Content uploaded by Michael Kaufmann
Author content
All content in this area was uploaded by Michael Kaufmann on May 23, 2015
Content may be subject to copyright.
9th International Workshop on Research and Education in Mechatronics
September 18th-19th 2008, Bergamo, Italy
DIFFERENT SCHEMES FOR BUMPLESS MANUAL/AUTOMATIC TRANSFER
Moritz Werling1, Michael Kaufmann1, Lutz Gr¨oll2
1Institute of Applied Computer Science/Automation, University of Karlsruhe (TH), 76128 Karlsruhe, Germany
2Institute of Applied Computer Science, Forschungszentrum Karlsruhe (FZK), 76021 Karlsruhe, Germany
Abstract: New technologies, such as Advanced Driver Assistance Systems (Adaptive Cruise Control, Auto-
matic Parking, Lane Keeping Assistance, etc.) lead to a cooperation of human and machine and thus require
manual/automatic transfer (MAT). Similar situations for human-machine interaction can be found in robotics
and process engineering. Therefore, the primary focus of study in this paper is giving an overview on so-called
bumpless transfer schemes for switching between automatic operation modi and the analysis of their applica-
bility to MAT.
Based on an analysis of bumpless transfer (roughly meaning continuous plant-input), different switching
schemes are rated with respect to their application in experimental vehicles. A practical realization is given
with AnnieWAY [11], an autonomous vehicle successfully entering the Urban Challenge 2007 final.
Key words: Manual/Automatic Transfer, Bumpless Transfer, Anti-windup, Override Control, Plant-input
Substitution, Autonomous Experimental Vehicle
I INTRODUCTION
Complex automation tasks require the combination of
multiple control modi (different control algorithms,
open/ closed loop, etc.) and suitable transfer strate-
gies, forming a control topology as depicted in Fig. 2.
These modern control systems often make use of the
complementary capabilities of human (flexibility, high
cognition) and machine (reliability). Hence, not only
transfer schemes between different automatic modes
are required, but also those between manual and auto-
matic.
-
-
y
d1
C
1
C
2
u
1
u2
u
y
1
y
2
y
d2
P
Figure 1: Hard switching between controllers
The most simple mechanism for mode transfer can
be seen in Fig. 1. Here, the transfer is carried out with-
out a synchronization of the manipulated variable and
hence in general bumpy (hard switching). For many
applications this mechanism is particularly chosen for
its simplicity and/or the associated safety concept.
In contrast to purely electronical applications,
mechatronical systems often involve mechanical ac-
tuators, which are much more sensitive to signal dis-
continuities and extremely high rates of signal change.
This is one reason for dealing with so-called bumpless
transfer. Others are safety [13], comfort aspects [8],
and tracking performance [9].
Bumpless transfer between different automatic con-
trol modes has been an intensively investigated field
of research for many years, as indicated by the vari-
ous literature cited in the previous and the following
paragraph. According to the knowledge of the authors
however, there is no literature available at all deal-
ing with the bumpless automatic-to-manual transfer,
which gives rise to this contribution.
The remainder of the paper proceeds as follows.
Section II gives an overview on classical bumpless
transfer schemes, such as anti-windup [2] (condition-
ing technique [6], realizable reference [6], observer
technique [1]), override control [4], and cross-fading
[5]. Section III takes on these schemes and verifies
them in terms of their transferability to manual/
automatic transfer (MAT). The derived schemes are
discussed and practical applications are given. Fur-
thermore, special MAT schemes, not derived from
bumpless transfer techniques are provided in Sec. IV
along with impact mitigation measures in case of con-
troller failure. Section V concludes and provides a
summary.
REM2008, Bergamo, Italy
(
. . .
)
Controller1
Controller2
ControllerN
Manual
Manual mode
- start-up/shut-down cycle
- set-point changes
- emergency intervention
Feedback control mode
- Change of control parameters
(tracking/setpoint controller)
- Change of control laws
(actuator/sensor failure, bifur-
cation, neglegibility of higher
dynamics)
- Change of reference variables
(di erent control objectives with
the same plant-input variable)
Open loop
Open-loop mode
- time program control
- experiments
Figure 2: Different operation modi of complex multi-control systems.
II DIFFERENT TECHNIQUES FOR
HANDLING
PLANT-INPUT-SUBSTITUTION
A. Anti-windup
Initially, windup was considered a negative effect of
the integral action of PID controllers which was en-
countered when saturation elements are located be-
tween the control output and the plant input. With
plain PID-control the control error is continuously in-
tegrated during the saturation phase without any effect
on the plant and the controller winds-up. Only when
the control error changes its sign, the integrator out-
put degrades gradually (”wind-down“), which results
in ample overshoots or even instability.
This integrator windup represents only a special case
of windup. A more general definition can be found
in [6,12], which define controller windup as a discrep-
ancy between the controller states and the plant input.
Besides saturation, other reasons for this discrepancy
can be so-called plant input substitutions which makes
anti-windup techniques not only suitable for internal
control mode changes but also for MAT.
Figure 3 shows the classical scheme, also known as
anti-reset-windup, which avoids the described
phenomenon by feedback-controlling the integral part
so that the mismatch between controller output u1and
plant input uvanishes.
Other anti-windup techniques are based on the ma-
nipulation of the reference signal (realizable Refer-
ence [6]) or apply observer theory in order to deter-
mine suitable control states [1]. An overview can be
found in [7].
A descriptive example application is given by the
velocity controller of a cruise control system of series
vehicles. As soon as the engine throttle is fully open,
the anti-windup scheme prevents the integrator, nec-
essary for stationary accuracy in the presence of road
slopes and wind, from winding-up.
-
-
-
Figure 3: Anti-reset-windup
B. Override Control
An override control problem, as described in [3, 4], is
characterized by multiple differently prioritized, par-
tially conflictive control objectives. As shown in Fig. 4,
min/max-operators often solve that problem. Typical
applications in process engineering are flow-rate con-
trol systems which ensure the compliance of a maxi-
mum pipe pressure. Both the flow and the (maximum)
pressure controller are combined via a min-operator
always forwarding the more conservative controller
output to the valve.
A straightforward usage of this override control
scheme to automotive application is adaptive cruise
control (ACC), a combination of distance and velocity
control [10]). During times of a free headway the ve-
locity controller influences the acceleration of the ve-
REM2008, Bergamo, Italy
-
-
Figure 4: Override control via min-operator
hicle so that the desired velocity given from the driver
is held. As soon as another vehicle is approached from
the rear, the distance controller overrides bumplessly
the velocity controller via the min-operator, and the
desired safety distance is kept.
C. Cross-fading of Static Controllers
The cross-fading of multiple controllers (see Fig. 6)
can be found in practice. Its proof of stability however
might be difficult. For this reason often only static
controllers (memoryless) are combined in the transi-
tion of their scopes. In [5] for example, ACC is real-
ized by state-scheduled cross-fading of multiple con-
trollers for different control objectives, such as speed
control, distance control, and e-braking.
-
-
Figure 5: Crossfading of multiple static controllers
D. Synchronization of the Reference Signal
In many applications with control modi aiming for dif-
ferent control objectives, there is the necessity for syn-
chronizing the reference signal with the current plant
states. Without this synchronization the control sys-
tem’s transient behavior would be characterized by big
amplitudes. One fairly successful anti-windup scheme,
called realizable reference carries out this idea by cal-
culating a virtual reference signal in a way, that the
discrepancy between the controller output and the plant
input vanishes [9]. This method as well as the slew-
rate limited reference signal can be regarded as a low-
level transient generation, which are not suitable for
complex trajectory tracking tasks, that assert collision
free movements. In the latter, a dynamic replanning is
required in order to avoid obstacles during the transi-
tion.
-
transient/
replanning
Figure 6: Synchronization of the reference signal
III TRANSFERABILITY TO MAT
From a controller’s point of view, there is little differ-
ence between switching from another automatic mode
and switching from manual operation. Therefore, the
previously introduced bumpless transfer schemes can
directly be applied to manual-to-automatic transfer.
The only precarious issue is, that the manual control
objective nearly always differs from the one of the au-
tomatic mode (why else would the operator have inter-
vened?), which necessitates the synchronization of the
reference signal as described in Sec. D. in most cases.
In the opposite direction however, meaning from
automatic to manual mode, the described automatic-
to-automatic transfer schemes cannot be directly ap-
plied. This arises from the fact, that
•the operator cannot be directly modified / ex-
tended,
•the haptics of the operating device has to be
maintained,
•the operator needs to be able to intervene at any
time, even if the system becomes instable.
Therefore, only the basic ideas of bumpless automatic-
to-automatic transfer are picked up and adapted to
automatic-to-manual transfer.
A. Synchronization of the Manual Operat-
ing Device
The automatic synchronization of the manual operat-
ing device can be regarded as a ”manual anti-windup“
scheme in the sense of eliminating the discrepancy be-
tween the manual control output and the plant input
REM2008, Bergamo, Italy
-
Figure 7: Synchronization of the Manual Operating
Device
of chapter A.. This synchronization scheme is ap-
plied in the process industry when a manual nulling
of the manipulated variable is impossible for temporal
or safety-related reasons. Therefore an additional ac-
tuator permanently positions the manual operating de-
vice according to the automatic manipulated variable
as indicated in Fig. 7. In critical situations the oper-
ating device is ready for takeover by the operator and
asserts so continuity, not necessarily bumplessness, of
the plant input signal during the transfer. Prime exam-
ples are the autopilot function of aircrafts or vehicular
park assist systems.
B. Manual Override Control
-
Figure 8: Manual Override Control
Manual override control is a direct transfer of au-
tomatic override control to MAT by substituting one
controller by the operator. Here, likewise, the min/max
operator is chosen in a way, such that the more con-
servative manipulated variable is put through to the
plant, enabling the operator to take corrective action in
critical situations. A practical implementation of this
scheme is the cruise control example of Sec. B.. Due
to the split-range1character of the vehicle’s acceler-
ation control elements, namely gas and break pedal
input, the driver is capable of overriding the automatic
manipulated variables in ”both directions“, meaning
he or she might brake harder or accelerate more than
the controller does. It is worth mentioning that in the
beginning cruise control utilized an additional actuator
for the synchronization of the gas pedal, which was,
1multiple plant inputs fordifferent input ranges
probably for safety reasons (deadlock), removed later
on.
C. Manual Cross-fading
-
Figure 9: Manual cross-fading
In situations where the focus is on bumpless trans-
fer rather than on regaining precise control over the
system as fast a possible, the cross-fading technique of
sec. C. will work for the transfer to manual control. As
depicted in Fig. 9, the detection of a manual interven-
tion initiates a cross-fading routine, which modulates
the weights from a 100% automatic to a 100% man-
ual operation gradually over time and asserts therefore
bumpless transfer.
IV SPECIAL MANUAL TRANSFER
TECHNIQUES
In contrast to the last sections, this section deals with
techniques, being rather intuitively motivated than de-
rived directly from classical bumpless transfer.
A. Additive Overlay
-
Figure 10: Additive overlay with continuous offset re-
duction
An obvious solution to the bumpless transfer be-
tween automatic and manual mode is the additive over-
lay of the respective control outputs, as depicted in
Fig.10. Since the manual part of the manipulating
variable is equivalent to a plant input disturbance, the
REM2008, Bergamo, Italy
controller may work against the operator and there-
fore ucneeds to be held constant with a triggered hold
element right at and after the moment of transfer. If
a permanent offset interferes with the haptics of the
manual operating device, the offset can be gradually
degraded over time according to Fig.10 and the oper-
ator has enough time to compensate for this reduction.
Taking a closer look at brake assist systems of mod-
ern trucks, one recognizes that this transfer scheme
has been implemented. The objective of the brake as-
sist is to prevent rear-end collision of vehicles travel-
ing ahead. As soon as the truck approaches another
vehicle from behind in a dangerous manner, the sys-
tem engages the brakes with up to 30% of the max-
imum pressure. If the driver agrees with this deci-
sion and steps on the brake, it is already pre-loaded
(offset), which safes time and eventually shortens the
brake distance.
-
Figure 11: Additive overlay with a differential manual
control variable
Another solution to cope with a permanent offset is
the change-over to a differential manual control vari-
able, as can be seen in Fig. 11. Since the integrator
state (between operator and plant) is marginally stable,
it cannot be distinguished from the permanent offset
of the operator’s point of view and therefore does not
change the haptics of the operating device. Left to say
that both overlay methods cannot be directly used for
bidirectional transients (e. g. automatic-to-manual-to-
automatic).
B. Contextbased manipulated variable lim-
itation
All previously described schemes provide a solution
to the bumpless transfer between automatic and man-
ual. Due to the limited reaction time of an operator,
critical situations may still occur (except for the man-
ual override control scheme) if the system destabilizes,
or, even worse, pursues the wrong control objective.
Therefore, counter measures in the form of manipu-
lated variable limitations need to be taken, which con-
fines the controller output to a certain ”save“ interval
-
Figure 12: Contextbased manipulated variable limita-
tion
and have to be combined with one of the previously in-
troduced bumpless transfer schemes. By implement-
ing a contextbased limitation, certain risks can be ana-
lyzed in advance and the limits can be dynamically ad-
justed, providing the best compromise between safety
and controller dynamics.
A simple but effective approach to scheduling the
limits is adopting them from an experienced opera-
tor. In cascaded control structures not only the final
manipulated variable but each single controller output
may be constrained.
An obvious example application is, again, an au-
tonomous experimental vehicle. On board of
AnnieWAY the lateral control algorithms have only lim-
ited access to the steering. At higher speeds the max-
imum permitted steering angle is limited to a few de-
grees, whereas at lower speeds, the full range can be
made use of for parking maneuvers etc. Hence, the
safety-driver will not be confronted with absurd steer-
ing amplitudes.
V IMPLEMENTATION AND
CONCLUSIONS
Figure 13: Application of bumpless transfer schemes
in autonomous experimental vehicle.
Bumpless MAT is a practical issue, that has gained
in relevance due to the increase in human-machine-
REM2008, Bergamo, Italy
interaction in automation, but may be easily under-
estimated. The introduced MAT schemes are derived
from classical bumpless transfer schemes between dif-
ferent control modi and solve the problem in multiple
ways. For the choice of the bumpless transfer scheme
however, no patent remedy can be offered. Instead, a
practical solution always has to take into account the
given hardware, the deployed controller, the require-
ments on the bumpless transfer, the safety concept,
and the haptics of the manual operating device of the
respective application.
For the autonomous driving application AnnieWAY, as
summarized by Table 1, override control (Override),
cross-fading (X-fading), classical anti-windup (Clas-
sical AW), bumpless replanning (Bumpl. repl.), syn-
chronization of the manual operating device (SyncMO)
and contextbased limitations on the manipulated vari-
able (lim) have very well worked out and contributed
by providing a save and convenient test driving to en-
tering the finals of the 2007 Darpa Urban Challenge.
Autonomous vehicle application
Transfer Task BT-Scheme
A⇄A Longitudinal control Override
A⇄A Lateral control X-Fading
M→A Longitudinal control Classical AW
M→A Lateral control Bumpl. repl.
A→M Longitudinal control Override+lim
A→M Lateral control SyncMO+lim
Table 1: Choice of bumpless transfer schemes in au-
tonomous test vehicle
VI ACKNOWLEDGEMENTS
The authors gratefully acknowledge the contribution
of the Transregional Collaborative Research Centre on
Cognitive Automobiles (SFB/Tr 28) granted by the
German Research Foundation (Deutsche Forschungs-
gemeinschaft, DFG).
REFERENCES
[1] K.J. ˚
Astrom and L. Rundqwist. Integrator
windup and how to avoid it. Proceedings of
the American Control Conference, 2:1693–1698,
1989.
[2] C. Edwards and I. Postlethwaite. Anti-windup
and bumpless transfer schemes. Control’96,
UKACC International Conference on (Conf.
Publ. No. 427), 1, 1996.
[3] A.H. Glattfelder and W. Schaufelberger. Sta-
bility of discrete override and cascade-limiter
single-loopcontrol systems. IEEE Transactions
on Automatic Control, 33(6):532–540,1988.
[4] A.H. Glattfelder and W. Schaufelberger. A
path from antiwindup to override control. In
IFAC Symposium on Nonlinear Control Systems,
Stuttgart, Germany, September 2004.
[5] D.N. Godbole and J. Lygeros. Longitudinal con-
trol of the lead car of a platoon. IEEE Trans-
actions on Vehicular Technology, 43(4):1125–
1135, 1994.
[6] R. Hanus, M. Kinnaert, and J.L. Henrotte. Con-
ditioning technique, a general anti-windup and
bumpless transfer method. Automatica (Journal
of IFAC), 23(6):729–739,1987.
[7] M.V. Kothare, P.J. Campo, M. Morari, and C.N.
Nett. A unified framework for the study of
anti-windup designs. Automatica, 30(12):1869–
1883, 1994.
[8] J.J. Martinez and C. Canudas-de Wit. A safe
longitudinal control for adaptive cruise control
and stop-and-go scenarios. IEEE transactions
on control systems technology, 15(2):246–258,
2007.
[9] Y. Peng, D. Vrancic, and R. Hanus. Anti-
windup, bumpless, and conditioned transfer
techniques for PID controllers. Control Systems
Magazine, IEEE, 16(4):48–57, 1996.
[10] R. Rajamani and C. Zhu. Semi-autonomous
adaptive cruise control systems. IEEE Trans-
actions on Vehicular Technology, 51(5):1186–
1192, 2002.
[11] C. Stiller, S. Kammel, B. Pitzer, J. Ziegler,
M. Werling, T. Gindele, and D. Jagszent. Team
AnnieWAY’s Autonomous System. Lecture
Notes in Computer Science, 4931:248, 2008.
[12] K.S. Walgama, S. Ronnback, and J. Sternby.
Generalisation of conditioning technique for
anti-windup compensation. IEEE Proceed-
ings on Control Theory and Applications,
139(2):109–118,1992.
[13] K. Zheng, A.H. Lee, J. Bentsman, and C.W.
Taft. Steady-state bumpless transfer under con-
troller uncertainty using the state/output feed-
back topology. IEEE Transactions on Control
Systems Technology, 14(1):3–17, 2006.
