217 views

Uploaded by joeni-42

save

- Optimal State Space Control
- Motorcycle Dynamics study
- Motorcycle Dynamics
- Motorcycle Modeling and Control
- Lecture2 - System Modeling, Standard Form, Laplace Transform.pdf
- Design of a Tuned Mass Damper for Seismic Excited Building
- 25
- Motorcycle Dynamics Library in Modelica
- Comparison of Pole Placement and LQR Applied.pdf
- Motorcycle Dynamics
- [13]
- Bogdan_1Thesis_1
- Assignment 1
- ee580_notes.pdf
- 04213169
- A51 1 Vrdoljak10 Applying Optimal Sliding Mode Based Load Frequency Control in Power Systems With Controllable Hydro Power Plants
- 10.1049@pe19980507
- Download
- Session 11
- Comandos de MatLab.docx
- Design of Decentralised Load Frequency
- 01359955
- Chaper 3_c b Bangal
- Category_Control theory - Wikipedia, the free encyclopedia.pdf
- 6SL3244 0BB00 1PA1 Datasheet En
- Load frequency analysis and its schematic control
- Zh 05110257261
- Wip
- signal and system Lecture 20
- A Process Control Experimen
- The Unwinding: An Inner History of the New America
- Yes Please
- Sapiens: A Brief History of Humankind
- The Innovators: How a Group of Hackers, Geniuses, and Geeks Created the Digital Revolution
- Dispatches from Pluto: Lost and Found in the Mississippi Delta
- Elon Musk: Tesla, SpaceX, and the Quest for a Fantastic Future
- Devil in the Grove: Thurgood Marshall, the Groveland Boys, and the Dawn of a New America
- John Adams
- The Prize: The Epic Quest for Oil, Money & Power
- The Emperor of All Maladies: A Biography of Cancer
- A Heartbreaking Work Of Staggering Genius: A Memoir Based on a True Story
- This Changes Everything: Capitalism vs. The Climate
- Grand Pursuit: The Story of Economic Genius
- The New Confessions of an Economic Hit Man
- Team of Rivals: The Political Genius of Abraham Lincoln
- Smart People Should Build Things: How to Restore Our Culture of Achievement, Build a Path for Entrepreneurs, and Create New Jobs in America
- The Hard Thing About Hard Things: Building a Business When There Are No Easy Answers
- Rise of ISIS: A Threat We Can't Ignore
- The World Is Flat 3.0: A Brief History of the Twenty-first Century
- Bad Feminist: Essays
- Steve Jobs
- Angela's Ashes: A Memoir
- How To Win Friends and Influence People
- Extremely Loud and Incredibly Close: A Novel
- The Sympathizer: A Novel (Pulitzer Prize for Fiction)
- The Silver Linings Playbook: A Novel
- Leaving Berlin: A Novel
- The Light Between Oceans: A Novel
- The Incarnations: A Novel
- You Too Can Have a Body Like Mine: A Novel
- The Love Affairs of Nathaniel P.: A Novel
- Life of Pi
- The Flamethrowers: A Novel
- Brooklyn: A Novel
- The Blazing World: A Novel
- The Rosie Project: A Novel
- The First Bad Man: A Novel
- We Are Not Ourselves: A Novel
- Bel Canto
- The Master
- A Man Called Ove: A Novel
- Beautiful Ruins: A Novel
- The Kitchen House: A Novel
- Interpreter of Maladies
- The Wallcreeper
- Wolf Hall: A Novel
- The Art of Racing in the Rain: A Novel
- The Cider House Rules
- A Prayer for Owen Meany: A Novel
- The Perks of Being a Wallflower
- Lovers at the Chameleon Club, Paris 1932: A Novel
- The Bonfire of the Vanities: A Novel
- Little Bee: A Novel

You are on page 1of 9

The main objectives of the work described are to devise an effective path-based motorcycle simulation capability and to add to understanding of how riders control motorcycles. Optimal linear preview control theory was previously applied to the tracking of a roadway by a car, using a simple car model operating in ﬁxed control. Similar theory is applied to path control of motorcycles. The simple car previously employed is replaced by a much more elaborate motorcycle. The steering angle control used previously is changed into steering torque control. Rider upper body lean torque is also allowed as a control input. The machine speed is considered constant but is a parameter of the motion. The objective of the optimal control is to minimize a weighted sum of tracking errors, rider lean angle and control power. The time-invariant optimal control corresponding to a white noise disturbance and to an inﬁnite optimization horizon is found for many situations, involving variations in machine speed and performance priorities. Tight controls, corresponding to high weightings on performance, and loose controls, corresponding to high weightings on control power, are identiﬁed. Results show the expected pattern for preview control, that information well into the future is of limited value in determining the present control inputs. Full system performance is achievable with only ﬁnite preview. The extent of the preview necessary for full performance is determined as a function of machine speed and performance priorities. This necessary preview is found to be in accord with conventional wisdom of motorcycle riding and rider training. Optimal path tracking preview controls are shown to represent the inverse dynamics of the motorcycle. New light is shed on the relative effectiveness of steering torque and body lean torque controls. Simulations of an optimally controlled motorcycle and rider combination are conducted. A typical lane change path and an S-shaped path from the literature are used. For a chosen speed, optimal controls are installed on the machine for which they were derived and simulation results showing tracking performance, control inputs, and other responses are included. Transformation of the problem from a global description, in which the optimal control is found, to a local description corresponding to the rider’s view, is described. It is concluded that a motorcycle rider model representing a useful combination of steering control capability and computational economy has been established. The model yields new insights into rider and motorcycle behavior. DOI: 10.1115/1.2745842 Keywords: motorcycle, steering, optimal control, tracking, preview, road

R. S. Sharp

Department of Electrical and Electronic Engineering, Imperial College, London SW7 2AZ, UK e-mail: robin.sharp@imperial.ac.uk

1

Introduction

Research on the steering control of cars has been reviewed by Guo and Guan 1 and MacAdam 2 . The latter also considered speed control. Some research is aimed at replicating the actions of human controllers, complete with their imperfections, including time delays and lack of skill but the main objective here is to devise a control scheme, representing the driver’s function to some extent, which is both simple and effective. Simplicity implies ease and speed of computation in simulations. Effectiveness implies accurate path following and speed maintenance throughout the operating range of the car, in the manner of a skilled driver. The main applications area of interest is virtual prototyping, which is growing in importance in vehicle manufacturing industries on a regular basis. The preoccupation in published work has been with improving path following control by increasingly elaborate use of the “path error” at a single preview point ahead of the vehicle, but the view taken here is that single-point preview, with no memory, is extremely limiting and multipoint preview opens up substantial possibilities for improved tracking performance. It is surmised that

Contributed by the Dynamic Systems, Measurement, and Control Division of ASME for publication in the JOURNAL OF DYNAMIC SYSTEMS, MEASUREMENT, AND CONTROL. Manuscript received September 26, 2005; ﬁnal manuscript received December 14, 2006. Review conducted by Hemant M. Sardar.

real drivers need knowledge of the whole of the road lateral proﬁle within a certain preview distance in order to control a vehicle well. Multipoint preview is presumed in 3–10 . Unless the control strategy is learned by copying a human driver, the great difﬁculty in this context is devising a strategy and deriving a set of control parameters, although notably in 8 complex nonlinear optimization was used. In more recent work, it has been shown how road proﬁle and vehicle dynamics models can be assembled into a discrete-time system, to which the theory of the linear quadratic optimal regulator LQR can be applied 11,12 . The objective function, to be minimized by the optimal control, is a sum of squares of path errors and possibly other road/vehicle metrics, together with the control power. Weighting parameters determine the relative importance to be attached to the cost components included. The optimal control, under the terms speciﬁed, can then be found easily. Model predictive control methods have been shown to reproduce the results of LQR theory with preview, when the problem is set up in a corresponding way 13 . The control is found to consist of a state-feedback part and a preview part. The state-feedback part is the same regardless of whether or not preview is included in the problem. It desensitizes the plant to disturbances in general. The preview part can be obtained by recursive use of the state-transition-matrix of the closedloop system and it consists of sets of gains, one for each control JULY 2007, Vol. 129 / 373

Journal of Dynamic Systems, Measurement, and Control Copyright © 2007 by ASME

Downloaded 06 Feb 2008 to 155.198.124.91. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

and closely related to the closed-loop system dynamics, to be applied to the previewed path errors, the products to combine linearly to yield the control inputs. An apparently effective but rather complex motorcycle steering controller for path tracking has been devised by Frezza et al. 14,15 . This nonlinear-feedback steer-angle controller was designed based on a very simple motorcycle model but has been applied to a full multibody simulation model. A rider control without preview has also been described in 16 . The absence of preview must be compensated by unrealistically high bandwidth in the assumed controller. Nevertheless, it is expected that, without path preview, the performance will not be competitive against models with preview. Single-point preview is utilized in the commercial package, BikeSim see http://www.carsim.com , together with a gain-scheduled PID roll angle to steer torque roll-tracking control and a user-selectable path-preview-error to target-rollangle gain. The user also needs to select the preview distance to be employed.

y k = Cz k + Du k

4

If y ri is a sample from a white-noise random sequence, the timeinvariant optimal control minimizing a cost function J, given that the pair A, B is stabilizable and the pair A, C is detectable 17 , is u* k = − Kz k with K = R + BT PB

−1 T

B PA

5

where the objective function to be minimized by the control, J n = limn→ k=0 zT k Qz k + TT k RT k and P satisﬁes the matrixdifference-Riccati equation, P = AT PA − AT PB R + BT PB

−1 T

B PA + Q

6

**2 Outline of Optimal Linear Preview Regulator Theory
**

The detailed development of the application of optimal linear preview control theory to vehicle steering control can be found in 11–13 , as stated above. In the interests of completeness, an outline is included here. The relevant linear vehicle model is translated to the discrete-time form, x v k + 1 = A vx v k + B vT k 1 y v k = C vx v k + D vT k and the lateral proﬁle of the road is considered in discrete sample value form, with sample values from past observations of the road ahead being stored as states of the full vehicle/road system. As the system moves forward in time, a new road sample value is read in and the oldest stored value is discarded, corresponding to the vehicle having passed the point on the road to which this oldest value refers. All the other road sample values are shifted through the time step, nearer to the vehicle. The dynamics of this shift register process are represented mathematically by y r k + 1 = Ary r k + Bry ri k where Ar is of the form 0 0 0 0 and Br is of the form 0 0 0 ... 1 Combining vehicle and road equations together, we obtain the full dynamic system description xv k + 1 yr k + 1 = Av 0 0 Ar xv k yr k + Bv 0 Tk + 0 Br y ri k 3 The complete problem is now in a standard form, z k + 1 = Az k + Bu k + Ey ri k 374 / Vol. 129, JULY 2007 1 0 0 0 0 1 0 0 ... ... 0 ... ... 0 1 0 ... 0 ... 0 2

... ... ... ... ... ...

Q = CTqC, containing the diagonal weighting matrix, diag q1 , q2 , q3 , with terms corresponding to the number of performance aspects contributing to the cost function, and R = diag r1 , r2 , corresponding to the two control inputs, steering torque and rider upper body lean torque. q1 is set to represent the importance of tracking errors, q2 relates to the absolute lean angle of the rider’s upper body, while q3 relates to the lean angle of the rider’s upper body relative to the motorcycle. r1 and r2 are set to reﬂect the costs associated with steering torque and rider lean torque respectively. High values of q2 promote the rider upper body leaning in opposition to the motorcycle, while high values of q3 restrict the rider leaning relative to the machine. From 12 , it is also known that the time-invariant optimal control given by 5 remains optimal if the white-noise sequence representing the road lateral proﬁle is low-pass ﬁltered, to make it more realistic, and “full” preview of the road 11,12 is available. It is implied that the white-noise disturbance assumption is not as restrictive as it might appear to be. As in 11 , the optimal control is found by partitioning the problem into non-preview and preview parts. First, the nonpreview, standard LQR optimal control is found. Then, the statetransition-matrix of the closed-loop system is applied recursively to the non-preview partition of the solution to the matrixdifference-Riccati equation to ﬁnd the preview gains. For path tracking, the optimal control solutions can be transformed from a ground-based framework into a vehicle-based one, as explained in 11 . The ground-based solution uses small angle theory, which requires that the path to be followed remains close to the ﬁxed line of the x-axis. The transformation allows the tracking of paths for which small angle theory applies only within the envelope that contains the vehicle and the part of the path immediately in front of the vehicle used for preview. After the transformation, the tracking potential of the controlled vehicle is much less restricted than before. Paths which change direction markedly, even circular paths, can be followed without special difﬁculties, provided that small angle theory applies over the preview distance. Such a condition will normally be satisﬁed naturally, since high path curvature implies restricted vehicle speed which, in turn, implies the need for only limited preview. Tracking is simulated by installing the state-feedback and preview controls to the vehicle, initializing so that the road states in the memory of the shift register correspond to the vehicle having traveled a straight path up to the present time, and designing a road lateral proﬁle which starts with a straight section at least as long as the preview distance. As the system steps through one time increment, the road proﬁle sample values within the preview distance are observed in the ground-based frame in which they are deﬁned, including a new value that was previously in front of the previewed path, and the position and attitude of the vehicle are updated. The vehicle position and orientation are used to transform the road samples into a vehicle-based reference frame, so that they represent the driver’s view of the road. The closed-loop Transactions of the ASME

Downloaded 06 Feb 2008 to 155.198.124.91. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

system is then completely deﬁned and the next discrete simulation step is taken. The process continues until a predeﬁned number of discrete time steps has been taken. For paths which follow the general direction of the earth x-axis, the transformation of the tracking problem from global-view to local-view is unnecessary. The MATLAB function DLSIM can be employed straightforwardly to carry out tracking simulations. This provides a useful check on the accuracy of the more elaborate local-view code.

3

Motorcycle Model Outline

Modeling of motorcycle handling is now a mature technology 18–23 . Especially, the advent of automated multibody mechanical systems software has enabled the construction of high-ﬁdelity models, which reproduce the behavior of real machines over a wide range of normal operating conditions. Measurements with instrumented vehicles have conﬁrmed the capabilities of properly conceived and executed models to mimic real behavior 24–36 . The model to be used here is considered to represent the state-ofthe-art and it is fully documented in 22 , with background material in 19 . The model has provided the basis for the commercial virtual motorcycle, BikeSim, which is routinely now used in product and component development, see http://www.carsim.com. The symbolic multibody software system AutoSim 37 , soon to be relaunched as VehicleSim, see http://www.carsim.com, was employed for the model building task and the code, a commented and understandable text ﬁle, is available at the web site http:// www.ic.ac.uk/controlandpower/motorcycles. The mathematical motorcycle model is considered representative of modern high-performance, road-going machines, traveling on ﬂat and level road surfaces. The model parameters, see 22 , are based on those of the Suzuki GSX-R1000, which is ﬁtted with the common telescopic front fork suspension and a swinging arm rear suspension with monoshock spring/damper arrangement. It is of tree-structure, except for one kinematic-loop-closure in the monoshock description. Alternative ways of representing the kinematics of the monoshock suspension are examined in 23 . The main frame is allowed unrestricted motion, while the rider’s upper body and the rear swing arm are pin-jointed to it. Torsional compliance at the steering head is allowed by the inclusion of a massless frame with roll freedom and spring/damper restraint to the main frame, the twist angle being treated as small. The steering system is pin-jointed to the massless twisting frame at the steering head. The lower front forks and front wheel can translate along the line of the forks, relative to the upper forks, and each road wheel is presumed axisymmetric and allowed to spin. Each tire has width and the contact points migrate, in general both circumferentially and laterally, so that the lowest point on each tire is taken to be the contact center. For each tire, the motion state of the contact center, the compression of the tire from the nominal static equilibrium state and the wheel camber angle are used in the “Magic Formula” tire model 20,22 to work out the steady state tire force and moment system. The lateral force and aligning moment, in each case, are lagged to allow for the tire carcass lateral compliance, with a relaxation length which is a function of speed. Suspension springs and dampers are deﬁned as linear but suspension and steering limit stops are included. A proportional/ integral speed controller giving a rear wheel spin torque in response to a speed error is included, so that target speed proﬁles can be followed. The speed controller has ﬁxed gains, which are non-critical. In particular, the speed controller allows equilibrium “trim” states to be found via straight-running simulations in which conditions change only very slowly. Steering control inputs can be considered to consist only of handlebar torque or to also include rider upper body lean torque. The model is considered valid, in a generic sense, for the full range of feasible speeds, wheel loads and lean angles, tire slip angles less than 8 deg, and vibration frequencies less than about 18 Hz. Journal of Dynamic Systems, Measurement, and Control

The AutoSim model ﬁle can be loaded in either of two ways, linear or nonlinear. In the latter case, a ready-to-compile-and-link simulation program can be obtained. In the former linear case, the system is symbolically linearized, for small perturbations about a general trim condition, and a MATLAB “M” ﬁle is written automatically, to set up the linear model in state space form. The symbolic MATLAB ﬁle can be used for the complete range of linear system operations encompassed by the MATLAB software. Such operations include automated conversion of the linear model to discrete-time form and discrete linear quadratic regulator optimization. The linear model generated by AutoSim has 27 states, including the longitudinal displacement from the origin and the wheel spin angles. For the optimal control calculations, it is necessary to reduce the model by the omission of those parts that relate to these ignorable coordinates. Thus 24 state variable values are used for feedback. When the trim condition involves straight running, symmetry demands that all the feedback gains associated with in-plane states are zero, which naturally turns out to be true.

4

Application of the Control Theory to Motorcycles

The theory of the discrete-time linear quadratic regulator with road preview, as described above, is now applied to steering control of motorcycles. According to the conclusions of 38 , in which it was established that a motorcycle with ﬁxed steering system has very unattractive stability properties in comparison with one having free steering, see also 18 , primary control is assumed to be through steering torque not steer angle, as with a car and rider upper body lean torque may also be included as a control. Solution of the optimization problem requires that a road model, of shift-register form, is appended to the motorcycle model as described in Sec. 2. Then, the standard linear quadratic regulator theory will yield time-invariant optimal tracking controls, if the road lateral disturbance is treated as white noise 11 or as low-pass ﬁltered white noise 12 with full preview being available. The optimal control has feedback and preview parts, in each case. The feedback part is the same as if no preview were considered. The preview part represents an inversion of the closed-loop system dynamics. It can be found conveniently by recursive application of the closed-loop system’s state-transition-matrix 11,12 . Even for a motorcycle of ﬁxed design, there are several parameters needed to deﬁne the control optimization problem. First, it can be presumed that the rider has only one control, the steering torque, or two, the steering torque and the rider’s upper body lean torque. Secondly, the machine speed is variable. Thirdly, we can choose different balances in the cost function between steer torque, represented by r1, rider lean torque if included , represented by r2, tracking error, represented by q1, absolute rider upper body lean angle, represented by q2, and relative rider body lean angle, represented by q3. Fourthly, the extent of the rider’s preview can be varied but, since the control is only optimal if the preview is sufﬁcient for full performance 11,12 , we will deal only in cases in which the gains fall to near zero within the preview distance.

5

Optimal Control Results

In the ﬁrst instance, let us consider the single control case, using 500 preview points. Figure 1 shows optimal preview gain sequences relating to steer torque for the nominal motorcycle/rider combination in straight-running at 10, 40, and 70 m / s speeds with q1 = 104, q2 = 4 104, q3 = 0. This selection of weights is informed, to some extent, by simulation results to be shown later. In each case, the linearization uses the appropriate trim state determined by simulation to equilibrium of the nonlinear system, and a set of optimal motorcycle state-feedback control gains applies. As expected, the only nonzero state-feedback gains are associated with out-of-plane state-variables. Symmetry of the straight-running condition implies zero gains for all in-plane states. Feedback gains JULY 2007, Vol. 129 / 375

Downloaded 06 Feb 2008 to 155.198.124.91. Redistribution subject to ASME license or copyright; see http://www.asme.org/terms/Terms_Use.cfm

Fig. 1 Optimal preview steer torque gains for tight singlecontrol system with 500 preview points, q1 = 104, q2 = 4 Ã 104, q3 = 0, r1 = 1, and three speeds, as functions of the distance ahead

are numerous and not so interesting, and they are not shown. The largest gains in the state-feedback control relate to roll angle and frame twist angle. The preview gain sequences are shown as continuous curves, although they are in fact discrete values corresponding to the ﬁxed discrete-time step of 0.02 s. The feedforward gains show convergence to zero as the preview distance increases, which is a well-known feature of preview control 39 . This convergence conﬁrms that the 500 preview points assumed are sufﬁcient. The relatively high values of q1 and q2 correspond to tight control 12 and the preview gains are high. With this tight control, at 10 m / s speed, a view of the road extending about 30 m forwards is all that is needed. Corresponding necessary preview distances for 40 and 70 m / s are 150 and 450 m, respectively. In terms of time, the necessary preview for full performance increases signiﬁcantly with speed. The most useful information in the path ahead is about 8 m in front of the motorcycle at 10 m / s, 35 m ahead at 40 m / s and at 65 m ahead at 70 m / s, a little under 1 s ahead in time, in each case. Figure 2 shows the contrasting results when the control is loose as represented by q1 = 100, q2 = 400, q3 = 0. The preview gains are reduced roughly in proportion to the change in the performance index weightings and, at each speed, the most useful path information is further ahead of the vehicle than with the tighter control of Fig. 1, at 15, 80, and 170 m for 10, 40, and 70 m / s, respectively. The inverse dynamics of the motorcycle and rider are apparent in the preview gain sequences, especially for 70 m / s speed. At this high speed, the weave mode of the machine is very lightly damped 19 , having a natural frequency of 3.2 Hz. The oscillations in the preview gain sequence, converted to a time basis, are at this same frequency. If the corresponding control is within the rider’s control bandwidth, it is implied that he/she will utilize control economically to excite a lightly damped mode to obtain a desired response, exactly as demanded in the optimization problem. Let us now allow rider upper body lean torque as a second control, running in parallel with steering torque. It can be surmised at the outset that most real riders will ﬁnd it difﬁcult to realize body lean torque controls at other than quite low frequen-

Fig. 3 Optimal preview gains for tight dual-control system with 500 preview points and with equal weights on the two controls, q1 = 104, q2 = 4 Ã 104, q3 = 0, r1 = r2 = 1, and three speeds, as functions of the distance ahead

cies and that should be remembered in interpreting results. A new issue is the relative weighting of the two control torques, as reﬂected in the value of the ratio of r2 to r1. In Fig. 3, these values are equal and the control is tight, as given by q1 = 104, q2 = 4 104, q3 = 0. The steering torque gains are almost the same as for the single-control case of Fig. 1 and the body lean torque gains are relatively small. The same pattern is clear in Fig. 4, corresponding to Fig. 2, in which the control is loose. The use of body lean torque control can be encouraged by reducing r2 to lower its cost. Figures 5 and 6 show the optimal control results. Again the steering torque gains are almost unchanged, while the body lean torque gains vary roughly in inverse proportion to the r2 value. Preview distances necessary for full performance can be quantiﬁed crudely by using a criterion introduced in 12 . According to this criterion, we consider the preview information ineffective beyond the point where, with a very long preview, 99% of the area under the total gain curve is enclosed. Results are shown in Figs.

Fig. 2 Optimal preview steer torque gains for loose singlecontrol system with 500 preview points, q1 = 100, q2 = 400, q3 = 0, r1 = 1, and three speeds, as functions of the distance ahead

Fig. 4 Optimal preview steer torque and upper body lean torque gains for loose dual-control system with 500 preview points and with equal weights on the two controls, q1 = 100, q2 = 400, q3 = 0, r1 = r2 = 1, and three speeds, as functions of the distance ahead

376 / Vol. 129, JULY 2007

Transactions of the ASME

Fig. 7 Required preview distances „99% controls, q1 = 104, q2 = 4 Ã 104, q3 = 0. Circles and crosses show lean torque for r1 = r2 = 1. torque and stars show lean torque for squares are for the single-control system.

criterion… for tight show steer torque Pluses show steer r1 = 1, r2 = 0.1. The

Fig. 5 Optimal preview steer torque and upper body lean torque gains for tight dual-control system with 500 preview points and with unequal weights on the two controls, q1 = 104, q2 = 4 Ã 104, q3 = 0, r1 = 1, r2 = 0.1, and three speeds, as functions of the distance ahead

7 and 8, for tight and loose controls, respectively, with ﬁve cases in each ﬁgure. As is clear from the inspection of the gain sequences, the necessary preview for full tracking performance increases more than in proportion to speed, with most cases being very similar to each other. However, for tight control, there are some curious differences between the requirements for body lean torque and for steering torque at high speeds. This result is thought not to be of any great signiﬁcance and no real understanding of it is claimed. The necessary preview distances are much greater than for a typical car at similar speeds 11,12 , as is well known to skilled motorcycle riders and riding instructors 40–42 .

6

Path Tracking Simulations

The simplest approach to path tracking is to use the optimal controls developed in a ﬁxed reference frame, as described above, to close the loop around the motorcycle/road system and to update

the state through time by repeatedly multiplying it by the closedloop system’s state-transition-matrix. At each update, a new road sample, from the furthest point along the preview distance, enters the problem as input. The MATLAB function DLSIM contains the process. This simple approach implies that the restriction to small angles in the linear theory of the control generation carries over into the tracking simulation. Paths which diverge substantially from the starting direction will be followed with imprecision, and not followed at all if the path direction changes by 90 deg or more. A better approach is to transform the problem to a vehicle basis. In such a basis, the reference axes are relocated to align with the motorcycle and the roadway is described relative to the motorcycle at each time step 11,12 . The small-angle theory now requires only that the path direction does not change radically over the preview distance. Bearing in mind the relationships between vehicle speed, feasible path curvature and required preview distance, it can be appreciated that this lesser need is likely to be satisﬁed in practical circumstances. The deﬁned path is designed to start with a straight section and the preview states of the combined roadway/motorcycle system are loaded initially with zeros, to correspond with the vehicle having traveled along a straight path, before the simulation commences, for the duration of the preview. The path proﬁle is then “seen” in the distance ﬁrst, as with a rider with a full view of the road ahead. The simulation proceeds by stepping through time, using the step length 0.02 s employed for the discrete-time optimal control calculations, until a speciﬁed number of steps has been completed. At each step, the absolute position and orientation of the motorcycle are updated and an account of the distance traveled from the start is kept. By comparison of the distance traveled by the motorcycle and the distance along the road from the start, it is known how the road relates to the machine at any time. The road data in front of the machine, covering the preview distance, is read in, transformed to the local view of the rider and used to replace the old values in that part of the system state vector that corresponds to the road proﬁle. Spline interpolation is

Fig. 6 Optimal preview steer torque and upper body lean torque gains for loose dual-control system with 500 preview points and with unequal weights on the two controls, q1 = 100, q2 = 400, q3 = 0, r1 = 1, r2 = 0.1, and three speeds, as functions of the distance ahead

Fig. 8 Required preview distances „99% criterion… for loose controls, q1 = 100, q2 = 400, q3 = 0. Circles show steer torque and crosses show lean torque for r1 = r2 = 1. Pluses show steer torque and stars show lean torque for r1 = 1, r2 = 0.1. The squares are for the single-control system.

Journal of Dynamic Systems, Measurement, and Control

JULY 2007, Vol. 129 / 377

Fig. 9 Lane change to the right at 30 m / s, with the upper graph showing the roadway and the motorcycle path and the lower graph showing the steering torque and the rider’s upper body lean torque applied. The controls presume 180 m rider preview and weightings of q1 = 104, q2 = 4 Ã 104, q3 = 0, and r1 = r2 = 1.

Fig. 11 Lane change to the right at 30 m / s, with the upper graph showing the roadway and the motorcycle path and the lower graph showing the steering torque and the rider’s upper body lean torque applied. The controls presume 180 m preview and weightings of q1 = 104, q2 = 4 Ã 104, q3 = 0 and r1 = 1, r2 = 0.1.

used to obtain the local lateral road proﬁle at the proper preview points. At the same time, the motorcycle states representing its lateral position and its attitude angle are set to zero, corresponding to the reference axes being relocated to align with the machine. Many maneuvers have been studied, including randomly generated, low-pass-ﬁltered lateral proﬁles as in 11 , lane changes with geometry from 43 , sudden direction changes as in 11,12 and an S-path maneuver devised in 14 . Exemplary results are shown next, ﬁrst for the lane change, then for the S-path. Figures 9 and 10 show a lane change at 30 m / s speed. Both steering torque and rider upper body lean torque controls are included with equal weights as indicated by r1 = r2 = 1, with 300 preview points corresponding to 180 m rider preview being assumed and tight control deriving from the high relative cost of tracking errors and rider total lean angle implied by q1 = 104, q2 = 4 104, q3 = 0. Despite the tight control, some corner-cutting occurs, the extent being controllable through the choice of the

weights. Figures 11 and 12 show similar results obtained for a reduced cost on rider lean torque in the control design, given by r1 = 1, r2 = 0.1. As should be expected, much more lean torque is utilized but the tracking accuracy is substantially unchanged and the steering torque history is nearly the same as in the previous case. Apparently, the tracking is hardly improved, even by the application of large rider lean torques. The main difference is in respect of the rider upper body relative rolling motion, which is in opposition to that of the motorcycle. The same maneuver performed with equal weights on the two controls and much looser control r1 = r2 = 1, q1 = 100, q2 = 400, q3 = 0 is shown in Figs. 13 and 14. With the higher priority on reducing the control power, the control activity starts earlier and the levels of steering torque and rider body lean torque used are much reduced. The lean torque is near zero, in fact. The cornercutting behavior is exaggerated and the maximum roll angle achieved is only about one third of its previous value. In general form, the time-histories of the motions shown agree with experi-

Fig. 10 Lane change to the right at 30 m / s, showing various aspects of the motions, scaled for clarity. The controls presume 180 m preview and weightings of q1 = 104, q2 = 4 Ã 104, q3 = 0, and r1 = r2 = 1.

Fig. 12 Lane change at 30 m / s, showing various aspects of the motions, scaled for clarity. The controls presume 180 m preview and weightings of q1 = 104, q2 = 4 Ã 104, q3 = 0, and r1 = 1, r2 = 0.1.

378 / Vol. 129, JULY 2007

Transactions of the ASME

Fig. 13 Lane change at 30 m / s, with the upper graph showing the roadway and the motorcycle path and the lower graph showing the steering torque and the rider’s upper body lean torque applied. The controls presume 250 m preview and weightings of q1 = 100, q2 = 400, q3 = 0 and r1 = r2 = 1.

Fig. 15 S-maneuver simulation at 30 m / s, with the upper graph showing the roadway and the motorcycle path and the lower graph showing the steering torque and the rider’s upper body lean torque applied. The controls presume 90 m preview and weightings of q1 = 104, q2 = 0, q3 = 105 and r1 = r2 = 1. With these weightings, the rider lean torque is small compared with the steering torque.

mental results for lane change maneuvering shown in 43–45 . The steer angles here are noticeably smaller than in the references, but that is likely to be a feature of the particular machine and tires simulated. When this trial is repeated with a lower cost associated with the use of body lean control r2 = 0.1 , the only signiﬁcant change is that rider lean torque is increased roughly by a factor of 10. Other changes of behavior are small, reinforcing the idea that body lean torque control is ineffective. These results are not shown for brevity. Clearly, within the optimal preview control theory framework, there is much scope for varying the rider’s strategy and actions according to his/her priorities. Figures 15–17 contain S-path results, again for a speed of 30 m / s, at which the maneuver is a severe challenge to a motorcycle. Tight controls as given by q1 = 104, q2 = 0, q3 = 105 and r1 = r2 = 1 are utilized, thereby keeping the necessary preview quite short at 90 m. Substantial control torques are applied and the motorcycle roll angle reaches about 50 deg in each direction. The roll

angle is inﬂuenced by the lateral migration of the tire to road contact points round the tire cross section. The simulation continues to rely on the linear model, as used to generate the optimal preview controls, and the results will not account for geometric nonlinearities in the real problem. However, the smallness of the tire slip angles suggests that even the linear model deals with the tire forces well. Most of the necessary side-force is generated through wheel camber. The front tire side force magnitude reaches 765 N, while that at the rear peaks at 1080 N. As compared with the results shown in 14 , the responses here are smoother and the tracking errors are smaller, with more evidence of control action ahead of the path change, but it should be remembered that varying the control tightness will smoothen or sharpen the motions shown above. Tracking errors will be inﬂuenced correspondingly. The steer angle history shows similar peak

Fig. 14 Lane change to the right at 30 m / s, showing various aspects of the motions, scaled for clarity. The controls presume 250 m rider preview and weightings of q1 = 100, q2 = 400, q3 = 0 and r1 = r2 = 1.

Fig. 16 S-maneuver simulation at 30 m / s, showing various aspects of the motion, scaled for clarity. The controls presume 90 m rider preview and weightings of q1 = 104, q2 = 0, q3 = 105 and r1 = r2 = 1, as in Fig. 15. The weightings cause the rider to lean just a little relative to the motorcycle.

Journal of Dynamic Systems, Measurement, and Control

JULY 2007, Vol. 129 / 379

values to those of 14 but a much smoother and, arguably, a more realistic, proﬁle. If the simulations of Figs. 9–17 are repeated using only steer torque as a control input, almost identical path tracking and vehicle motion results are obtained. This is conﬁrmatory of the opinion of Weir and Zellner 46 , formed by theoretical analysis, and the ﬁndings of Aoki 25 and of Katayama et al. 44 , who concluded from experimentation that rider lean control of relatively massive single-track vehicles is completely secondary to steering torque control. It is apparent that a motorcycle rider’s movements relative to the machine contribute to improving the dynamics in a largely passive manner. Such movements can often be observed to occur, in racing for example, even before the application of signiﬁcant steering torque, in preparation for a maneuver and not as a part of it.

7

Conclusions

The kind of optimal linear preview control theory previously applied to path following by cars can be applied to motorcycles without particular difﬁculty. The high order of the motorcycle model used here contrasts with the simplicity of the car models employed in previous work. The linearity of the model is its key feature in the application. The time-invariant optimal control that corresponds to minimization of a weighted sum of path errors, rider upper body lean angle and control power over an inﬁnite horizon with a white noise disturbance is of familiar form. Diminishing returns are obtained for preview beyond a certain distance in front of the vehicle and the notion of full preview applies. There is no point in “seeing” further forwards of the motorcycle than a certain distance. The necessary preview distance increases more than in proportion to speed and is much greater, for a given speed, than is necessary for a typical car. This consequence of the theory accords perfectly with riding experience and conventional wisdom. The optimal preview control represents a kind of inversion of the state-feedback controlled system’s dynamics, such that an oscillatory plant requires a correspondingly oscillatory preview gain sequence for effective tracking. At high speeds, when a highperformance motorcycle typically will become oscillatory, the rider apparently needs to employ a gain sequence showing the same frequency content as the lightly damped machine modes. If the oscillations are within the rider’s control bandwidth, with high skill, he/she will utilize control at the eigenfrequency to get good response with little control power. Bandwidth limitations of the rider have not been included in this treatment, so that further work needs to be done to complete the developing picture. As with the car, tight and loose controls can be generated by putting more or less priority on tracking accuracy in the cost function. Loose controls are more demanding of preview than are tight ones. It is implied that, under restricted preview conditions, riders will adopt tighter control strategies, which involves working harder at the riding task, again, in line with common experience. Similar steering torque and rider upper body lean torque magnitudes are typical, but they are very much to the rider’s choice. The optimal lean torque can be made larger by placing a lower cost on its use. The optimal steering torque control is hardly affected by the inclusion of lean torque as a control or by the

weighting attached to it. Neither is the tracking performance much inﬂuenced by the lean torque. It is clear in the results that lean torque control is quite a minor issue, mainly inﬂuencing the rolling motions of the rider’s upper body relative to the motorcycle. It appears that riders’ body motions are relatively prosaic, contributing to improving the view of the road and improving the system dynamics in a very low-frequency passive manner. With various sets of time-invariant optimal controls installed, the motorcycle will, in general, successfully track a deﬁned path. As the control is made looser, so the corner-cutting and the application of control inputs ahead of the main features of the path change increase. This early application of control, especially in relaxed driving, is a well-known feature of the real-world activity and is nicely reproduced here. The controls become smoother and are of reducing magnitude as the control power increases in importance in the cost function relative to the path tracking. The path tracking is economical computationally, since the optimal controls are obtained off-line and are invariant for a ﬁxed speed. Variable speed cases and optimal controls obtained for linearizations about cornering trim states remain to be dealt with, but it is easy to imagine a set of gain scheduled controllers with adaptation to speed and lean angle as extensions of the work so far completed. Further studies are being conducted to discover how the design of the motorcycle inﬂuences the requirement for preview, on the basis that advantage to the rider concerned with high performance will accrue from minimizing the need for planning the control inputs well ahead of a maneuver. It is hoped that these studies will bring increased understanding of the potential conﬂict between stability and responsiveness, which is a part of two-wheeled vehicle folklore.

Acknowledgment

The support of the Engineering and Physical Sciences Research Council for the author’s work is gratefully acknowledged.

Nomenclature

A, B, C, D, E A r, B r A v, B v, C v, D v J K P q R T k u xv yv y yr y ri z matrices in a general state-space system representation matrices for road model matrices for vehicle model cost function gain vector solution of the matrix-Ricatti difference equation q1 0 0; 0 q2 0; 0 0 q3 weighting matrix for performance r1 0; 0 r2 weighting matrix for controls control torque vector discrete-time marker control vector with optimal value u* vehicle state vector vehicle output output vector road state vector new road displacement value, input to system general system state vector

References

1 Guo, K., and Guan, H., 1993, “Modelling of Driver/Vehicle Directional Control System,” Veh. Syst. Dyn., 22 3-4 , pp. 141–184. 2 MacAdam, C. C., 2003, “Understanding and Modelling the Human Driver,” Veh. Syst. Dyn., 40 1-3 , pp. 101–134. 3 Kraiss, K. F., and Kuettelwesch, H., 1992, “Identiﬁcation and Application of Neural Operator Models in a Car Driving Situation,” Proceedings of the 5th IFAC/IFIP/IFORS/IEA Symposium on Analysis, Design and Evaluation of Man-Machine Systems. 4 Modjtahedzadeh, A., and Hess, R. A., 1993, “A Model of Driver Steering Control Behavior for Use in Assessing Vehicle Handling Qualities,” ASME J. Dyn. Syst., Meas., Control, 115, pp. 456–464.

Fig. 17 Path tracking error in S-maneuver of Figs. 15 and 16

380 / Vol. 129, JULY 2007

Transactions of the ASME

5 Peng, H., and Tomizuka, M., 1993, “Preview Control for Vehicle Lateral Guidance in Highway Automation,” ASME J. Dyn. Syst., Meas., Control, 115, pp. 679–686. 6 Shimura, A., and Yoshida, K., 1998, “Steering Control for Car Cornering by Means of Neural Network and Genetic Algorithm,” Proceedings of the IFAC Symposium on Intelligent Components for Vehicles, Seville. 7 Sharp, R. S., Casanova, D., and Symonds, P., 2000, “A Mathematical Model for Car Steering, With Design, Tuning and Performance Results,” Veh. Syst. Dyn., 33 5 , pp. 289–326. 8 Prokop, G., 2001, “Modelling Human Vehicle Driving by Model Predictive On-Line Optimisation,” Veh. Syst. Dyn., 35 1 , pp. 19–35. 9 Guo, K., Pan, F., Cheng, Y., and Ding, H., 2002, “Driver Model Based on the Preview Optimal Artiﬁcial Neural Network,” Proceedings of the AVEC’02. 10 Guo, K., Cheng, Y., and Ding, H., 2004, “Analytical Method for Modelling Driver in Vehicle Directional Control,” in The Dynamics of Vehicles on Roads and on Tracks, M. Abe, ed., Supplement to Vehicle System Dynamics, Taylor and Francis, London, Vol. 41, pp. 401–410. 11 Sharp, R. S., and Valtetsiotis, V., 2001, “Optimal Preview Car Steering Control,” Supplement to Vehicle System Dynamics, P. Lugner and K. Hedrick, eds., Vol. 35, pp. 101–117. 12 Sharp, R. S., 2005, “Driver Steering Control and a New Perspective on Car Handling Qualities,” J. Mech. Eng. Sci., 219, pp. 1041–1051. 13 Cole, D. J., Pick, A. J., and Odhams, A. M. C., 2006, “Predictive and Linear Quadratic Methods for Potential Application to Modelling Driver Steering Control,” Veh. Syst. Dyn., 44 3 , pp. 259–284. 14 Frezza, R., and Beghi, A., 2003, “Simulating a Motorcycle Driver,” in New Trends in Nonlinear Dynamics and Control, W. Kang et al., eds., SpringerVerlag, Berlin, Heidelberg, pp. 175–186. 15 Frezza, R., Beghi, A., and Saccon, A., 2004, “Model Predictive and Hierarchical Control for Path Following With Motorcycle: Application to the Development of the Pilot Model for Virtual Prototyping,” Proceedings of the 43rd Conference on Decision and Control, Dec. 14–17, Atlantis, Bahamas, Paper No. TuB09.4, CD Rom. 16 Huyge, K., Ambrósio, J., and Pereira, M., 2005, “A Control Strategy for the Dynamics of a Motorcycle, Including Rider,” Proceedings of the ENOC-2005, Eindhoven, Netherlands, 7–12 August, CDRom. 17 Anderson, B. D. O., and Moore, J. B., 1971, Linear Optimal Control, PrenticeHall, Englewood Cliffs, New Jersey. 18 Sharp, R. S., 2001, “Stability, Control and Steering Responses of Motorcycles,” Veh. Syst. Dyn., 35 4-5 , pp. 291–318. 19 Sharp, R. S., and Limebeer, D. J. N., 2001, “A Motorcycle Model for Stability and Control Analysis,” Multibody Syst. Dyn., 6 2 , pp. 123–142. 20 Pacejka, H. B., 2002, Tyre and Vehicle Dynamics, Butterworth Heinemann, London. 21 Cossalter, V., and Lot, R., 2002, “A Motorcycle Multi-Body Model for Real Time Simulations Based on a Natural Coordinates Approach,” Veh. Syst. Dyn., 37 6 , pp. 423–447. 22 Sharp, R. S., Evangelou, S., and Limebeer, D. J. N., 2004, “Advances in the Modelling of Motorcycle Dynamics,” Multibody Syst. Dyn., 12 3 , pp. 251– 283. 23 Sharp, R. S., Evangelou, S., and Limebeer, D. J. N., 2005, Multibody Aspects of Motorcycle Modelling With Special Reference to Autosim, Advances in Computational Multibody Systems, J. G. Ambrósio ed., Springer-Verlag, Dordrecht, The Netherlands, pp. 45–68. 24 Eaton, D. J., 1973, “Lateral Dynamics of the Uncontrolled Motorcycle,” Proceedings of the 2nd International Congress on Automotive Safety, San Francisco.

25 Aoki, A., 1979, “Experimental Study on Motorcycle Steering Performance,” SAE 790265. 26 Weir, D. H., and Zellner, J. W., 1979, “Experimental Investigation of the Transient Behavior of Motorcycles,” SAE 790266. 27 Verma, M. K., Scott, R. A., and Segel, L., 1980, “Effect of Frame Compliance on the Lateral Dynamics of Motorcycles,” Veh. Syst. Dyn., 9 3 , pp. 181–206. 28 Thomson, B., and Rathgeber, H., 1984, “Automated Systems Used for Rapid and Flexible Generation of System Models Exempliﬁed by a Veriﬁed Passenger Car and a Motorcycle Model,” The Dynamics of Vehicles on Roads and on Railway Tracks, J. Karl Hedrick, ed., Swets and Zeitkinger, Lisse, pp. 645– 654. 29 Takahashi, T., Yamada, T., and Nakamura, T., 1984, “Experimental and Theoretical Study of the Inﬂuence of Tires on Straight Running Motorcycle Weave Response,” SAE 840248. 30 Hasegawa, A., 1985, “Analysis of Controllability and Stability of Motorcycles - Analysis of Stability at High Speed Driving, 10th International Technical Conference on Experimental Safety Vehicles, Oxford, UK, July 1985, pp. 480–500. 31 Nishimi, T., Aoki, A., and Katayama, T., 1985, “Analysis of Straight Running Stability of Motorcycles, 10th International Technical Conference on Experimental Safety Vehicles, Oxford, UK, July 1985, pp. 1080–1094. 32 Bayer, B., 1988, “Flattern und Pendeln bei Krafträdern,” Automobil_Industrie, 2, pp. 193–197. 33 Sugizaki, M., and Hasegawa, A., 1988, “Experimental Analysis of Transient Response of Motorcycle Rider Systems,” SAE 881783. 34 Iffelsberger, L., 1991, “Application of Vehicle Dynamics Simulation in Motorcycle Development,” Safety Environment Future, R. Brenndicke, ed., pp. 511–525. 35 Wisselmann, D., Iffelsberger, L., and Brandlhuber, B., 1993, “Einsatz Eines Fahrdynamik-Simulationsmodells in der Motorradentwicklung bei BMW,” ATZ, 95 2 , pp. 56–63. 36 Bortoluzzi, D., Doria, A., Lot, R., and Fabbri, L., 2000, “Experimental Investigation and Simulation of Motorcycle Turning Performance,” Sicherheit Umwelt Zukunft 111, Tagungsband der 3 Internationalen Motorradkonferenz, Institut für Zweiradsicherheit, Essen. 37 Sayers, M. W., 1999, “Vehicle Models for RTS Applications,” Veh. Syst. Dyn., 32 4-5 , pp. 421–438. 38 Sharp, R. S., 1971, “The Stability and Control of Motorcycles,” J. Mech. Eng. Sci., 13 5 , pp. 316–329. 39 Tomizuka, M., 1976, “Optimal Linear Preview Control With Application to Vehicle Suspension - Revisited,” ASME J. Dyn. Syst., Meas., Control, 98, pp. 309–315. 40 Watanabe, Y., and Yoshida, K., 1973, “Motorcycle Handling Performance for Obstacle Avoidance,” 2nd International Congress on Automotive Safety, San Francisco. 41 Code, K., 1993, “A Twist of the Wrist II - The Basics of High Performance Motorcycling,” California Superbike School, Technical Report. 42 Coyne, P., Mayblin, W., and Mares, P., 1998, Motorcycle Roadcraft: The Police Riders’ Handbook, Home Ofﬁce, London. 43 Rice, R. S., 1978, “Rider Skill Inﬂuences on Motorcycle Maneuvering,” SAE SP-428, pp. 79–90. 44 Katayama, T., Aoki, A., and Nishimi, T., 1988, “Control Behaviour of Motorcycle Riders,” Veh. Syst. Dyn., 17, pp. 211–229. 45 Zellner, J. W., and Weir, D. H., 1978, “Development of Handling Test Procedures for Motorcycles,” SAE SP-428, pp. 91–100. 46 Weir, D. H., and Zellner, J. W., 1978, “Lateral-Directional Motorcycle Dynamics and Rider Control,” SAE SP-428, pp. 7–31.

Journal of Dynamic Systems, Measurement, and Control

JULY 2007, Vol. 129 / 381

- Optimal State Space ControlUploaded byfilipetaveiros
- Motorcycle Dynamics studyUploaded byJuan Lopez
- Motorcycle DynamicsUploaded byananth95nandu
- Motorcycle Modeling and ControlUploaded bykeeblerkhan
- Lecture2 - System Modeling, Standard Form, Laplace Transform.pdfUploaded byBurak Osman Yuce
- Design of a Tuned Mass Damper for Seismic Excited BuildingUploaded byDevang Bhavsar
- 25Uploaded bysouvik5000
- Motorcycle Dynamics Library in ModelicaUploaded bySaufiy Sarmin
- Comparison of Pole Placement and LQR Applied.pdfUploaded byAnonymous Gmk3t4Q
- Motorcycle DynamicsUploaded bySai Sujan
- [13]Uploaded byFernando Ramos
- Bogdan_1Thesis_1Uploaded byPhan Minh
- Assignment 1Uploaded byauuser2011
- ee580_notes.pdfUploaded bynagesh
- 04213169Uploaded bygiba_c
- A51 1 Vrdoljak10 Applying Optimal Sliding Mode Based Load Frequency Control in Power Systems With Controllable Hydro Power PlantsUploaded byStevan Milivojevic
- 10.1049@pe19980507Uploaded byPavan Singh Tomar
- DownloadUploaded bySaikat Gope
- Session 11Uploaded byabolfazl
- Comandos de MatLab.docxUploaded byJose Arturo Torres
- Design of Decentralised Load FrequencyUploaded byCarlos Martinez
- 01359955Uploaded bydoud98
- Chaper 3_c b BangalUploaded byRakesh Polimera
- Category_Control theory - Wikipedia, the free encyclopedia.pdfUploaded byjjj881
- 6SL3244 0BB00 1PA1 Datasheet EnUploaded byshailesh284
- Load frequency analysis and its schematic controlUploaded byIRJET Journal
- Zh 05110257261Uploaded byAlion
- WipUploaded byAnas M. M. Arif
- signal and system Lecture 20Uploaded byali_rehman87
- A Process Control ExperimenUploaded byasepsow