An Investigation of Vibration Feedthrough and Feedthrough Cancellation in Joystick Controlled Vehicles

The inertial forces acting on operators of joystick controlled machinery in moving vehicles can produce unwanted control signals through the joystick. These forces tend to deteriorate continuous tracking performance and further, when the machinery in control is the vehicle itself, they may lead to unstable oscillations that jeopardize that vehicle’s safe operation. In this paper, we propose the use of a force-reflecting joystick and a model-based controller to cancel the effects of inertia forces. Using a physical model of human biomechanics, we experimentally investigate the effectiveness of a cancellation controller in stabilizing a driving task. A second experiment involving a human subject on board a motion base investigates the ability of the cancellation controller to improve performance in a continuous tracking task. Results indicate that the cancellation controller enhances stability and improves tracking. INTRODUCTION Human operators onboard moving vehicles are subjected to inertial forces due to vehicle accelerations and these forces, when they are coupled through the operator’s body into the control interface, induce unwanted control signals that degrade tracking performance. The phenomenon has been called vibration ∗R. Brent Gillespie (corresponding author) is with the Department of Mechanical Engineering, University of Michigan, Ann Arbor, MI 48109, USA. (e-mail: [email protected]) feedthrough or biodynamic feedthrough. If the controlled plant is the vehicle itself (in which case the operator may be considered a pilot or driver), then a closed loop exists that involves the biomechanics of the operator’s body. These closed loop dynamics may give rise to unstable vehicle oscillations that jeopardize safe operation of the vehicle. Examples of vehicles whose piloting may suffer from vibration feedthrough include tanks, electric wheelchairs, frontloaders, fighter jets and helicopters [1]. But even when the controlled plant is a remote vehicle or on-baord machine, where no closed loop involving the human biomechanics exists, tracking performance will be degraded by vibration feedthrough. The present study focuses on vibration feedthrough, and it does not include pilot induced oscillations. The first one is caused by inertial effects, typically in the frequency range between 1Hz and 5Hz. The second one is caused by the instability of the system comprised by the vehicle and the volitional control activity of the pilot on frequencies below 1Hz. To eliminate the effects of vibration feedthrough, the use of a motorized, force-reflecting joystick has been proposed [2]. The aim is to cancel the non-volitional force caused by inertial effects with a torque injected by a DC motor on the force-reflecting joystick. This is expected to improve tracking performance in both the open loop (remote control) and closed loop (driving) cases and to improve stability in the closed loop (driving) case. Force feedback in manual control interfaces has been shown to improve human/machine performance in various tasks. The 1 Copyright c © 2003 by ASME classic example of the benefits of force-feedback is the bilateral telemanipulator, wherein force feedback in a local “master” manipulator carries information about the interactions taking place between the remote “slave” manipulator and its environment. But force-feedback has also proven beneficial in the performance of vehicle control tasks. Force-reflecting devices improve the information content of manually controlled vehicles within virtual environments according to Repperger and Chandler [3], enabling improved operation of the vehicle and improved performance in tracking tasks. Yuhara et al. [4] used a structural driver-vehicle model to design a force feedback steering wheel. The added kinesthetic information improves vehicle handling, improves lane following both in compensatory and in pursuit control, and reduces the mental and physical load of the driver. A force feedback joystick was used to give information about the motion of an unmanned air vehicle (UAV) to its pilot operating from a remote site in the work presented by Korteling and Borg [5]. The force-feedback system improved the performance of the pilot and reduced the mental workload associated with maneuvering a simulated UAV. Parker et al. [6] built a force feedback system for heavy duty hydraulic machines. The system gives a feel to the human operator for the load acting on the tip of the boom of an excavator. The benefits of force-feedback in these examples, however, accrue because of the improved information about the controlled element’s behavior available to the human operator. We are interested in benefits to be reaped by motorizing the interface device that do not involve cognitive control or volition on the part of the human operator. Vibration feedthrough can occur (and, with a motorized interface can be compensated) without any participation of cognitive processing. Vibration feedthrough has been identified as a cause of deteriorated human/machine performance and investigated in various scenarios. A comprehensive overview of biodynamic effects on continuous tracking performance is available in Griffin [7]. The dynamics of both motion-type and force-type joystick interfaces and the associated human-machine system was analyzed by Hess [8], [9]. A structural pilot-aircraft model was constructed to analyze the roll-ratchet phenomenon. This includes a simple biodynamic feedthrough model, a continuous tracking model, a model for manipulator-feel system dynamics and a model for vestibular motion feedback. The resulting Bode magnitude plots of the pilot-vehicle transfer functions follow trends similar to those of experimental Bode plots. The analysis and simulation of vibration feedthrough and feedthrough cancellation through signal processing of a joystick controlled aircraft is presented by Verger et al. [10]. The inertial effects acting on the pilot are estimated by an adaptive filter and they are subtracted from the control signal. The results of an experimental study were published in another paper by Verger et al. [11]. A joystick controlled motion platform was used for demonstrating the solution. The need to predict the continuous tracking performance of pilots of vehicles and machine operators was first identified during the second World War [12]. Since that time, ever more accurate models of human tracking performance have been sought by the military and by industry [13]. A comprehensive summary of such models is given by Reid [14] and by McRuer [15]. The most frequently used types of models are the structural model and the optimal control (also known as algorithmic) model. Alternative modeling approaches also exist, e.g. the fuzzy control model [12]. The structural model evolved from McRuer’s crossover model, and methods for measuring or estimating its parameters have been documented in numerous articles, e.g. HessModel. McRuer et al. [16] describes new experimental results in the context of this model. The human operator is modeled as a linear, time invariant system in most of the studies. The algorithmic model uses LQR techniques and Kalman filters, and its implementations both in simulation and hardware are described in great detail in the literature [17]. The tracking performance of human operators has been investigated for a variety of scenarios, including tracking tasks carried out in more than one dimension at a time [18]. McRuer and Schmidt [19] investigated the behavior of pilots when carrying out a secondary task in addition to tracking. To eliminate vibrations induced by inertial effects, adaptive filtering of the control signal was implemented by Verger et al. [10], [11]. In this work, however, the cancellation was effected by injecting a cancelling signal to the joystick output, rather than imposing a cancelling torque on the joystick. Thus the feel of the joystick to the operator was not affected directly. Also, this signal processing solution cuts off the high frequency components of the control signal above 1Hz, which somewhat deteriorates the performance of the vehicle. An acceleration feedforward control approach that imposed a cancelling torque on the joystick using a force-reflecting joystick was proposed by Gillespie et al. [2] A robust controller was implemented using force-feedback by Sirouspour and Salcudean, [20]. An alternative approach involving increased joystick damping with decreased loop gain was proposed by Arai et al. [21]. Venneri and Noor [22] predict the appearance of EMG signal detectors mounted on the pilot’s hand for obtaining vehicle control signals in the future. This solution is expected to eliminate biodynamic feedthrough, however, as pointed out by Xu and Hollerbach [23], voluntary and nonvoluntary movements can not be separated based on EMG using current technology. In this paper, the use of a force reflecting joystick and a model based controller is further investigated as a means of solving the vibration feedthrough problem. The transfer function of the human operator from vehicle acceleration to nonvolitional moment imposed on the joystick will be determined based on human subject tests. The proposed controller will identify this transfer function and will impose a moment on the joystick in the opposite sense as a function of measured vehicle accelerations. This approach is expected to improve tracking performance and improve safety. 2 Copyright c © 2003 by ASME The body of this paper is organized in three parts. First, the principles of modelling the human/machine dynamics are explained. The next section presents how vibration feedthrough was reproduced experimentally and in simulation for the closed loop case in which the human biomechanics were modelled using a stand-in physical second order system. Also for this case, a vibration feedthrough cancellation controller was tested using a force reflecting joystick and the stand-in model for the biomechanics. At last, human subject tests of tracking performance carried out on a single axis motion platform are discussed. The degradation of tracking performance in moving vehicles and the effectiveness of a simple cancellation controller were demonstrated by these tests. This is a groundwork for the investigation of vibration feedthrough in the driving task, the results of which are expected to lead to the design of a feedthrough cancellation controller. MODELING THE HUMAN/MACHINE COUPLED DYNAMICS We are interested in developing a haptic or force-reflecting interface for the purpose of suppressing biodynamic feedthrough and thereby imp roving human/machine performance on-board moving vehicles. We have chosen to investigate biodynamic feedthrough and its cancellation in the context of human tracking performance. We distinguish between two types of tracking task. The first we call remote control, which involves tracking, from on-board a moving host vehicle, of a moving target with a piece of machinery whose motion is independent of the motion of the host vehicle. The motion (or vibration) of the host vehicle can be considered a disturbance to the task. The second type of tracking task, which we call driving, involves tracking of a moving target with the vehicle itself. In the driving task, the motion of the machinery used for tracking is intimately coupled to the motion which acts as disturbance to the tracking task: they are one and the same. We consider a ’motion stick’, a joystick with a pivot that rotates around one axis without dead-zone and without backlash, giving an electrical signal as a function of angular displacement. By contrast, a ’force stick’ is a rigid joystick providing electrical signals as a function of the force imposed on it. Our joystick can be modelled as a second order transfer function from the force applied on it to angular displacement which involves its moment of inertia and a virtual return spring and a virtual damper. The joystick angle is multiplied by a scalar to produce the output of the virtual plant to be controlled. The controlled plant has no dynamics, thus our tracking task is zeroth order. Fig.1 shows the block diagram of a general system operating in remote control mode. The human operator (HO) is characterized by the double input, single output transfer function H(s), with the reference signal of the tracking task Xr and vehicle acceleration Ẍv as inputs and a torque T as output. The quantity the HO intends to control, by acting through the joystick is the output of the plant, that is, the plant position Xp. The torque acting on the joystick T , has two components, one of them we call the nonvolitional torque Tnv which is the output of the transfer function Hnv(s) describing the nonvolitional effects of the vehicle acceleration Ẍv on the torque T . This is a consequence of the biodynamic (primarily inertia) forces acting on the pilot in the moving vehicle. The other component we call the volitional torque Tv, which is the output of the transfer function describing the action of the volitional controller Hv(s) (involving perception, cognition and muscle action) on an error signal or other combination of Xr and Xp . Finally, a gain Cp multiplies the joystick angle to produce a command Xc of the plant P(s).