An Innovative method for Two-Level Autonomous Emergency Braking

Algorithm Design
Hamidrezae Rezaei Nedamani, Parisa Masnadi Khiabani, Shahram Azadi
1-ADAS Project Manager at AIRIC Saipa
Abstract
Recently Advanced Driver-Assistance Systems (ADAS) have played a decisive role in the automotive
active safety system. Autonomous emergency braking (AEB) as one of the features of these
systems has been considered in this research. This study defines decision space in a two-level
braking AEB algorithm design. The purpose of the two-level braking method is first to give awareness
of collision risk by the jerk effect of partial braking and, secondly, to create enough delay for the driver
to escape collision by steering. The driving situation regarding two main indexes, the last point
to brake (LPB) and the last point to steer (LPS), is analyzed precisely. Then, the decision space is
introduced for selecting two key parameters that guarantee collision avoidance. The validation of the
developed control algorithm based on the proposed method is evaluated in different scenarios in
the computer simulation (IPG-CarMaker environment) and in field tests. The results of tests confirm
the effectiveness of the suggested method in both constant speed and accelerated motion.
Keywords: ADAS, AEB, IPG-CarMaker, decision-making algorithm, last point to brake (LPB), last
point to steer (LPS)

1
Introduction
The role of active safety systems inspires this research in preventing loss of life and property, especially
in the rear-end collisions that has a significant part in accident statistics. Active safety alerts the driver
to an imminent hazard and prevents an accident. This paper has studied autonomous emergency braking
(AEB) as one of the active safety features. Over time, the decision-making algorithms in this system
have become more complex due to computational costs and challenges. The three main factors
associated with this issue are sensors accuracy enhancement, complicated environmental conditions,
and customers’ expectations from the system performance in the critical situations. Therefore, this
paper is mainly focused on the control algorithm design, and this matter has been studied in the
reviewed researches;
References [1-7] present adaptive AEB systems based on a fuzzy controller. The general approach in
this set of references is to improve the efficiency of AEB performance. The Factors such as driver
attention (reaction time) or the braking efficiency (maximum braking acceleration, as a combination of
road and vehicle factors) have been calculated in the fuzzy logic controller. In [7-13], the relationship
between the braking force and slip speed has been investigated by considering the road conditions (such
as friction and slope) and driving behaviors. This set of papers claim that the performance and
efficiency of AEB systems have been improved significantly by the mentioned parameters. The
researches in [14-16] present a detailed estimation of time to collision (TTC), distance to collision (𝑑𝐵 )
and integration of them in the longitudinal safety control. These factors have been utilized as the basic
criteria for designing and developing the AEB system. Another category of references [17-20] have
studied the challenges of active and passive safety systems integrity and interaction. For instance,
finding the appropriate time of a seat-belt activation within AEB system action time has been discussed.
The common aspect in these articles is to minimize the negative effects of this interaction. [21-25] have
described the emergency situation identification and fast reaction to prevent a collision. In these papers,
decision-making algorithms have been designed based on artificial intelligence techniques and
reinforcement learning methods. Sensors' data and environmental/simulation experiments have been
utilized to train the algorithm about evasion policies and braking in a critical situation. In [26-30],
braking controllers for the AEB system have been proposed based on a driver braking deceleration
behavior analysis. These references have modeled TTC from a driver braking status in field/simulator
tests and extracted warning and braking strategy in one/multiple levels.
In the reviewed researches, one-level braking approach (in rare cases, two-level braking) has been
considered commonly. In addition, these studies claim that utilizing various techniques and conditions
improves the AEB decision-making algorithms performance. The scope of this study focuses on
introducing the decision space in developing AEB algorithm. This paper introduces a method for AEB
algorithm designers that guarantees longitudinal collision avoidance. It allows the algorithm designer
to select the basic parameters (consistent with the driver behavior pattern). The solution has been
derived from two fundamental concepts: the last point to steer (LPS) and the last point to brake (LPB).

2
In the present study, the two-level braking is recommended under specific conditions, in which the in-
time alert draws a subject driver's attention to avoid a collision by steering.
Basic Concept
In general, the AEB two-level braking algorithm warns a driver about the probable collision. If the
driver does not take corrective action and a collision is imminent, the AEB system will apply two-level
braking to avoid or mitigate a collision. System developers may consider different criteria in collision
avoidance algorithms. In this study, inspired by reference [31], the distance criterion has been
considered. Figure 1 represents the involved parameters and the status of the subject vehicle with
respect to the obstacle vehicle.

Figure 1 variable description and schematic of the subject and the obstacle vehicle
The proposed method applies two key concepts: the last point to brake distance (LPB) and the last point
to steer distance (LPS). As noted in [32], two types of scenarios are more apparent than the others,
which will be explained as follows;
Scenario 1: an Un-accelerated Obstacle Vehicle movement
In this scenario, the obstacle vehicle moves at a constant speed. By assuming constant velocity for the
subject vehicle, there is no acceleration in motion equations. Therefore, the stopping distance in this
scenario is defined as Eq. (1), where 𝑣𝑑𝑖𝑓𝑓 = −𝑣𝑟𝑒𝑙 , 𝜏𝐵 is the time delay for adequate braking
deceleration, which is equal to 0.05 sec based on [32], and 𝐷𝑚𝑎𝑥 is the subject vehicle's maximum
deceleration. When a driver avoids collision by steering maneuver, the required time (𝑡𝑒𝑣𝑎 ) and
distance (𝑑𝑒𝑣𝑎 ) can be defined as Eq. (2) and Eq. (3), respectively. In Eq. (2), 𝑦𝑒𝑣𝑎 is the necessary
deviation offset for evasion, 𝑎𝑦,𝑚𝑎𝑥 is the maximum lateral acceleration and 𝜏𝑠 is the steering delay
which is around 0.1 sec.
𝑣𝑑𝑖𝑓𝑓 2
𝑑𝐵 (𝑣𝑑𝑖𝑓𝑓 ) = 𝑣𝑑𝑖𝑓𝑓 . 𝜏𝐵 + (1)
2𝐷𝑚𝑎𝑥
2𝑦𝑒𝑣𝑎
𝑡𝑒𝑣𝑎 = √ + 𝜏𝑠 (2)
𝑎𝑦,𝑚𝑎𝑥
𝑑𝑒𝑣𝑎 = 𝑣𝑑𝑖𝑓𝑓 . 𝑡𝑒𝑣𝑎 (3)

3
Scenario 2: Obstacle vehicle movement at constant deceleration
In this scenario, contrary to the first scenario, the obstacle vehicle moves with a constant deceleration;
therefore, motion equations will depend on the 𝐷𝑟𝑒𝑙 . The maximum relative deceleration, the stopping
distance, and the evasion distance are defined in Eq. (4), Eq. (5), and Eq. (6), respectively.
𝐷𝑚𝑎𝑥,𝑟𝑒𝑙 = 𝐷𝑚𝑎𝑥 − 𝐷𝑜𝑏𝑠 (4)
2
𝜏𝐵 (𝑣𝑑𝑖𝑓𝑓 + 𝐷𝑟𝑒𝑙 . 𝜏𝐵 )
𝑑𝐵 (𝑣𝑑𝑖𝑓𝑓 , 𝐷𝑟𝑒𝑙 ) = (𝑣𝑑𝑖𝑓𝑓 + 𝐷𝑟𝑒𝑙 . ) 𝜏𝐵 + (5)
2 2𝐷𝑚𝑎𝑥,𝑟𝑒𝑙
𝑡𝑒𝑣𝑎 2
𝑑𝑒𝑣𝑎 (𝑣𝑑𝑖𝑓𝑓 , 𝐷𝑟𝑒𝑙 ) = 𝑣𝑑𝑖𝑓𝑓 . 𝑡𝑒𝑣𝑎 + 𝐷𝑟𝑒𝑙 . (6)
2
Main Concept
As mentioned earlier, LPB and LPS are two fundamental concepts in designing the proposed AEB
algorithm. LPB represents 𝑑𝐵 behavior in Eq. (1) and Eq. (5) and LPS describes 𝑑𝑒𝑣𝑎 behavior in Eq.
(3) and Eq. (6). The fundamental algorithm concept can be extracted from the simultaneous
representation of LPB and LPS in the same coordinate system, which is depicted in Figure 2.

Table 1 describes the region segmentation of this figure in summary. In this study, the relative velocity
at the intersection of LPB and LPS is called 𝑉𝑖𝑛𝑠 , which is a criterion for separating high-speed and
low-speed areas. As it is shown in Figure 2 and mentioned in Table 1 within zone (I), the subject vehicle
does not meet LPB and LPS lines. Therefore the driver can avoid collision either by braking or by
steering, apart from relative velocity. In the high-speed area within zone (II), the traveling vehicle
enters between LPB and LPS lines. As a result, collision avoidance could not be possible by braking,
and a driver could escape from collision only by steering maneuver.
In the low-speed area within the zone (III), the subject vehicle first passes LPS then
reaches LPB; therefore, there is no time to avoid collision with steering. The driver has to
exert maximum braking force to avoid the crash. In zone (IV), because of passing LPB and LPS
boundary, apart from relative velocity, neither the braking nor steering can prevent the accident. In this
area, applying full braking force leads to collision mitigation.

4
 Figure 2. The schematic diagram illustrates the AEB algorithm

Table 1. The four identified areas' description in Figure 3

Collision Avoidance by Collision Avoidance by Collision Mitigation by

Zone
Full Braking Steering Full Braking
I ⚫ ⚫ N/A
II  ⚫ ⚫
III ⚫  N/A
IV   ⚫
Based on the defined situation in Table 1 the simultaneous longitudinal and lateral movement is
probable. In other words, there are conditions that, besides braking, a driver has to use steering
maneuvers to avoid the collision. Hence, dynamical considerations should be measured in this
concurrency. However, it will not conflict with the proposed algorithm.
The proposed method in algorithm design
In this research, two-level braking has been utilized in the suggested AEB algorithm. The goal of two-
level braking (partial braking and full braking) is mainly to give a prognosis to the subject vehicle's and
rear vehicle's (if present) driver. In other words, partial braking and following its jerk aware a driver of
the imminent risk. Besides that, partial braking gives a driver enough time to avoid collision by steering.
However, there are fundamental questions in the design of two-level braking. What are the appropriate
time and partial braking values? This study proposes a method to define the selection space for these
parameters to answer the questions.
The outline of the proposed algorithm in the constant speed scenario can be described through the zones
defined in Figure 2. The algorithm applies partial braking in the high-speed area within zone II until

5
the traveling vehicle encounters the LPS line. This decrease in speed gives time to the driver to escape
by steering. If the driver does not react appropriately and faces the LPS line, the algorithm will employ
full braking because there is no time to avoid collision by steering. In low-speed areas within zone III,
the situation is slightly different. There is no time for the driver to escape collision by steering. The
traveling vehicle encounters with the LPS line and then with the LPB line. Therefore once the vehicle
faces LPB, the algorithm should command full braking. In the proposed method 𝑉𝑖𝑛𝑠 (Refer to Figure
2) is a boundary between high speed and low-speed areas on the relative velocity-distance plane. 𝑉𝑖𝑛𝑠
is the relative velocity in the intersection of Eq. (5) and Eq. (6) that is a nonlinear function of two
deceleration (𝐷𝑜𝑏𝑠 , 𝐷𝑠𝑢𝑏 ). 𝑉𝑖𝑛𝑠 is defined based on Eq. (9).

𝑉𝑖𝑛𝑠 = (𝑡𝑒𝑣𝑎 − 𝜏𝐵 )𝑎 ((𝐷𝑚𝑎𝑥 − 𝐷𝑜𝑏𝑠 )

(9)
∓ √((𝐷𝑚𝑎𝑥 − 𝐷𝑜𝑏𝑠 )2 + (𝐷𝑚𝑎𝑥 − 𝐷𝑜𝑏𝑠 )(𝐷𝑜𝑏𝑠 − 𝐷𝑠𝑢𝑏 )))

− (𝐷𝑜𝑏𝑠 − 𝐷𝑠𝑢𝑏 ). 𝜏𝐵
Figure 3 represents 𝑉𝑖𝑛𝑠 plot in 3-D space based on changes on 𝐷𝑜𝑏𝑠 and 𝐷𝑠𝑢𝑏 . It is obvious that 𝑉𝑖𝑛𝑠
reaches to its maximum value when the motion is non-accelerated. In other words, 𝐷𝑜𝑏𝑠 and 𝐷𝑠𝑢𝑏 can
be zero simultaneously. This situation evokes the first scenario in which an obstacle vehicle moves at
a constant speed. Larger value of 𝑉𝑖𝑛𝑠 in the non-accelerated mode respect to accelerated mode means
that the low-speed area will be a larger region. Large low-speed area increases the probability of an
initial relative velocity in this region, which leads to full braking command at a higher relative velocity.
In other words, the algorithm's behavior in this non-accelerated scenario is more conservative than the
accelerated scenario. Meanwhile, this situation can prevent collision in accelerated scenarios. In the
non-accelerated scenario, 𝑉𝑖𝑛𝑠 is defined as Eq. (10)
𝑉𝑖𝑛𝑠 = 2𝐷𝑚𝑎𝑥 (𝑡𝑒𝑣𝑎 − 𝜏𝐵 ) (10)

Figure 3. V_ins drawn in terms of 0≤D_obs≤D_max, 0≤D_sub≤D_max, D_max=μg[m/s^2],

μ=0.7, τ_B=0.05 [s] and t_eva=0.6 [s]

6
In this paper, to avoid the calculation complexity, just the non-accelerated scenario has been considered.
The algorithm's operation depends on the partial braking command time and its deceleration level.
Since the present article uses the distance criterion, the time of partial braking engagement is translated
to partial braking activation. For a deeper understanding, the partial braking behavior has been depicted
in Figure 4 and
Figure 5 within the 𝑥𝑟𝑒𝑙 − 𝑣𝑑𝑖𝑓𝑓 graph. In Table 2 the two key parameters with the titles of ∆ and 𝐷𝑝
have been introduced.
Table 2. Description of critical variables in partial braking
Variable Description Unit
∆ Distance from LPB [𝑚]
𝐷𝑝 Partial braking deceleration [𝑚/𝑠 2 ]
∆ + 𝑑𝐵 Partial braking distance index [𝑚]

In Figure 4, partial braking behavior by assuming constant ∆ and variable 𝐷𝑝 and in

Figure 5 partial braking behavior with constant 𝐷𝑝 and variable ∆ has been depicted. In each of these
two figures, the behavior imposed by the control algorithm is visible. Thus, partial braking in the high-
speed area will change to full braking as soon as the traveling vehicle encounters the LPS line, and in
the low-speed area, this will happen when it meets the LPB. In
Figure 6, the view of decision-making algorithm with different initial conditions (∆ and 𝐷𝑝 ) has been
shown. It is obvious that inappropriate initial conditions may lead to collision. An example of this issue
can be seen in the graph (1), however, in the graph (2) and (3) of
Figure 6, the relative speed and relative distance are converged to zero simultaneously, and no accident
occurs. With these descriptions, the main innovation of this research can be the selection space
introduction for ∆ and 𝐷𝑝 parameters in a way that guarantees collision avoidance.

7
 Figure 4. Schematic representation of 𝑫𝒑 changes per constant values of ∆

Figure 5. Schematic representation of ∆ changes per constant values of 𝑫𝒑

Figure 6. An overview of the decision-making algorithm behavior with two different initial
conditions
The behavior of graph (3) in
Figure 6 displays the ideal boundary trajectory. It means that if ∆ and 𝐷𝑝 are chosen appropriately,
there will be no collision based on the suggested algorithm. Eq. (11) condition should be met to
establish graph (2) in
Figure 6. Otherwise, the behavior of the ego vehicle will change to graph (1).
𝑥 𝑝 (𝑣𝑑𝑖𝑓𝑓 ) ≥ 𝑑𝐵 (𝑣𝑑𝑖𝑓𝑓 ) 𝑣𝑑𝑖𝑓𝑓 @𝑣𝑖𝑛𝑠 (11)
The behavior of graph (3) in
Figure 6 can be extracted from the Eq. (12) integrating, which is the velocity equation of the graph.
The obtained motion equation is defined in Eq. (13).
𝑖𝑛𝑖𝑡𝑖𝑎𝑙
𝑣𝑝 = 𝐷𝑝 𝑡 + 𝑣𝑑𝑖𝑓𝑓 (12)

8
 𝑖𝑛𝑖𝑡𝑖𝑎𝑙 1 𝑖𝑛𝑖𝑡𝑖𝑎𝑙
𝑥𝑝 = ∫(𝐷𝑝 𝑡 + 𝑣𝑑𝑖𝑓𝑓 ) 𝑑𝑡 = 𝐷 𝑡 2 + 𝑣𝑑𝑖𝑓𝑓 𝑡 + 𝑥𝑝𝑖𝑛𝑖𝑡𝑖𝑎𝑙 (13)
2 𝑝
𝑖𝑛𝑖𝑡𝑖𝑎𝑙 2
𝑣𝑑𝑖𝑓𝑓
Where in 𝑥𝑝𝑖𝑛𝑖𝑡𝑖𝑎𝑙 = ∆+ .
2𝐷𝑚𝑎𝑥
By replacing 𝑥𝑝𝑖𝑛𝑖𝑡𝑖𝑎𝑙 value and 𝑡 from Eq. (12), the Eq. (13) can be rewritten in the form of Eq. (14)
𝑖𝑛𝑖𝑡𝑖𝑎𝑙2 𝑖𝑛𝑖𝑡𝑖𝑎𝑙 2
𝑖𝑛𝑖𝑡𝑖𝑎𝑙
1 𝑣𝑝 − 𝑣𝑑𝑖𝑓𝑓 𝑖𝑛𝑖𝑡𝑖𝑎𝑙
𝑣𝑝 − 𝑣𝑑𝑖𝑓𝑓 𝑣𝑑𝑖𝑓𝑓
𝑥𝑝 (𝑣𝑝 ) = 𝐷𝑝 ( ) + 𝑣𝑑𝑖𝑓𝑓 ( )+∆+ (14)
2 𝐷𝑝 𝐷𝑝 2𝐷𝑚𝑎𝑥
Finally, partial braking behavior with respect to ∆ and 𝐷𝑝 in 𝑥𝑟𝑒𝑙 − 𝑣𝑑𝑖𝑓𝑓 plane has been described in
Eq. (15)
1 1 1 𝑖𝑛𝑖𝑡𝑖𝑎𝑙 2
𝑥 𝑝 (𝑣𝑑𝑖𝑓𝑓 ) = 𝑣𝑑𝑖𝑓𝑓 2 + ( − ) 𝑣𝑑𝑖𝑓𝑓 +∆ (15)
2𝐷𝑝 2𝐷𝑚𝑎𝑥 2𝐷𝑝
Two critical parameters ∆ and 𝐷𝑝 values should be defined in a way that the partial braking behavior
in the suggested algorithm performs a similar behavior in the graph (2) or (3) in
Figure 6. In these two graphs 𝑥 𝑝 (𝑣𝑑𝑖𝑓𝑓 ) is guided in a manner that encounters the LPB line in the low-
speed zone, and collision avoidance is guaranteed. By looking at Eq. (11) and after simplification, Eq.
(15) can be obtained. It is assumed that the maximum partial braking deceleration is 60% of the
maximum deceleration (𝐷𝑝 = 0.6 𝐷𝑚𝑎𝑥 ) [32], therefore the 𝐷𝑝 boundary is defined.
𝑖𝑛𝑖𝑡𝑖𝑎𝑙2
(𝑣𝑑𝑖𝑓𝑓 − 𝑣𝑖𝑛𝑠 2 ) 𝐷𝑚𝑎𝑥
≤ 𝐷𝑝 ≤ 𝐷𝑝,𝑚𝑎𝑥 (15)
𝑖𝑛𝑖𝑡𝑖𝑎𝑙 2
(𝑣𝑑𝑖𝑓𝑓 − 𝑣𝑖𝑛𝑠 2 ) − 2𝑣𝑖𝑛𝑠 𝜏𝐵 𝐷𝑚𝑎𝑥 + 2∆𝐷𝑚𝑎𝑥
Considering the velocity limit defined in the Euro NCAP standard for AEB test, the relative velocity at
𝑖𝑛𝑖𝑡𝑖𝑎𝑙
the initial moment is considered between (𝑣𝑑𝑖𝑓𝑓 ∈ [10 − 80]). In Eq. (15), determining the boundary
𝑘𝑚
for ∆ will lead to a better drawing. If the radar range is considered 200 m, with 80 ℎ initial speed and
maximum deceleration, the stopping distance will be around 30 m. As a result, the maximum ∆ can be
extracted according to Eq. (16)
𝑖𝑛𝑖𝑡𝑖𝑎𝑙 2
𝑣𝑑𝑖𝑓𝑓 ⁄
( 3.6)
2
(80⁄3.6)
∆𝑚𝑎𝑥 = RADAR 𝑟𝑎𝑛𝑔𝑒 − = 200 − ( ) ≈ 170 (16)
2(𝜇𝑔) 2(0.85 × 9.81)
( )
Hence, generally, its range is considered ( ∆∈ [10 − 170]).
Based on the purpose of this study, Eq. (15) represents the decision space for a designer to select the
key parameters in the AEB decision-making algorithm. To better understand Eq. (15), Figure 7
illustrates a schematic of the section space in the algorithm design. In this figure 𝐷𝑝 partial braking
𝑖𝑛𝑖𝑡𝑖𝑎𝑙
surface with respect to (∆ & 𝑣𝑑𝑖𝑓𝑓 ) changes has been depicted. The limitation and boundaries on

9
 𝑖𝑛𝑖𝑡𝑖𝑎𝑙
𝑣𝑑𝑖𝑓𝑓 , ∆, 𝐷𝑝 are clearly visible (red cube). The upper part of the cube that is separated by a surface
(green volume) is a parameters' (∆ and 𝐷𝑝 ) selection space, which ensure collision avoidance.

Figure 7. Display the key parameters selection in the 𝑫𝒑 drawing for 𝒗𝒊𝒏𝒔 ≤ 𝒗𝒊𝒏𝒊𝒕𝒊𝒂𝒍
𝒅𝒊𝒇𝒇 ≤ 𝟖𝟎, 𝟏𝟎 ≤
𝒎
∆≤ 𝟏𝟕𝟎, 𝑫𝒎𝒂𝒙 = 𝝁𝒈 [ 𝟐 ] , 𝝁 = 𝟎. 𝟖𝟓, 𝝉𝑩 = 𝟎. 𝟎𝟓 [𝒔], 𝒕𝒆𝒗𝒂 = 𝟎. 𝟔 [𝒔]
𝒔
This research is mainly focused on guiding the AEB algorithm designer to design a two-level braking
algorithm. With this approach, ∆ parameter has been utilized as a criterion to determine the driving
mode (aggressive, normal, or conservative driving mode). This category choice requires detailed driver
behavioral studies, which is out of the scope of current research. Therefore, just a general concept of
this categorization has been provided.
Small values of ∆ parameter (distance criterion) are equivalent to aggressive driving. Indeed, drivers
with aggressive behavior with higher risk tend to brake later and in closer distances to the obstacle
vehicle as much as possible and brake with higher intensity. This tendency is quite in contrast with a
conservative type of driving. They tend to keep a higher distance from obstacle vehicles and avoid
potential hazards with lower braking intensity. According to this, the AEB algorithm designer can
design the system based on the determined driving mode.
Realization of the proposed algorithm
Some assumptions have been considered for algorithm realization. In collision avoidance by steering,
which requires lane-change maneuver, the sideline of the ego vehicle assumes to be free of any obstacle.
In addition, there is no vehicle behind the subject vehicle in partial and full braking, which lead to
another accident.
Before considering the AEB decision-making flowchart, it is necessary to consider one of the probable
scenarios that may disrupt decision algorithm implementation. There are situations where the absolute

10
velocity of both vehicles is non-zero and equal, so the relative velocity is close to zero. From the TTC
calculation point of view, there is no danger. However, if the relative distance of two vehicles is
considerably small, this scenario will have a high accident potential. Once the obstacle vehicle brakes
slightly, the situation becomes destabilized, and the collision probability increases significantly.
Therefore, the decision algorithm should consider the time headway criterion and time to the collision
to prevent collision in these situations.
In Figure 8 the AEB algorithm pre-activation condition flowchart has been depicted based on [33].
Some conditions should be considered in the implementation phase at the vehicle level before activating
the system.

Figure 8 Precondition flowchart for AEB activation

Figure 9 shows the hazard levels in the two-level braking algorithm. As seen in this figure, the decision-
making process must be considered from a safe level with zero hazards to the most dangerous level.
Figure 10 describes the rules that the algorithm switches between the different situations based on these
safety levels. The values of warning level, braking level, and required decelerations should be assigned
in each state. In the safe state, all the three mentioned variables equal to zero. The values assigned in
the warning state are similar to the safe state, except that the warning level variable equals one. Two
other states are partial braking and full braking states. There is a common situation in these two states
that braking must not take place. When the ego vehicle overtakes the obstacle vehicle in a two-way
road and encounters an obstacle vehicle in the opposite direction, the AEB algorithm should not be
𝑣
activated. The occurrence of (1 − 𝑣𝑑𝑖𝑓𝑓 ) ≥ 0 in the partial braking and full braking states confirms this
𝑠𝑢𝑏
issue. In other states, this condition is used to release braking after a complete vehicle stop.

11
 Figure 9. Warning levels definition in AEB algorithm

Figure 10. The proposed algorithm flowchart in AEB active mode

Computer Simulation
This study has used V model-based development, one of the standard approaches in automotive
software development [34]. A co-simulation platform has been utilized to create a test environment and
decision-making algorithm based on the model in the loop (MIL). The MIL simulation facilitates

12
building the specific behavior of the control algorithm that will be implemented in C code. This
situation eases any modification or debugging during the development phase. Test vehicles dynamics,
sensor simulation, and test scenarios are simulated in IPG-CarMaker [35], while the decision-making
algorithm is implemented in MATLAB/Simulink and State flow[36, 37]. MATLAB/State Flow
provides concise and clear descriptions of complex systems using the theory of finite state machines,
flow diagrams, and state transfer diagrams. IPG-CarMaker is a well-known commercial tool and virtual
test drive which facilitates the development process from the early conceptual stage to rapid prototype
testing [35]. It provides SIL operation to HIL tests. In this paper, test scenarios based on NCAP
assessments and local traffic situations are designed in the road interface of CarMaker. In addition, the
Demo Renault Megan model is selected for the car model. It is noteworthy that all the driving test
procedures have been implemented through an office driving simulator, so a predefined "driver model"
has not been used. During the aforementioned tests, only the cruise control is activated. The utilized
radar sensor's parameters are described below:
Table 3. RADAR specifications in IPG-CarMaker software

Radar Operating Frequency Band +76...77 GHz

Azimuth Angle Augmentation -9.0°...+9.0° far field
Sensitivity (min. RCS@x m) 10 𝑚2 @250m far range,
Cycle time app. 60 ms

Figure 11. View of IPG-CarMaker software

Computer simulation has been done to validate the proposed design method and algorithm of the two-
level AEB system. The simulation tests are designed with two approaches based on Euro NCAP tests
and operational tests resulting from innovative ideas presented in Figure 7. In Table 4, the results of
the first series of tests have been represented. The tests have been categorized into three different
scenarios, which are CCRs, CCRm, and CCRb. The table has shown the tests' initial conditions, key
parameters values (∆ and 𝐷𝑝 ) and the minimum value of the relative distance after AEB algorithm

13
activation. The tests validate the algorithm performance for the selected parameters in accordance with
Eq. (15) in the selected Euro NCAP scenarios. As it can be seen in all tests, there is a non-zero and
positive relative distance with zero relative velocity. This means that the proposed algorithm prevents
collision. Figure 12 shows the braking deceleration, relative speed, and relative distance changes in one
of the mentioned tests. The traveling vehicle's behavior in the 𝑥𝑟𝑒𝑙 − 𝑣𝑑𝑖𝑓𝑓 graph has been depicted in
Figure 13. In this figure, the starting point of partial braking and full braking has been shown. The
similarity between this graph and the ideal graph (3) in
Figure 6 validates the proposed two-level braking algorithm design method. It is worth mentioning that
𝑘𝑚
in Table 4, because of 𝑣𝑖𝑛𝑠 ≈ 35 ( ℎ ), the initial relative velocities in the tests are considered faster
𝑘𝑚
than 40 ( ℎ ).
In Table 5, the results of the second set of tests have been indicated. As briefly mentioned, in the design
of this series of tests, ∆ parameter is considered a criterion for determining the driving mode. For
aggressive mode ∆ is equal to 15 [m], and for normal and conservative mode is determined 40 [m] and
90 [m] respectively. It should be noted that these numbers are only representative of driving mode and
their use in the design of tests in Table 5 is merely a confirmation of the proposed method. Defining
these numbers requires precise behavioral studies and investigations that are beyond the scope of this
study. As can be seen in the results of this table by stabilizing the ∆ value at different relative velocities,
the minimum distance after two-level braking indicates no collision.
Table 4. Euro NCAP tests' results
[𝑘𝑚]
𝑖𝑛𝑖𝑡𝑖𝑎𝑙 [ ]

[𝑘𝑚]

[𝑘𝑚]

[𝑚 ]
𝑚
Scenarios

[𝑚 ]

𝑠2
ℎ

ℎ
∆ 𝑚]

𝑠2

Braking

Braking

𝑥𝑟𝑒𝑙 𝑚]
𝑖𝑛𝑖𝑡𝑖𝑎𝑙

Partial
[

𝐷𝑜𝑏𝑠
𝑣𝑑𝑖𝑓𝑓

[
𝑣𝑠𝑢𝑏

𝑣𝑜𝑏𝑠
𝐷𝑝

Min
Full
𝑥𝑟𝑒𝑙

_ 10 _ _ 10 0 _ _ _ _
13 12 11 10 9 8 7 6 5 4 3 2 1

_ 20 _ _ 20 0 _ _ _ _
_ 30 _ _ 30 0 _ _ _ _
200 40 15 1.2 40 0 0   0.11
200 50 25 1.8 50 0 0   0.17
200 60 50 1.6 60 0 0   0.18
CCRs

200 70 80 1.7 70 0 0   0.13

200 80 20 4.6 80 0 0   0.50
_ 10 _ _ 30 20 _ _ _ _
_ 20 _ _ 40 20 _ _ _ _
_ _ _ _ _ _ _
CCRm

30 50 20
200 40 10 1.65 60 20 0   0.29
200 50 40 1.18 70 20 0   0.12

14
  
19 18 17 16 15 14

200 60 60 1.36 80 20 0 0.16

200 60 60 2 80 20 0   0.11
12 0 _ _ 50 50 2   0.01
12 0 _ _ 50 50 6   1.13
CCRb

40 _ _ _ 50 50 2   0.22
40 _ _ _ 50 50 6   0.25

Figure 12. The simulation test result related to row 7 of Table 4

15
Figure 13. The vehicle behavior on the 𝒙𝒓𝒆𝒍 − 𝒗𝒅𝒊𝒇𝒇 page in the simulation test related to row 7
of Table 4

Table 5. Validation results of the proposed method by determining ∆ parameter as a criterion

for the driving mode
[𝑘𝑚]

[𝑘𝑚]

[𝑘𝑚]
ℎ

[𝑚 ]
[𝑚 ]

𝑠2

Braking

Braking
ℎ

𝑥𝑟𝑒𝑙 𝑚]
∆ 𝑚]
Driving

𝑠2

Partial
Mode

𝑖𝑛𝑖𝑡𝑖𝑎𝑙

[
[

𝐷𝑜𝑏𝑠

Min
𝑣𝑑𝑖𝑓𝑓

Full
𝑣𝑠𝑢𝑏

𝑣𝑜𝑏𝑠
𝐷𝑝

40 15 1.2 40 0 0   0.11
 
Aggressive mode

50 15 2.6 50 0 0 0.37
60 15 3.7 60 0 0   0.44
70 15 4.6 70 0 0   0.52
80 15 5.2 80 0 0   0.47
35 15 0.3 55 20 0   0.17
45 15 1.9 65 20 0   0.35
55 15 3.2 75 20 0   0.40
40 40 0.5 40 0 0   0.07
mo
No
rm

de
al

50 40 1.2 50 0 0   0.19

16
 60 40 1.9 60 0 0   0.29
70 40 2.6 70 0 0   0.27
80 40 3.2 80 0 0   0.28
35 40 0.1 55 20 0   0.06
45 40 0.8 65 20 0   0.12
55 40 1.5 75 20 0   0.30
40 90 0.21 40 0 0   0.04
Conservative mode

50 90 0.57 50 0 0   0.07
60 90 0.96 60 0 0   0.08
70 90 1.37 70 0 0   0.11
80 90 1.80 80 0 0   0.17
35 90 0.06 55 20 0   0.11
45 90 0.38 65 20 0   0.01
55 90 0.76 75 20 0   0.1711
Experimental Results
HIL preparation
The XPack4 is a real-time system with remote desktop capabilities. Its operating system is Xeno, and
it provides digital /analog input/output cards, CAN communication, and Flex ray. The inputs/outputs
of the designed system are determined for this hardware. All necessary data from CAN bus are defined
in the form of a data dictionary through a DBC file. This rapid prototype hardware allows testing and
validation of implemented algorithms.
Field Tests
The main purpose of these tests is to validate the theory presented in the decision algorithm. Due to the
constraints and limitations of the real world in these series of tests, some simplifications have been
considered with respect to the designed tests. It is worth noting that simplifications do not lead to the
loss of the expected core performance. Figure 14 shows some views of the field test steps.

17
 Figure 14. Views of field tests
The important point in designing field tests is that the physical constraints will impose restrictions on
the field tests performance, which makes them different from the simulation environment tests. In
addition, studying these numbers in the real environment requires behavioral studies and several other
factors that are not in the scope of this research. With these descriptions, the parameters initial
estimations have been utilized for validation, which has been proposed in accordance with the theory.
The coefficient of friction in the test ground is considered to be μ≈0.6, the braking time delay 𝜏𝐵 ≈0.4
[sec] and the escape time 𝑡𝑒𝑣𝑎 ≈0.6 [sec]. Corresponding to the decision space of Figure 7, by taking
these estimated numbers into account, Figure 15 is drawn.

Figure 15. Display the key parameters' selection in the 𝑫𝒑 drawing for 𝒗𝒊𝒏𝒔 ≤ 𝒗𝒊𝒏𝒊𝒕𝒊𝒂𝒍
𝒅𝒊𝒇𝒇 ≤ 𝟐𝟎, 𝟓 ≤
𝒎
∆≤ 𝟐𝟎, 𝑫𝒎𝒂𝒙 = 𝝁𝒈 [𝒔𝟐] , 𝝁 = 𝟎. 𝟔, 𝝉𝑩 = 𝟎. 𝟒 [𝒔], 𝒕𝒆𝒗𝒂 = 𝟎. 𝟔 [𝒔]

18
In this research, many tests have been performed to validate the proposed AEB algorithm with two
levels of braking. Figure 16 shows the results for one of the field tests. In this figure, the represented
features are; A) Relative distance, speed, and acceleration with constant Dummy obstacle, b) Relative
distance-relative velocity plate, c) AEB actuator activation signal, and normalized acceleration of the
vehicle under test. As mentioned earlier, each of the tests’ ∆ is considered a criterion for determining
the driving mode. ∆ is determined 10 [m] for aggressive driving mode, 11 [m] for normal driving mode,
and 13 [m] for the conservative driving mode. The rational of these values stems from the fact that in
response to the ∆ lower values (shorter distance), the system should choose a higher partial braking
level to avoid the collision. According to the determined value of ∆ and the initial relative velocity, the
partial deceleration is obtained from the selection space of Figure 15. When there is a potential danger
and the braking starts, relative distance and velocity are gradually reduced, which can be seen in the
first part of the figure (part A). The third diagram of portion A represents the two-level braking
command. Part B validates the ideal diagram 3 in
Figure 6 and Figure 13. Due to the field tests' limitation, there is a small deviation from ideal diagrams;
however, the graph in this part still shows the desired behavior. The last graph also illustrates the delay
of the braking acceleration system with the AEB operator's activation signal. The results validate the
theoretical discussions and the key parameters' selection for collision avoidance.

19
Figure 16. Field test results, ∆ = 10 and initial relative speed 40 [km / h]

20
Conclusion
This research structurally provides a way to design two-level automatic braking algorithms. The
decision-making algorithm has been extracted and analyzed step-by-step, and the key parameters
selection space for two-level AEB has been defined. This method guarantees collision avoidance along
with stabilizing the driving mode. The results of the computer simulation and field tests have shown
that the proposed method can avoid collisions with two levels of braking in different scenarios.
Acknowledgment

21
Reference
[1] V. Milanés, J. Pérez, J. Godoy, and E. Onieva, "A fuzzy aid rear-end collision warning/avoidance
system," Expert Systems with Applications, vol. 39, no. 10, pp. 9097-9107, 2012.
[2] D. F. Llorca et al., "Autonomous pedestrian collision avoidance using a fuzzy steering controller,"
IEEE transactions on intelligent transportation systems, vol. 12, no. 2, pp. 390-401, 2011.
[3] L. G. Alarcon, M. E. V. Recalde, M. Marcano, and E. Marti, "Adaptable emergency braking based
on a fuzzy controller and a predictive model," in 2018 IEEE International Conference on Vehicular
Electronics and Safety (ICVES), 2018: IEEE, pp. 1-6.
[4] X. Xiong, M. Wang, Y. Cai, L. Chen, H. Farah, and M. Hagenzieker, "A forward collision
avoidance algorithm based on driver braking behavior," Accident Analysis & Prevention, vol. 129,
pp. 30-43, 2019.
[5] Z. Lei and S. Qin, "Research on Forward Collision Warning System and Fuzzy Control of
Automatic Emergency Braking System," SAE Technical Paper, 0148-7191, 2020.
[6] L. Gonzalez, J. A. Matute-Peaspan, J. P. Rastelli, and I. Calvo, "Longitudinal Collision Avoidance
Based on Model Predictive Controllers and Fuzzy Inference Systems," in 2020 IEEE 23rd
International Conference on Intelligent Transportation Systems (ITSC), 2020: IEEE, pp. 1-6.
[7] J. Herzfeld, S. Thottathodhi, M. Jonasson, L. S. Muppirisetty, S. Roychowdhury, and J. Sjöberg,
"Collision Avoidance by Utilizing Dynamic Road Friction Information," in 2020 54th Asilomar
Conference on Signals, Systems, and Computers, 2020: IEEE, pp. 1425-1429.
[8] T. Chen, K. Liu, Z. Wang, G. Deng, and B. Chen, "Vehicle forward collision warning algorithm
based on road friction," Transportation research part D: transport and environment, vol. 66, pp.
49-57, 2019.
[9] I.-C. Han, B.-C. Luan, and F.-C. Hsieh, "Development of autonomous emergency braking control
system based on road friction," in 2014 IEEE International Conference on Automation Science and
Engineering (CASE), 2014: IEEE, pp. 933-937.
[10] S. Jeon, G. Kim, and B. Kim, "Study on V2V-based AEB system performance analysis in
various road conditions at an intersection," International Journal of Software Engineering and Its
Applications, vol. 9, no. 7, pp. 1-10, 2015.
[11] H. Mun, G. Kim, and B. Kim, "AEB System for a curved road considering V2V-based road
surface conditions," Ubiquitous Science and Engineering, vol. 86, pp. p8-13, 2015.
[12] H. Kim, K. Shin, I. Chang, and K. Huh, "Autonomous emergency braking considering road
slope and friction coefficient," International Journal of Automotive Technology, vol. 19, no. 6, pp.
1013-1022, 2018.
[13] M. Lin, J. Yoon, and B. Kim, "Proposal and validation of AEB system algorithm for various
slope environments," in Advanced Multimedia and Ubiquitous Engineering: Springer, 2016, pp.
241-246.

22
[14] T. Lee, T. Kim, B. Kim, K. Yi, and J. Lee, "Advanced braking algorithm for robust longitudinal
risk management," Proceedings of the Institution of Mechanical Engineers, Part D: Journal of
Automobile Engineering, vol. 230, no. 11, pp. 1488-1503, 2016.
[15] D. Lee, S. Kim, C. Kim, and K. Huh, "Development of an autonomous braking system using
the predicted stopping distance," International Journal of Automotive Technology, vol. 15, no. 2,
pp. 341-346, 2014.
[16] K. D. Kusano and H. Gabler, "Method for estimating time to collision at braking in real-world,
lead vehicle stopped rear-end crashes for use in pre-crash system design," SAE International
Journal of Passenger Cars-Mechanical Systems, vol. 4, no. 2011-01-0576, pp. 435-443, 2011.
[17] E.-S. Kim, S.-K. Min, D.-H. Sung, S.-M. Lee, and C.-B. Hong, "The AEB System with Active
and Passive Safety Integration for Reducing Occupants' Injuries in High-Velocity Region," in 24th
International Technical Conference on the Enhanced Safety of Vehicles (ESV) National Highway
Traffic Safety Administration, 2015, no. 15-0335.
[18] M. Edwards, A. Nathanson, J. Carroll, M. Wisch, O. Zander, and N. Lubbe, "Assessment of
integrated pedestrian protection systems with autonomous emergency braking (AEB) and passive
safety components," Traffic injury prevention, vol. 16, no. sup1, pp. S2-S11, 2015.
[19] J. K. Lee, H. J. Chu, and K. R. Hurh, "A development of the CAE process for the AEB-occupant
integrated safety system," in 25th International Technical Conference on the Enhanced Safety of
Vehicles (ESV) National Highway Traffic Safety Administration, 2017.
[20] P. Puthan, P. Thalya, and N. Lubbe, "Active and passive safety passenger car technologies:
Potentials to save lives in India," in Proceedings of IRCOBI Asia Conference, 2018.
[21] G. Castania, "Artificial Intelligence Controls for Vehicle Emergency Maneuvering,"
Politecnico di Torino, 2020.
[22] H. Porav and P. Newman, "Imminent collision mitigation with reinforcement learning and
vision," in 2018 21st International Conference on Intelligent Transportation Systems (ITSC), 2018:
IEEE, pp. 958-964.
[23] M. Kim, S. Lee, J. Lim, J. Choi, and S. G. Kang, "Unexpected collision avoidance driving
strategy using deep reinforcement learning," IEEE Access, vol. 8, pp. 17243-17252, 2020.
[24] H. Chae, C. M. Kang, B. Kim, J. Kim, C. C. Chung, and J. W. Choi, "Autonomous braking
system via deep reinforcement learning," in 2017 IEEE 20th International conference on intelligent
transportation systems (ITSC), 2017: IEEE, pp. 1-6.
[25] C. Arvind and J. Senthilnath, "Autonomous RL: Autonomous vehicle obstacle avoidance in a
dynamic environment using MLP-SARSA reinforcement learning," in 2019 IEEE 5th International
Conference on Mechatronics System and Robots (ICMSR), 2019: IEEE, pp. 120-124.
[26] J. Duan et al., "Driver braking behavior analysis to improve autonomous emergency braking
systems in typical Chinese vehicle-bicycle conflicts," Accident Analysis & Prevention, vol. 108,
pp. 74-82, 2017.

23
[27] Z. Han, X. Li, and J. Wang, "Longitudinal collision warning system based on driver braking
characteristics," in 2020 3rd International Conference on Algorithms, Computing and Artificial
Intelligence, 2020, pp. 1-7.
[28] N. Lubbe, "Brake reactions of distracted drivers to pedestrian Forward Collision Warning
systems," Journal of safety research, vol. 61, pp. 23-32, 2017.
[29] C.-N. Boda, Driver interaction with vulnerable road users: Understanding and modelling
driver behaviour for the design and evaluation of intelligent safety systems. Chalmers Tekniska
Hogskola (Sweden), 2017.
[30] R. Chen, "Driver behavior in car following-the implications for forward collision avoidance,"
Virginia Tech, 2016.
[31] D. S. Jeon, S. K. Min, and D. H. Sung, "Apparatus and method for preventing vehicle collision,"
ed: Google Patents, 2016.
[32] H. Winner, "Fundamentals of Collision Protection Systems," in Handbook of Driver Assistance
Systems: Basic Information, Components and Systems for Active Safety and Comfort, H. Winner,
S. Hakuli, F. Lotz, and C. Singer Eds. Cham: Springer International Publishing, 2016, pp. 1149-
1176.
[33] Intelligent transport systems ISO, ISO (the International Organization for Standardization),
2013-07 2013.
[34] V. Socci, "Implementing a model-based design and test workflow," in 2015 IEEE International
Symposium on Systems Engineering (ISSE), 2015: IEEE, pp. 130-134.
[35] A. Arcidiacono, "ADAS virtual validation: ACC and AEB case study with IPG CarMaker,"
Politecnico di Torino, 2018.
[36] R. Colgren, Basic MATLAB®, Simulink®, and Stateflow®. American Institute of Aeronautics
and Astronautics, 2006.
[37] J. Friedman, "MATLAB/Simulink for automotive systems design," in Proceedings of the
Design Automation & Test in Europe Conference, 2006, vol. 1: IEEE, pp. 1-2.
[38] H. Winner, B. Danner, and J. Steinle, "Adaptive cruise control," in Handbuch
Fahrerassistenzsysteme: Springer, 2009, pp. 478-521.

24