Professional Documents
Culture Documents
During the last few years, a lot of results of the lane keeping assist (LKA) systems have been
reported and a few manufacturers applied LKA systems to commercial vehicles. These systems,
however, only work in limited situations when the vehicle departs from the lane, by introducing a
gentle steering torque to help the driver keep the vehicle in the lane. In this paper, a lane guidance
control (LGC) system, which is more active than the LKA, is designed in three of way and
investigated via computer simulations and real vehicle tests. The performance of the proposed
LGC systems are analyzed and compared to each other.
Topics / Vehicle Dynamics, Modeling and Simulation, Autonomous Vehicle, Advanced Driver Assistance System
As mentioned in the section 1, the three different a wheel base, and lf and lr are the distance from CG to
LGC methodologies are designed and investigated in front and rear of the vehicle, respectively. m is the mass
this research. and ay is a lateral acceleration of the vehicle.
control input as follows: compliance of the vehicle, which is induced by side slip
tf angle, is considered to improve the control performance
J { X QX u Ru}dt
T T
(8)
as follows:
0
where Q and R are weigh matrices that penalize states or des , K des ,com Vx
inputs. By finding a proper KLQR that minimizes the (12)
given quadratic cost function, the optimal control input where, des ,com des f
to follow the desired path is yielded as follows: Finally, the steering command, K , is yielded to
u LQR K LQR X (9) make the desired yaw motion of the vehicle as follows:
1
coordinates of A and B in the triangle ABO. The
coordinates of point A and B are calculated by using the 0
lateral offset, heading angle, and radius of road
curvature obtained from camera sensor as follows: -1
road ( L )
2 2
x L2 y 2 , y (10) -2
2(1 road )
0 10 20 30 40 50
Time [sec]
where road=1/Rroad. L is a target distance and is the Fig. 6 Curvature of actual road
lateral offset. The radius of the desired turning circle is
also calculated by using trigonometric functions in the
triangle ABO` as follows: Figure 7 shows the simulation results of the
2 proposed LGC systems in the similar situations to actual
des 1/ Rdes 2 ( y cos x sin ) road environment. The vehicle speed was set to 70 km/h,
L and the vehicle model was set as a sedan. The lateral
(11)
x L sin , y L cos offset, heading angle, steering wheel angle, and yaw
where,
Rdes cos L / 2
rate are shown in Figure 7, respectively.
0.2
LQR
0.1 Kinematics model-based LGC strategy shows the best performance
0 among the three. The kinematics-based LGC strategy
-0.1 shows the equal or a bit better performance than the
-0.2 model-based LQR method. Although the predictive
model-based LGC strategy shows the best performance,
0 10 20 30 40 50
Time [sec]
it requires precise vehicle parameters and many tuning
(b) Heading angle parameters. This makes it practically difficult for the
engineers to develop such system. The model-based
5
LQR controller has the same drawbacks with the
Predictive predictive model-based control strategy. The kinematics
Steering angle [deg]
LQR
-based LGC strategy, however, is the simplest and
Kinematics
0
practically useful among the three as it requires only
four tuning parameters without any precise knowledge
of vehicle dynamics parameters. Furthermore, the
performance of the kinematics-based LGC strategy is
-5
0 10 20 30 40 50 also within the acceptable range.
Time [sec]
(c) Steering wheel angle 4. VEHICLE TEST RESULTS
2
Predictive The real vehicle tests have been also performed
LQR under the same road conditions as shown in Figure 5
Yaw rate [deg/s]
1
Kinematics and Figure 6. The test vehicle is comprised of the
0 camera sensor, MDPS, human-machine interface (HMI),
-1
and rapid control prototyping (RCP) equipment. Figure
8 shows the configuration of the test vehicle and its
-2 components. The camera sensor for detection of the lane
0 10 20 30 40 50
Time [sec] is located at the top of the windshield. The MDPS
(d) Yaw rate module, which makes it possible to control the steering,
is equipped with test vehicle. The HMI informs whether
Fig. 7 Simulation results of the LGC the lane detection, and the status of control system to
the driver. The control algorithms are implemented in
Figure 7-(a) and (b) shows the time-histories of the the RCP equipment, ‘Micro AutoBox’. All of data from
lateral offset from the road center line, and heading each component are transferred or received by CAN
angle relative to the lane, respectively. All of the lateral communications.
offset show under 0.1 m, and the heading angles are also
maintained 0.1 deg, that is, all three LGC control
strategies show good performances for path following
as shown in Figure 7. Among them, especially, the
predictive model-based one shows the best performance
in the view point of the lateral offset and heading angle.
Table 1 summarizes the simulation results and the
number of tuning parameters required to design the
controllers. The differences of the steering wheel angle
and yaw rate are small enough to be neglected as shown
in Figure 7-(c) and (d).
Fig. 8 Configuration of the test vehicle
60
too many tuning parameters, and it is too hard to tune by
40 developer as results. In contrast, it is easy to tune in the
Predictive
case of the kinematics-based LGC strategy, because of
20
LQR small and intuitive tuning parameters. The kinematics-
0
Kinematics based LGC, therefore, can be guaranteed consistent
0 10 20 30 40 50 performance.
Time [sec]
(a) Vehicle speed Table 2. Summary of vehicle test results
(1) (2) (3)
2
Lateral offset
0.34 0.19 0.11
RMS [m]
Lateral offset [m]
1
error Heading angle
0.22 0.32 0.20
0 [deg]
0.5
model-based control, (2) Model-based LQR control, (3)
0 Kinematics-based control. The performance of designed
Predictive
control strategies were investigated and analyzed via
-0.5
LQR computer simulations and vehicle tests. The simulations
-1
Kinematics were conducted with road model, which is modeled
0 10 20 30 40 50 based on the actual road. The vehicle tests were also
Time [sec]
performed under same road.
(c) Heading angle From simulation and vehicle test results, it is shown
that all of proposed control strategies are good enough
AVEC ’12
REFERENCES