Pmt Zaal

Modeling Human Multimodal Perception and Control Using Genetic Maximum Likelihood Estimation

This paper presents a new method for estimating the parameters of multi-channel pilot models that is based on maximum likelihood estimation. To cope with the inherent nonlinearity of this optimization problem, the gradient-based Gauss-Newton algorithm commonly used to optimize the likelihood function in terms of output error is complemented with a genetic algorithm. This significantly increases the probability of finding the global optimum of the optimization problem. The genetic maximum likelihood method is successfully applied to data from a recent human-in-the-loop experiment. Accurate estimates of the pilot model parameters and the remnant characteristics were obtained. Multiple simulations with increasing levels of pilot remnant were performed, using the set of parameters found from the experimental data, to investigate how the accuracy of the parameter estimate is affected by increasing remnant. It is shown that only for very high levels of pilot remnant the bias in the parameter estimates is substantial. Some adjustments to the maximum likelihood method are proposed to reduce this bias.

Use of Pitch and Heave Motion Cues in a Pitch Control Task

During pitch rotation of the aircraft, a pilot, seated in front of the aircraft center of gravity, is subjected to rotational pitch and vertical heave motion. The heave motion is a combination of the vertical motion of the aircraft center of gravity and the heave motion as a result of the pitch rotation. In a pitch tracking task, all of these cues could potentially have a positive effect on performance and control behavior, as they are all related to the aircraft pitch attitude. To improve the tuning of flight simulator motion filters, a better understanding of how these motion components are used by the pilot is required. First, the optimal use of the different motion components was evaluated using an optimal control analysis. Next, an aircraft pitch attitude control experiment was performed in the SIMONA Research Simulator, investigating the effects of pitch rotation, pitch heave, and center of gravity heave on pilot control behavior. Pilot performance significantly improved with pitch motion, with an increased crossover frequency for the disturbance open loop. The increase in performance was a result of an increased visual gain and a reduction in visual lead, allowed for by the addition of pitch motion. Pitch heave motion showed similar but smaller effects. The center of gravity heave motion, although taking up most of the simulator motion space, was found to have no significant effects on performance and control behavior.

Effects of Heave Washout Settings in Aircraft Pitch Disturbance Rejection

In most moving-base flight simulators, the simulated aircraft motion needs to be filtered with motion washout filters to keep the simulator within its limited motion envelope. Translational motion in particular requires filtering, as the low frequency components of the vehicle motion tend to quickly drive simulators toward their motion bounds. Commonly, linear washout filters are therefore used to attenuate the simulated motion in magnitude and in phase. It is found in many studies that the settings of these washout filters affect pilot performance and control behavior. In most of these studies, no comparison to a case with one-to-one motion cues is performed, as a result of the limited motion envelope of the used simulators. In the current study, an experiment was performed in the SIMONA Research Simulator at Delft University of Technology to investigate the effects of heave washout settings on pilot performance and control behavior in a pitch attitude control task. In addition to rotational pitch motion, heave accelerations at the pilot station that result directly from aircraft pitch were evaluated. This heave motion component could be supplied one-to-one in the simulator due to the modest size of the aircraft model, a Cessna Citation I business jet. The experiment revealed that pilot performance and control activity both increased significantly with increasing heave motion fidelity. An analysis of pilot control behavior using pilot models indicated that the enhanced performance was caused by an increase in the magnitude with which pilots responded to visual and physical motion stimuli and a decrease in the amount of visual lead that was generated by the pilots.

Multimodal Pilot Control Behavior in Combined Target-Following Disturbance-Rejection Tasks

Investigating how humans use their perceptual modalities while controlling a vehicle is important for the design of new control systems and the optimization of simulator motion cueing. For the identification of separate pilot response functions to the different perceived cues, multiple forcing functions need to be inserted into the manual control loop. An example of a task with multiple forcing functions is a combined target-following disturbance-rejection task, where a target and disturbance signal are used to separate the human visual and vestibular motion responses. The use of multiple forcing functions, however, also affects the nature of the control task and how the motion cues are used by the pilot to form a proper control action. This paper presents the results of an experiment where possible effects of using multiple forcing functions on pilot control behavior in an aircraft pitch control task are investigated. The results indicate that pilot performance and control activity are significantly lower when the relative power of the target forcing function is increased. This is caused by a significant change in multimodal pilot control behavior. With an increase in relative target power, the visual-perception gain is reduced and the visual time delay becomes higher. The motion-perception gain reduces if both forcing functions have significant power. It is also found that multimodal pilot control behavior in a pure target or disturbance task can be analyzed by adding a small additional disturbance or target signal, respectively. In this case, the effects on control behavior are found to be minimal, while still being able to accurately estimate the parameters of the multichannel pilot model.

Pilot equalization in manual control of aircraft dynamics

In continuous manual control tasks, pilots adapt their control strategy to the dynamics of the controlled element to yield adequate performance of the combined pilot-vehicle system. For a controlled element representing the linearized pitch dynamics of a small jet aircraft, the pilot models described in literature were found to lack the required freedom in the pilot equalization term to accurately model the adopted pilot compensation. An additional lead term in the pilot equalization transfer function was found to significantly increase the accuracy in modeling manual control behavior of aircraft pitch dynamics.

Modeling Wide-Frequency-Range Pilot Equalization for Control of Aircraft Pitch Dynamics

In continuous manual control tasks, human controllers adapt their control strategy to the dynamics of the controlled element. This compensation for the controlled-element dynamics is performed around the pilot–vehicle system crossover frequency, in order to obtain satisfactory performance of the combined pilot–vehicle system, but is also seen to extend to frequencies well above crossover. For a controlled element representing the linearized pitch dynamics of a small conventional jet aircraft, an extension to the models for pilot equalization described in the literature was found to be needed for the modeling of the adopted pilot equalization dynamics over a wide frequency range. Measured pilot describing functions revealed that pilots use a combination of low-frequency lag and high-frequency lead equalization to compensate for the characteristics of these typical aircraft pitch dynamics around the short-period mode. An additional high-frequency lead term in the pilot equalization transfer function was found to allow for the modeling of these adopted equalization dynamics over a wide frequency range, thereby also yielding a significant increase in the percentage of measured control inputs that is explained by the pilot model. Furthermore, for this controlled element the extended model for the equalization dynamics was found to be important for the interpretation of the changes in pilot control behavior that occur due to the presence of physical motion feedback.

Conducting multi-modal pilot model identification - Results of a simulator experiment

*† ‡ § As part of the research project “A Cybernetic approach to assess simulator fidelity” an experiment was conducted in the SIMONA Research Simulator. The experiment is performed in preparation of an in-flight experiment with the objective to perform multimodal pilot model identification during real flight in the Citation II laboratory aircraft of the Delft University of Technology. To capture the pilot’s behavior in a mathematical model, the pilot is asked to perform a well defined pitch attitude control task. The multi-modal identification of the pilot requires the pilot to perform a simultaneous target tracking and disturbance rejection task. Two forcing functions are therefore inserted into the closed-loop system. One of the primary goals of the experiment is to gain more insight in the design requirements of the disturbance and tracking signals. The simulator experiment is also used investigate the effect of turbulence on the accuracy of the identification results. The final goal of the simulator experiment was to investigate the possibility of modifying the aircraft dynamics by using a stick shaping filter. The experiment revealed that the power that can be inserted by the forcing functions is limited by the torque limiters of the autopilot. Turbulence did not affect the identification results significantly. The pilots showed crossover regression for the conditions with the stick-shaping filter. This is probably caused by the high bandwidth of the forcing functions used in combination with double integrator dynamics.

Design and Certification of a Fly-By-Wire System with Minimal Impact on the Original Flight Controls

The Faculty of Aerospace Engineering of the Delft University of Technology operates a Cessna Citation II laboratory aircraft. The aircraft is equipped with an advanced flight test instrumentation system and is used for research and student educational flights. The aircraft has a conventional reversible control system, combined with an analog electric automatic control system. To accommodate new research projects a novel fly-by-wire control system is developed. The fly-by-wire system is based on the original automatic control system of the aircraft, reducing the impact on the original flight controls, minimizing aircraft modifications, and inheriting the safety features of the original control system. As a result, the complexity of the certification process and costs could be significantly reduced. The new system will be used for research into novel flight control algorithms, optimal input signals for aircraft model identification, the identification of multimodal pilot control behavior, and novel human-machine interfaces.

Preparation for conducting multi-modal pilot model identification in real flight

*† ‡ § Currently, flight simulator requirements are often stated in terms of hardware, properties of the visual and motion systems like time delay and bandwidth. We postulate that simulator fidelity can be defined in terms of human behavior. Then, for any specific task, a high-fidelity simulator would induce the same behavior as in the real world, whereas a low-fidelity simulator would induce behavior that is significantly different. An essential step in assessing behavioral fidelity is to have a quantitative model of pilot behavior in flight. The current lack of baseline models is perhaps the main obstruction in fidelity research. Although previous work was already successful in measuring behavior in real flight, it only measured a lumped single-input single-output pilot model, prohibiting the proper investigation of pilot visual-vestibular perception. It is the goal of this project to identify a multimodal pilot model in real flight. The comparison of a pilot model obtained in real flight with a pilot model obtained in a simulator is expected to be difficult. Pilots can and will adapt to any difference between the task as conducted in real flight as in the simulator. It is therefore of the utmost importance to obtain a very accurate model of the aircraft and all the components in the closed loop pilot/aircraft system. This paper describes our efforts to obtain such a model and investigate the consequences of adding non-linear components to the closed loop pilot/aircraft system.

Maximum likelihood estimation of multi-modal pilot control behavior in a target-following task

Modeling multi-modal perception and control of pilots can be an important tool when designing new manual control systems or assessing simulator fidelity. This paper introduces a new pilot model parameter estimation technique based on maximum likelihood estimation. The method provides new possibilities for analyzing multi-modal control behavior in a target-following task. The new method is applied to data from an experiment comparing pilot control behavior in real flight and in a flight simulator. Previously, the data from this experiment could only be used to estimate a lumped, single modality, pilot model. The results from the maximum likelihood estimation technique indeed give new insights into the pilots use of different modalities in the aircraft and in the simulator.