncorrelated control inputs accounted for by the remnant n (McRuer and Jex, 1967). Many skill-based vehicle control tasks can be represented as either the following of a desired reference state, the compensation for a disturbance (turbulence) signal, or a combination of both. In Fig. 1 these driving or forcing function signals are depicted using the symbols ft and fd , respectively. Note that for both target following and
2.3 Evaluating Simulator Motion Cueing Fidelity Evaluating simulator motion cueing fidelity is a difficult problem that the flight simulation community has struggled with for years. As indicated in Fig. 2, there are different levels at which the performance of a flight simulator could be compared to what is provided in a real aircraft (indicated with ∆s in Fig. 2). Perhaps the most workable definition of fidelity is physical fidelity, which is determined directly by the quality of the simulated aircraft dynamics and the simulator cueing systems. In fact, the few available flight simulator fidelity requirements (Anon., 2009) are largely stated in terms of physical fidelity. However, as clearly pointed out by Brown et al. (1989), human motion perception has its limitations, which can (and are) frequently exploited in simulator motion cueing. Therefore, perfect fidelity at the physical level may not be required for achieving a motion sensation that is indistinguishable from the aircraft. When evaluating the motion performance of a simulator it would therefore be more prudent to evaluate its perceptual fidelity. aircraft pilot aircraft dynamics
aircraft systems
simulated aircraft dynamics
simulator cueing systems
aircraft pilot perception
∆
aircraft pilot control
∆ simulator pilot perception
∆ simulator pilot control
flight simulator physical fidelity
perceptual fidelity
behavioral fidelity
Fig. 2. Evaluating flight simulator fidelity at physical, perceptual and behavioral levels.
Problems, however, arise when attempting to evaluate perceptual fidelity. Human motion perception is a result of a number of subconscious perception and integration processes that take place in the brain, which makes measuring perceptual fidelity problematic. The current industry standard for evaluating simulator motion fidelity is to judge it using subjective evaluation-pilot comments (Grant and Reid, 1997). Such subjective evaluation of simulator motion systems has, however, been found to yield surprising and inconclusive results at a number of different occasions (Parrish and Martin Jr., 1976; Beukers et al., 2009). In addition, without knowledge of how the perceived motion affects pilot behavior, it is impossible to weigh an observed discrepancy in perceptual motion fidelity against another. This paper therefore proposes a method for evaluating simulator motion at the behavioral fidelity level depicted in Fig. 2. The advantage of determining fidelity at the behavioral level is that pilot control behavior can be measured objectively, thereby allowing for objective assessment of simulator motion cueing performance. 3. THE CYBERNETIC APPROACH Obtaining meaningful objective measurements of pilot control for evaluating simulator fidelity at the behavioral level is by no means straightforward. The approach taken in this project relies heavily on the modeling of pilot control behavior using quasi-linear pilot models, as for instance proposed by McRuer and Jex (1967). As also illustrated by Fig. 1, McRuer and Jex noted the similarities between closed-loop automatic control systems and skillbased manual vehicle control. In addition, they described one of the key properties of human manual control: its adaptability to a myriad of factors. In order to yield optimal dynamical characteristics of the combined pilotvehicle system, human controllers are seen to adapt their control strategy to defining features of the control task. Low simulator motion fidelity, resulting in degraded motion feedback during manual control, is therefore highly likely to cause changes in pilot control strategy. This allows for optimization of simulator motion cueing using a cybernetic approach: by using mathematical identification techniques and behavioral pilot models, objective measurements of pilot control behavior can be used as the metric for evaluating the level of behavioral simulator motion fidelity. As illustrated in Fig. 3, the cybernetic approach proposed in this paper involves measuring pilot control behavior during real flight, and using these measurements as a baseline for evaluating behavioral discrepancies (“∆”) sensor dynamics
Hpv (jω)
1 multimodal pilot models & identification techniques
2
3 identify pilot model in
identify pilot model in real flight
the simulator
∆
4
5
improve the simulator motion cueing
framework for human-centered motion fidelity metric
Fig. 3. Framework for optimization of flight simulator motion cueing using a cybernetic approach. that occur in a flight simulator. These behavioral discrepancies are then used to improve simulator motion cueing and are expected to yield a framework for the definition of a human-centered motion fidelity metric. This model-based cybernetic approach requires models of pilot control behavior that incorporate the contributions of the different perceptual modalities. An example of such a multimodal pilot model, typical for manual control of aircraft roll attitude (Hosman, 1996), is depicted in Fig. 4. Note that the dominant visual and vestibular motion perception processes are included as separate channels. For each of these separate channels, the perceptual sensor dynamics, the adopted pilot equalization and relevant physical limitations of the human controller are included in the model. Note from Fig. 4 that the dynamics of the visual sensor (eyes) are assumed to be unity. The dynamics corresponding to the perception of angular motion using the semicircular canals of the vestibular system, Hsc (jω), are taken from previous research (Hosman, 1996). The main free parameters of the pilot model depicted in Fig. 4 are the parameters of the pilot equalization model (the visual gain Kv , visual lead constant TL and the motion gain Km ), and the visual and motion perception delays τv and τm . The dynamics of the neuromuscular actuation required for giving control inputs are described by the transfer function Hnm (jω), which is typically modeled as a second-order mass-spring-damper system with two additional parameters: the neuromuscular system frequency ωnm and the neuromuscular damping factor ζnm .
equalization
limitations
n +
e
Kv (1 + jωTL)
−
2 ωnm 2 2 (jω) + 2ζnmωnm + ωnm
φw
(jω)2 Hpm (jω)
1 + jω 0.11 (1 + jω 5.9)(1 + jω 0.005) |
{z
}
Km
u
+
e−jωτv
e−jωτm |
Hsc(jω)
Fig. 4. A typical multimodal pilot model for manual control of aircraft roll attitude.
{z
Hnm(jω)
}
Especially the values of the parameters associated with the pilot equalization (Kv , TL and Km ) reveal how pilots internally weigh visual and vestibular information to achieve a certain control action. Degraded motion fidelity is therefore expected to yield a shift in control strategy, which can then be measured objectively by evaluating changes in these multimodal pilot model parameters. 4. SIMULATOR EXPERIMENT This model-based approach to investigating changes in pilot control behavior has been successfully applied in many previous experiments (Zaal et al., 2009b; Pool et al., 2010). Here, the same method will be utilized to analyze data from a recent experiment performed in the SIMONA Research Simulator (SRS) at Delft University of Technology. The objective of this experiment was to evaluate the effects of roll motion washout on pilot behavior. 4.1 Control Task
disturbance, fd ωd Ad rad/s deg 0.383 0.672 0.844 0.508 1.764 0.253 2.838 0.129 3.912 0.078 5.446 0.048 7.747 0.030 10.508 0.022 13.116 0.018 17.334 0.015
φd rad -0.269 4.016 -0.806 4.938 5.442 2.274 1.636 2.973 3.429 3.486
target, ft ωt At rad/s deg 0.460 0.698 0.997 0.489 2.071 0.220 3.145 0.119 4.065 0.080 5.599 0.049 7.900 0.031 10.661 0.023 14.880 0.018 17.564 0.016
nt − 6 13 27 41 53 73 103 139 194 229
interest (0.1–20 rad/s) and the amplitude distributions Ad,t (k) of both forcing function signals were defined using the same low-pass filter described by Zaal et al. (2009b) to yield signals with reduced power at the higher frequencies. The sinusoid frequencies were defined to be integer multiples of the measurement time base frequency ωm = 2π/Tm : ωd,t (k) = nd,t (k)ωm . The numerical values of all multisine forcing function signal parameters defined in Eq. (3) are listed in Table 1. 4.3 Simulator Motion
Fig. 5. Compensatory visual display.
The controlled aircraft dynamics were the aileron-to-roll attitude dynamics of a Cessna Citation II business jet, linearized at an altitude of 10,000 ft and an airspeed of 160 kts with the yaw damper active. The transfer function of this controlled element is given by: 4.627(s + 2.038) s(s2 + 4.646s + 7.937)
(2)
Note that the system defined by Eq. (2) is approximately equal to K/s for frequencies below 2.8 rad/s, and to K/s2 above this frequency. 4.2 Forcing Functions The disturbance and target forcing functions signals fd and ft were constructed as quasi-random multisine signals according to:
The experiment was performed in the SRS at Delft University of Technology. The SRS has a hydraulic six degreeof-freedom hexapod motion system, which was used to supply participants with roll motion cues. Four different roll motion cueing conditions of different levels of motion fidelity were evaluated in the experiment, defined by a variation in motion filter gain Kφ and break frequency ωφ (see Eq. (1)). Fig. 6 depicts an evaluation of the well-known (physical) motion fidelity criterion proposed by Schroeder (1999) for the four different motion conditions, which are depicted by the square markers. Note from Fig. 6 that this criterion states that increased gain and phase attenuation caused by the motion filter result in a degradation in motion fidelity. The motion filter parameters corresponding to the different conditions are depicted between brackets alongside each marker in the following format: (Kφ ,ωφ ). Note from Fig. 6 that the four selected motion filter parameter sets range from high to medium/low fidelity according to the Schroeder criterion. Furthermore, note that the conditions labeled (0,∞) and (1,0) correspond to no simulator motion and 1-to-1 roll motion cues, respectively. 70
(0,∞)
60
low
50
fidelity medium
40
fidelity
30 (0.5,0.5)
20
(1,0.5) high
10
fidelity
0
X
0
Nd,t
fd,t (t) =
φt rad 1.288 6.089 5.507 1.734 2.019 0.441 5.175 3.415 1.066 3.479
e
An electrical sidestick was used to give control inputs u to the controlled dynamics. The sidestick was tuned to function as a force-stick, where ±36 N stick force was required for attaining maximum positive and negative control inputs to the controlled element.
Hφ,u (s) =
nd − 5 11 23 37 51 71 101 137 171 226
phase error at 1 rad/s
A combined target-following and disturbance-rejection task as depicted in Fig. 1 was performed in this experiment. Each experiment run lasted 90 seconds, of which the last 81.92 seconds were used as the measurement interval, Tm . The roll tracking error e, which participants were to minimize during the control task, was depicted on a compensatory visual display, see Fig. 5.
Table 1. Multisine forcing function properties.
Ad,t (k) sin [ωd,t (k)t + φd,t (k)]
(3)
k=1
Both signals were sums of 10 sinusoids. Sinusoid frequencies ωd,t (k) were distributed over the frequency range of
0.2
0.4
0.6
0.8
1 (1,0)
rotational gain at 1 rad/s
Fig. 6. Evaluation of the rotational simulator motion fidelity criterion proposed by Schroeder (1999) for the different experimental conditions.
4.4 Participants and Experimental Procedures Six subjects performed the roll-tracking task for the four experimental conditions depicted in Fig. 6. All participants were active Cessna Citation II pilots employed by the Faculty of Engineering and most had experience with similar control tasks from previous simulator experiments. Participants performed a significant number of training runs – typically 5-6 repetitions of each experimental condition – until their proficiency in performing the task had clearly reached an asymptote. Then five more repetitions of each experimental condition were collected as the measurement data. The different roll motion conditions were presented in random order (Latin square) throughout both the training and measurement phases of the experiment. Participants were instructed to minimize the roll-tracking error e presented on the visual display, by counteracting the rotation of the target line with respect to the aircraft symbol (see Fig. 5). After each run participants were informed of their tracking score, defined as the root mean square of the error signal e, to motivate them to constantly control at their maximum level of performance. 5. EXPERIMENT RESULTS To investigate changes in pilot control behavior due to the differences in roll motion cueing illustrated by Fig. 6, the multimodal pilot model depicted in Fig. 4 was fit to (a) Visual gain
(c) Motion gain
0.6
0.3 0.2 within-subject data mean data 0.0
1.0
0.5
motion gain Km , −
0.4
visual lead constant TL , s
visual gain Kv , −
Fig. 7(a) and (b) clearly show that the presence of roll motion cues affects pilots’ responses to visual information compared to the no-motion condition (0,∞). Higher values for the visual gain Kv and markedly reduced values for the visual lead constant TL are found for the conditions where roll motion was present. The reduction in visual lead results from the parallel vestibular perception of roll motion, that is the nonzero value of Km (Fig. 7(c)), which yields a more efficient (compare the values of τv and τm in Fig. 7(d) and (e)) source of pilot lead (Hosman, 1996). This more efficient lead generation allows for control with a higher gain (increased Kv ) and crossover frequency, which typically yields an increase in task performance. Highly similar effects of vestibular motion feedback are reported in many previous investigations (Zaal et al., 2009b; Pool et al., 2010).
(b) Visual lead constant
0.5
0.1
measured time traces of e, φw and u. The average recorded time traces for each subject were used to estimate the seven free pilot model parameters using a maximum likelihood estimation algorithm (Zaal et al., 2009a). Fig. 7(a) to (e) show the estimates obtained for the five most defining parameters of the pilot model: the visual gain Kv and lead constant TL , the motion gain Km and the visual and motion delays τv and τm , respectively. Note that the motion conditions (horizontal axis) are ordered from low to high motion fidelity and that parameter estimates for the individual subjects are depicted in addition to the means and 95% confidence intervals.
0.4 0.3 0.2 0.1 0.0
(0,∞) (0.5,0.5) (1,0.5)
(1,0)
0.8 0.6 0.4 0.2 0.0
(0,∞) (0.5,0.5) (1,0.5)
(1,0)
(0,∞) (0.5,0.5) (1,0.5)
motion condition (Kφ ,ωφ )
motion condition (Kφ ,ωφ ) (d) Visual delay
(e) Motion delay
(f ) Variance accounted for
0.4
0.35
(1,0)
motion condition (Kφ ,ωφ )
100
0.25 0.20 0.15
80
0.3
70
VAF, %
motion delay τm , s
visual delay τv , s
90 0.30
0.2
60 50 40 30
0.1
20
total visual motion
10 0.10
0.0 (0,∞) (0.5,0.5) (1,0.5)
(1,0)
motion condition (Kφ ,ωφ )
0 (0,∞) (0.5,0.5) (1,0.5)
(1,0)
motion condition (Kφ ,ωφ )
(0,∞) (0.5,0.5) (1,0.5)
(1,0)
motion condition (Kφ ,ωφ )
Fig. 7. Effects of roll motion cueing on pilot control behavior in a roll-attitude control task evaluated using a multimodal pilot model. Figs. (a) to (e) show changes in the most important pilot model parameters as a function of roll motion cueing setting (Kφ ,ωφ ). Fig. (f) shows the contributions of the visual and motion channels of the pilot model to the overall response, expressed in terms of model variance accounted for (VAF).
Furthermore, note that the differences in the level of simulator motion fidelity for the different experimental conditions also yield appreciable differences in pilot control behavior. Even though the motion gain Km and the visual delay τv are not found to change over the different motion conditions, the other pilot model parameters show clear trends with increasing motion fidelity. If roll motion is attenuated less, the pilot visual gain Kv is clearly increased, while less lead is generated visually, as indicated by the marked reduction in TL . More specifically, note that the measured values for these two parameters are highly similar for the (1,0) and (1,0.5) conditions, indicating behavioral motion fidelity is hardly affected when the filter break frequency ωφ is increased from 0 to 0.5 rad/s. When the motion is also scaled down for the (0.5,0.5) condition, Kv and TL are seen to move toward their values for the no-motion condition (0,∞), indicating a reduction in behavioral motion fidelity for the (0.5,0.5) condition compared to those where Kφ was unity. Finally, Fig. 7(f) shows the average pilot model variance accounted for (VAF) achieved for the different motion conditions. The VAF indicates the percentage of the measured model output (u) that is explained by the model (Zaal et al., 2009a). In addition to the total pilot model VAF (black markers), also the VAFs of only the visual and motion channels of the pilot model are indicated with the white and gray markers, respectively. Note from Fig. 7(f) that the total model VAF is more or less constant, at approximately 80%. In addition, note that the VAF of the different perceptual channels of the pilot model follows the same trends observed in Kv and TL for the different motion conditions. This is further indication that the contribution of roll motion feedback to pilot control is highly similar for the (1,0) and (1,0.5) conditions, but that pilots make significantly less use of motion feedback for the (0.5,0.5) condition, due to the reduction in motion fidelity. 6. CONCLUSIONS AND FUTURE WORK This paper proposed an objective method for evaluation of flight simulator motion fidelity based on measurements of multimodal pilot control behavior. This method relies on pilot models and mathematical identification techniques to quantify changes in pilot control behavior due to degraded motion fidelity. Using data from a recent simulator experiment, it was shown that degraded roll motion fidelity clearly affected pilot roll attitude control behavior. In addition, it was shown that no discrepancies in pilot control behavior occurred compared to the 1-to-1 motion case when roll motion was filtered by a unity-gain highpass filter with a break frequency of 0.5 rad/s. The findings described in this paper will be refined in future experiments by evaluating other washout filters and combinations of filter parameters. In addition, in-flight measurements of pilot control behavior will be collected to obtain a true baseline for simulator motion fidelity evaluations as described in this paper, allowing for optimization of behavioral flight simulator motion fidelity. REFERENCES Anon. (2009). ICAO 9625: Manual of Criteria for the Qualification of Flight Simulation Training Devices.
Volume 1 – Aeroplanes. ICAO. 3rd edition. Baarspul, M. (1986). Flight Simulation Techniques with Emphasis on the Generation of High Fidelity 6 DOF Motion Cues. Memorandum M-553, Delft University of Technology, Department of Aerospace Engineering. Beukers, J.T., Stroosma, O., Pool, D.M., Mulder, M., and Van Paassen, M.M. (2009). Investigation into Pilot Perception and Control During Decrab Maneuvers in Simulated Flight. In Proceedings of the AIAA Modeling and Simulation Technologies Conference and Exhibit, Chicago (IL), AIAA-2009-6030. Brown, Y.J., Cardullo, F.M., and Sinacori, J.B. (1989). Need-Based Evaluation of Simulator Force and Motion Cuing Devices. In AIAA Flight Simulation Technologies Conference and Exhibit, Boston (MA), Technical Papers (A89-48376), AIAA-1989-3272. Grant, P.R. and Reid, L.D. (1997). Motion Washout Filter Tuning: Rules and Requirements. Journal of Aircraft, 34(2), 145–151. Hosman, R.J.A.W. (1996). Pilot’s Perception and Control of Aircraft Motions. Ph.D. thesis, Delft University of Technology, Faculty of Aerospace Engineering. Hosman, R., Hamman, B., Lehman, C., Pelchat, Y., and Schroeder, J. (2001). Summary of the Panel Discussion on Motion Cueing Requirements. In Proceedings of the AIAA Modeling and Simulation Technologies Conference & Exhibit, Montreal, Canada, AIAA-2001-4253. McRuer, D.T. and Jex, H.R. (1967). A Review of QuasiLinear Pilot Models. IEEE Transactions on Human Factors in Electronics, HFE-8(3), 231–249. Parrish, R.V. and Martin Jr., D.J. (1976). Comparison of a Linear and a Nonlinear Washout For Motion Simulators Utilizing Objective and Subjective Data from CTOL Transport Landing Approaches. Technical Report NASA TN D-8157, Langley Research Center. Pool, D.M., Zaal, P.M.T., Mulder, M., and van Paassen, M.M. (2010). Effects of Heave Washout Settings in Aircraft Pitch Disturbance Rejection. Journal of Guidance, Control, and Dynamics, 33(1), 29–41. Reid, L.D. and Nahon, M.A. (1985). Flight Simulation Motion-Base Drive Algorithms: Part 1 – Developing and Testing the Equations. Technical Report UTIAS 296, Institute for Aerospace Studies, University of Toronto. Schmidt, S.F. and Conrad, B. (1970). Motion Drive Signals for Piloted Flight Simulators. Technical Report NASA CR-1601, NASA Ames Research Center. Schroeder, J.A. (1999). Helicopter Flight Simulation Motion Platform Requirements. Technical Report NASA TP-1999-208766, NASA Ames Research Center. Telban, R.J., Wu, W., and Cardullo, F.M. (2000). Motion Cueing Algorithm Development: Initial Investigation and Redesign of the Algorithms. Technical Report NASA CR-2000-209863, Dept. of MechanicaI Engineering, State University of New York, Binghamton (NY). Zaal, P.M.T., Pool, D.M., Chu, Q.P., Van Paassen, M.M., Mulder, M., and Mulder, J.A. (2009a). Modeling Human Multimodal Perception and Control Using Genetic Maximum Likelihood Estimation. Journal of Guidance, Control and Dynamics, 32(4), 1089–1099. Zaal, P.M.T., Pool, D.M., De Bruin, J., Mulder, M., and Van Paassen, M.M. (2009b). Use of Pitch and Heave Motion Cues in a Pitch Control Task. Journal of Guidance, Control and Dynamics, 32(2), 366–377.