Assessing he Sa e y Impac s o Coope a i e
Awa eness o Au oma ed D i ing
1s Ma cel Sonn ag
Ins i u e o Au omo i e Enginee ing (ika)
RWTH Aachen Uni e si y
Aachen, Ge many
[email p o ec ed]
2nd P amod Mallipudi
RWTH Aachen Uni e si y
Aachen, Ge many
[email p o ec ed]
3 d Lu z Ecks ein
Ins i u e o Au omo i e Enginee ing (ika)
RWTH Aachen Uni e si y
Aachen, Ge many
[email p o ec ed]
Abs ac —Connec ed and au oma ed d i ing is expec ed o
inc ease oad sa e y. Sa e y impac assessmen is an app oach
o p ospec i ely assess he po en ial sa e y bene i s o sys ems
be o e hei in oduc ion o ma ke . As hese assessmen s usually
le e age model-in- he-loop simula ions, i is equi ed o model
he in ol ed oad use s’ beha io ealis ically. Cu en esea ch
on p ospec i e sa e y impac assessmen mainly ocusses on
au oma ed d i ing unc ions (ADFs) no aking ehicle com-
munica ion in o accoun . When assessing ADFs inco po a ing
collec i e awa eness o o he oad use s based on ehicle
communica ion, he e ec s o he communica ion a e o be
modeled in addi ion. Coope a i e awa eness aims a o e coming
pe cep ion sho comings o ADFs due o limi ed senso anges
o isual obs uc ions, as oad use s communica e hei own
dynamic s a e. This wo k p esen s a me hodology o p ospec i e
sa e y impac assessmen o coope a i e awa eness-enabled ADFs.
The me hodology includes coope a i e awa eness da a analyses,
gene a ing ealis ic baseline scena ios, as well as modelling he
accu acies o he communica ed in o ma ion. The de eloped
me hodology is e alua ed o ehicle- o- ehicle communica ion
o ETSI Coope a i e Awa eness Messages (CAM) based on
eal-wo ld da a om he V2AIX da ase . This use case is
assessed o di e en ADFs a u ban in e sec ions a ec ed by
isual obs uc ions. The esul s show an inc ease in he c ash
occu ence o ealis ic CAM modelling compa ed o ideal CAMs.
This implies ha assessing coope a i e awa eness in simula ions
equi es modelling i ca e ully o achie e eliable esul s.
Index Te ms—Connec ed and Au oma ed D i ing, Sa e y Im-
pac Assessmen , Coope a i e Awa eness, V2X
I. INTRODUCTION
Connec ed and au oma ed d i ing unc ions a e expec ed o
inc ease he sa e y o oad a ic. Achie ing his sa e y gain
equi es ca e ul design o d i ing unc ion beha io s. P ospec-
i e sa e y impac assessmen usually le e ages model-in- he-
loop simula ions o assess he sa e y po en ials o d i ing
unc ions be o e in oduc ion o ma ke [1]. Usually ma u e
sys ems a e assumed as his is equi ed o be in oduced o
he ma ke . Those sys ems a e compa ed o human d i e s o
o he sys em implemen a ions as a baseline. Based on cu en
acciden da a, he o e all sa e y impac s, e.g., a he Eu opean
le el, a e es ima ed [2]. Howe e , o a alid s a emen abou
The esea ch leading o hese esul s has ecei ed unding om he
Eu opean Union’s Ho izon 2020 esea ch and inno a ion p og am unde g an
ag eemen No 101006664. The sole esponsibili y o his publica ion lies wi h
he au ho s. The au ho s would like o hank all pa ne s wi hin he Hi-D i e
p ojec (hi-d i e.eu) o hei coope a ion and aluable con ibu ion.
he impac o a d i ing unc ion, he scena ios as well as he
beha io o he sys em unde es (SuT) and o he oad use s
a e o be modeled ealis ically.
One key enable o au oma ed d i ing is ehicle- o-
e e y hing (V2X) communica ion. Especially, when V2X is
u ilized o coope a i e awa eness and collec i e pe cep ion
use cases, posi i e e ec s on oad sa e y a e expec ed [3].
Coope a i e awa eness in his con ex means ha oad use s
a e in o med abou each o he ’s posi ion, dynamics, and a -
ibu es [4]. Connec ed oad use s migh use his in o ma ion
o inco po a e i in hei beha io planning, achie ing he
abili y o o e come hei pe cep ion limi a ions due o limi ed
senso ange and isual obs uc ions. While sa e y bene i s
o au oma ed d i ing gi en ideal coope a i e awa eness seem
ob ious, i is no clea i such bene i s can be achie ed gi en
ac ual coope a i e awa eness wi h e o s in he ansmi ed
da a, e.g., due o imp ecise localiza ion. This migh lead o a
dec ease in o e all a ic sa e y, depending on he magni ude
o hese e o s.
Cu en esea ch on coope a i e awa eness and collec i e
pe cep ion mainly ocusses on echnical aspec s, while ying
o de i e ough sa e y es ima es based on hese aspec s [3],
[5]–[7]. Thus, no p ope sa e y impac assessmen is pe -
o med. Howe e , o he esea ch assesses sa e y impac s o
o he V2X use cases, ha canno be conside ed as coope a i e
awa eness o au oma ed d i ing. One s udy assesses he sa e y
impac s o wa nings o ulne able oad use s (VRUs) ia
V2X [8]. I coope a i e awa eness use cases a e in es iga ed
in sa e y impac assessmen , i is assumed as ideal, neglec ing
e o s in he ansmi ed da a, which a e also he e o ma u e
sys ems ha can be in oduced o he ma ke .
The au ho s a e no awa e o p ope p ospec i e sa e y
impac assessmen s udies o espec i e me hods in es iga ing
coope a i e awa eness-enabled ADFs inco po a ing ealis ic
modelling o e o s. As he conside a ion o coope a i e awa e-
ness a ec s di e en s eps o p ospec i e sa e y impac assess-
men , s a ing om da a p epa a ion owa ds he simula ion
i sel , he ull p ocess is o be ex ended.
This wo k complemen s cu en esea ch wi h he ollowing
con ibu ions:
•Ex ension o he me hodology o he main s eps o
p ospec i e sa e y impac assessmen o he conside a ion
This ull- ex pape was pee - e iewed a he di ec ion o IEEE Ins umen a ion and Measu emen Socie y p io o he accep ance and publica ion.
o e o s in coope a i e awa eness.
•Exempla y applica ion o he ex ended me hodology o
he sa e y impac assessmen o ETSI Coope a i e Awa e-
ness Messages (CAMs) [4] a u ban in e sec ions based
on ac ual acciden da a.
•Compa ison o ADFs wi h no connec i i y, coope a i e
awa eness wi h e o s, and ideal coope a i e awa eness
a u ban in e sec ions.
The de eloped me hodology is i s applied o CAMs as he e
a e al eady se ies p oduc ion ehicles on he ma ke sending
hese messages, making u ilizing hese messages in ADFs
ealis ic in he nea u u e. Beyond ha , he de eloped me hod-
ology is alid o assessing collec i e pe cep ion use cases,
oo, ex ending i o mul iple e oneous objec in o ma ion om
di e en sou ces.
A e elabo a ing on ela ed wo k (Sec ion II), he de eloped
me hodology is p esen ed (Sec ion III). I ocusses speci ically
on da a p epa a ion, scena io gene a ion, and sys em mod-
elling. The p esen ed me hodology is applied in an expe imen
conduc ing sa e y impac assessmen simula ions o di e en
SuTs a an u ban in e sec ion based on ac ual acciden da a
and eal-wo ld V2X da a (Sec ion IV). The esul s o he
expe imen a e p o ided and discussed in Sec ion V. Finally,
he wo k is concluded (Sec ion VI).
II. RELATED WORK
Cu en ela ed esea ch ocusses on gene al p ospec i e
sa e y impac assessmen o au oma ed d i ing on he one
hand. On he o he hand, i ocuses on coope a i e awa eness
and collec i e pe cep ion analysis, as well as he modelling
o i . To p epa e combining bo h ields in his wo k, hey a e
p esen ed in he ollowing.
A. P ospec i e Sa e y Impac Assessmen
P ospec i e sa e y impac assessmen aims a assessing
he po en ial sa e y bene i s o a echnology be o e b oad
in oduc ion o ma ke [1]. The sys ems inco po a ing his
echnology a e assumed as ma u e sys ems, hus, ul illing,
e.g., unc ional sa e y equi emen s. By u ilizing simula ions,
he change in equency o de ined scena ios is combined
wi h de ailed analyses on changed acciden and inju y isks
o he indi idual scena ios. Based on s a is ics o he a ge
acciden s, i is scaled up o assess he po en ial impac s o a
de ined egion, such as he EU [2].
When assessing coope a i e awa eness and collec i e pe -
cep ion use cases, he changed inju y isk including ull
a oidance o con lic s in de ined scena ios is o pa icula
in e es . Based on da a and sys em speci ica ions, conc e e
scena ios, as well as models, a e de i ed (see Fig. 1). Based
on ha , he baseline and ea men a e simula ed and he inju y
isk o he indi idual conc e e scena ios is e alua ed u ilizing
inju y isk unc ions (IRFs).
The baseline gene a ion is mainly based on in-dep h acci-
den da a, such as GIDAS [9]. A dis inc ion is made be ween
h ee gene al app oaches based on he da a usage [1]:
•App oach A: o iginal cases wi hou modi ica ions.
Scena io simula ion Simula ion
Change in
scena io
equency
Changed
scena io inju y
isk
Impac on he
numbe o
acciden s pe
se e i y
Da a
Sys em
speci ica ion
Scena ios
Models
Baseline
T ea men
Inju y isk
calcula ion
Scaling up
&
Fig. 1. Abs ac ed sa e y impac assessmen low. Changes in scena io
equency and inju y isk a e combined in he scaling up wi h acciden
s a is ics. The lowe pa de ails he scena io simula ion o he inju y isk
calcula ion.
•App oach B: o iginal cases wi h modi ica ions.
•App oach C: syn he ically gene a ed cases.
These app oaches can be u he speci ied based on he de-
ailed case gene a ion p ocess as well as he case ins an ia ion
p ocess [1].
B. V2X and Coope a i e Awa eness
The s anda d o coope a i e awa eness implemen ed in cu -
en se ies p oduc ion ehicles is he ETSI CAM [4]. Messages
a e p o ided by ehicles wi h equencies be ween 1 Hz and
10 Hz based on de ined igge s, such as changes in he speed.
Among o he in o ma ion, CAMs include in o ma ion on he
cu en posi ion and he dynamic s a e o he co esponding
oad use . In addi ion, o each o he alues, co a iances a e
p o ided o e lec he p ecision.
Coope a i e awa eness, as well as collec i e pe cep ion,
a e analyzed on a ious laye s in simula ion and in eal-
wo ld es ing. O en, he ocus is on echnical analyses o he
communica ion [3], [5]. The ocus o his wo k is on analyzing
coope a i e awa eness in eal-wo ld applica ions as a basis o
modeling o simula ion.
Di e en me ics a e used in li e a u e o desc ibe he
collec i e pe cep ion quali y, which can be applied o coop-
e a i e awa eness use cases, oo. Fo example, he age o
in o ma ion agg ega es all a ising la encies, and he objec
acking accu acy desc ibes he accu acy o he ansmi ed
objec cha ac e is ics [3]. In addi ion, me ics on signal le el,
e.g., based on packe loss, a e used in some s udies ocussing
on de ailed echnical analyses.
Rela ed o he objec acking accu ancy, one s udy in es-
iga ed he accu acy o di e en da a ansmi ed ia CAMs
o se ies p oduc ion ehicles [10]. While di e en ehicles
a e analyzed based on a p ecise e e ence GNSS sys em, only
a e age e o s a e p o ided.
Fo de ailed in es iga ion o V2X use cases, di e en
da ase s a e a ailable [11]. The da ase used in his wo k is he
V2AIX da ase [12], which includes CAMs p o ided by se ies-
p oduc ion ehicles in eal-wo ld a ic in di e en loca ions.
This allows de ailed in es iga ions, e.g., on pa e n o e o s
and ansmi ed co a iances.
C. Modelling Coope a i e Awa eness
Coope a i e awa eness can be modeled o simula ions on
di e en le els, s a ing a modelling he indi idual senso s
including he objec de ec ion and de ailed message ans-
mission [13]. Fo sa e y impac assessmen , his can be
abs ac ed o achie e a model-in- he-loop simula ion. This
sa es compu a ion esou ces and elimina es e o s caused by
cu en imma u i ies in simula ion en i onmen s, pe cep ion
algo i hms, o signal ansmission simula ion.
One way o abs ac ing is applying coope a i e pe cep ion
e o models [14] o coope a i e awa eness. This app oach
includes modelling he indi idual pe cep ion e o s wi hou
simula ing senso s in de ail, as well as using pe cep ions om
mul iple sou ces, such as he ansmi ed objec s and he SuT’s
own pe cep ion.
III. METHODOLOGY
Ex ending p ospec i e sa e y impac assessmen o he con-
side a ion o V2X in gene al o coope a i e awa eness in
pa icula equi es ex ending he ele an s eps by laye 6
(digi al in o ma ion) o he 6-laye -model (6LM) [15]. This
includes ex ending he sa e y impac assessmen by e o s
which a e inhe en in coope a i e awa eness, e en in ma u e
sys ems. The main s eps ha equi e ex ended app oaches a e
he da a p epa a ion, he scena io gene a ion, and he model
gene a ion and simula ion.
As he de ails o he applica ion o he me hodology a e
highly dependen on he a ailable da a and he conside ed
SuTs, his sec ion p o ides i s guidance while i is applied
o an example in Sec ion IV.
A. Da a p epa a ion
Gene a ing scena ios o sa e y impac assessmen is usually
based on acciden da a. As coope a i e awa eness aims a
o e coming limi ed sensing, e.g., because o isual obs uc-
ions, i is essen ial o ha e elemen s in luencing sensing
inco po a ed as ep esen a i e as possible in he cases ha a e
o be simula ed.
In addi ion o ensu ing applicable da a o he scena io
gene a ion i sel , ac ual coope a i e awa eness da a need o be
analyzed o model he coope a i e awa eness in he simula ion.
This includes modelling he ansmi ed co up ed ehicle
s a es, hei co a iances, he logic o sending messages, and
he la encies in ansmission.
Fo he e o model o he ansmi ed ehicle s a es, i
is impo an o conside he e o dis ibu ion as well as he
unde lying sys ema ics, i.e., is he e o independen pe sen
message o a e he e o s ac oss he indi idual messages
co ela ed. This migh be he case when he e is, e.g., a
cons an la e al e o in he posi ioning o a gi en ime. As he
ansmi ed co a iances pe s a e in eal applica ions a e only
es ima ed by he sys em, hey can di e om he ac ual e o
models. Thus, i is equi ed o model hese co a iances as well
based on eal-wo ld da a. As he model o apply is dependen
on he da a, he e canno be a gene al e o model de ined. On
example model is de eloped in Sec ion IV-B. In addi ion o he
imp ecisions in he a iables, he age o in o ma ion o he
ansmi ed da a is o be modeled. This includes modelling he
equencies o igge s o ansmission as well as he la encies
in communica ion. While he i s can be de i ed om message
speci ica ions, he la e migh be de i ed om eal-wo ld da a.
B. Scena io gene a ion
A c ucial pa o he sa e y impac assessmen is gene a -
ing he d i ing scena ios o be simula ed o achi e a alid
baseline. When p ospec i ely assessing coope a i e awa eness
o au oma ed d i ing, digi al in o ma ion is o be conside ed
especially in he scena io gene a ion, as laye 6 o he 6LM
needs o be added. This includes especially in o ma ion ha
migh di ec ly in luence he SuT’s eac ion, such as he objec
acking accu acy and he age o in o ma ion. Applying his o
app oach A is no an op ion, as o iginal cases ha include co-
ope a i e awa eness da a do no exis ye . Ins ead, app oaches
B and C a e applicable based on a ailable da a and ools. Fo
app oach B, he o iginal cases a e o be ex ended by he V2X
in o ma ion. In addi ion o modi ica ions o he scena ios i sel ,
an e o model is o be added o gene a e he e o s o he
indi idual cases. Fo app oach C, he e o model can be added
al eady o he ini ial case gene a ing, no jus sampling, e.g.,
ehicle s a es, bu also conc e e e o s.
Fo bo h app oaches, i is impo an o conside he sys em-
a ics behind he e o , i.e., is he e o andomly gene a ed o
each ime s ep o is i consis en o mul iple ime s eps.
C. Models and simula ion
Based on he gene a ed scena ios, he SuT can be analyzed
in simula ion. I equi es modelling he beha io o he SuT
based on he pe cei ed in o ma ion. The beha io is mainly
de ined by he pe cep ion o he en i onmen including he
digi al in o ma ion p o ided and he ajec o y planning based
on he pe cei ed in o ma ion. The pe cep ion can be modeled
based on coope a i e pe cep ion e o models, which a e based
on indi idual pe cep ion e o models [14]. This includes he
usion o mul iple pe cep ions.
As sa e y impac assessmen usually in es iga es abs ac
unc ions as model-in- he-loop simula ions, he ajec o y plan-
ning is usually abs ac ed as well. Ex ending o coope a i e
awa eness, he main change in he ajec o y planning is how
o conside ecei ed loca ions o su ounding dynamic objec s
inco po a ing he a ailable co a iances. This can be done o
in e sec ion con lic s based on iden i ying he posi ion o he
objec (ˆx, ˆy), conside ing he co a iance, which would lead o
he ea lies con lic assuming cons an eloci ies and he e o e
equi es he ea lies eac ion by he SuT. Fig. 2 illus a es he
ele an a iables. The iming o he con lic is de e mined by
he ime- o-con lic -poin TTCP o he SuT:
TTCP(ˆy) = sSuT,en y −ˆy
SuT
(1)
SuT
𝑠𝑜𝑏𝑗,𝑒𝑛𝑡𝑟𝑦
𝑠𝑜𝑏𝑗,𝑒𝑥𝑖𝑡
𝑠𝑆𝑢𝑇,𝑒𝑛𝑡𝑟𝑦
𝑠𝑆𝑢𝑇,𝑒𝑥𝑖𝑡
𝜉 ∙ 𝑐𝑜𝑣𝑦
𝑥
𝑦
(ො𝑥, ො𝑦)
𝜉 ∙ 𝑐𝑜𝑣𝑥
Fig. 2. Ske ch o abs ac ing he ajec o y planning p oblem based on he
TTCP and pPET. SuT in blue, con lic ing objec (obj), which is sending he
CAM, in o ange.
I is dependen on he dis ance o a el sSuT,en y o
he SuT o en e he con lic a ea, he y-coo dina e o he
conside ed objec posi ion ˆy, and he SuT’s eloci y SuT .
The SuT needs o s op be o e en e ing he con lic a ea o
a oid a p edic ed collision. (1) ep esen s an ac ual con lic ,
only when he objec ehicle is occupying he con lic a ea o
he gi en iming o is en e ing i while he SuT is occupying
his a ea.
Adding a sa e y ma gin in he o m o a minimum p edic ed
pos -enc oachmen - ime pPETmin esul s o he case, ha
he objec a i es i s , in he cons ain o a con lic :
sobj,exi −ˆx
obj
≧sSuT,en y −ˆy
SuT
−pPETmin (2)
I is de e mined by he dis ance sobj,exi he objec needs o
a el o clea he con lic a ea and i ’s eloci y obj as well
as he dis ance sSuT,en y he SuT needs o a el o en e he
con lic a ea and i ’s eloci y SuT .
Simila ly, in case he SuT would each he con lic a ea i s ,
he cons ain o a con lic is:
sSuT,exi −ˆy
SuT
≧sobj,en y −ˆx
obj
−pPETmin (3)
While he dis ances sa e ela ed o he ansmi ed poin
o he objec , bo h equa ions, (2) and (3), include he ac ually
conside ed objec posi ion in ela ion o he ansmi ed posi-
ion in he CAM o conside he co a iances in longi udinal
di ec ion co xand la e al di ec ion co y.
The objec posi ion conside ed o he con lic p edic ion
needs o lie inside he scaled co a iance ellipse:
ˆx
ξ·co x2
+ˆy
ξ·co y2
≦1(4)
The a iable ξis used o scale he ellipse o de e mine which
p obabili ies o possible poin s should s ill be conside ed.
These h ee equa ions de ine he cons ain s o inding he
posi ion o he objec o minimize he TTCP:
min{TTCP(ˆy) : (2),(3),(4)}(5)
Wall
Fig. 3. Scena io isualiza ion - SuT in black, con lic ing objec app oaching
om igh in ed. Visual obs uc ion due o wall. Sc eensho om CARLA
simula o .
This way, a cau iously d i ing SuT would be conside ed,
as i eac s based on he posi ion wi hin a gi en co a iance
equi ing he ea lies eac ion o a oid a con lic .
IV. EXPERIMENT
The implemen a ion o he desc ibed me hod depends on he
in es iga ed use case, he sys em, he in es iga ed scena ios
as well as he a ailable da a. Fo his eason, he me hod
is illus a ed based on a conc e e example. A connec ed and
au oma ed d i ing unc ion is in es iga ed ha u ilizes CAMs
a u ban in e sec ions o o e come isual obs uc ions. The
ocus o his expe imen is on in es iga ing he in luence o
ansmi ed posi ioning e o s as well as he age o in o -
ma ion. In addi ion, no a ull sa e y impac assessmen is
conduc ed, including he scaling up o achie e absolu e e ec s.
Ins ead, he ocus is on assessing he ela i e e ec s in a
de ined scena io o di e en SuTs.
A. Expe imen se up
The scena io in es iga ed in his expe imen is an in e -
sec ion scena io. The SuT in ends o c oss he in e sec ion
s aigh , while he e a e po en ially con lic ing ehicles ha
y o c oss he in e sec ion om he igh (see Fig. 3).
The in e sec ion has ou a ms and is loca ed in an u ban
a ea, leading o in es iga ing he speed limi s o 30 km/h
and 50 km/h. Visual obs uc ions a e placed based on ac ual
acciden da a om GIDAS. The SuT en e s he in e sec ion
o some scena ios om a mino and o o he s om a
majo oad, in luencing he igh -o -way si ua ion. The SuT
is con olled by an ADF (c . Sec ion IV-D). The con lic ing
ehicle is human-d i en and p o ides CAMs based on he
ETSI speci ica ions [4].
B. CAM Da a P epa a ion
The V2AIX da ase is u ilized o analyze CAM da a and
de i e he e o model. While ansmi ed co a iances o he
indi idual ehicle s a es can be di ec ly de i ed om he da a,
he ac ual e o s canno be de i ed di ec ly due o lacking
g ound u h. This especially holds ue o he posi ioning
e o , which is in he ocus o he in es iga ions o his wo k.
This sho age can be o e come by looking a he la e al e o s
i s o speci ic oad segmen s in he da a.
We ex ac CAMs om a oad segmen which allows only
e y limi ed la e al de ia ion om he e e ence pa h, i.e.
na ow one-way-s ee s. Once he oad sec ion is de e mined,
28 ehicle acks, iden i ied by he s a ion ID in he CAM, a e
selec ed o a mo e de ailed analysis (see Fig. 4, p o iding
an exce p ). The dis ance om he oad cen e line o each
ins ance o he ansmi ed posi ion in his oad segmen
is calcula ed. Fo each ehicle, he a e age e o in he
ansmi ed CAMs along he oad sec ion is calcula ed. The
esul ing his og am o he mi o ed absolu e alues including a
i ed Gaussian dis ibu ion is isualized in Fig. 5. The de i ed
s anda d de ia ion is 1.41 m which is in line wi h he indings
in [10]. Addi ionally, i ma ches he ansmi ed co a iances
in he analysed CAMs o on a e age 3 m. The co a iance
alue ep esen s co e ing 95 %, which is equal o a ound
wo s anda d de ia ions. The ansmi ed s anda d de ia ions
a e always a ound 3 m o la e al as well as longi udinal
co a iance, wi h only mino a ia ions o a ew cen ime e s.
Based on his, he ansmi ed co a iances o he e o model
a e assumed cons an wi h 3 m o he expe imen .
In addi ion o he a e age e o , om Fig. 4 we de i e ha
he e is a sys ema ic e o p esen in he messages and no
a andom e o pe ime s eps. This ma ches he assump ion
ha he p o ided posi ions in he CAMs we e il e ed using a
Kalman il e .
Based on he indings om [10], which in es iga e simila
e o s o la e al and longi udinal e o s, he indings o his
wo k o la e al e o s a e also applied o he longi udinal
e o s. This way, a comple e e o model o he posi ioning
including he ansmi ed co a iances is de ined o model he
objec acking accu acy.
In addi ion, he age o in o ma ion is conside ed. Fo his
wo k, i is modeled ia wo main componen s: he la ency in
in o ma ion ansmission and p ocessing, and he igge s o
sending CAMs. While he i s is modeled wi h a cons an
alue o 50 ms based on li e a u e [16], he la e is modeled
using he ETSI speci ica ions o CAMs [4]. The messages a e
sen wi h a equency be ween 1 Hz and 10 Hz, depending on
he igge s de ined in he s anda d: a new CAM is gene a ed
i he heading changes by mo e han 4◦, he posi ion changed
by mo e han 4 m, o he eloci y changed by mo e han
0.5 m s−1. This esul s in addi ional AoI o up o 0.1 s o
1.0 s, depending on he CAM igge ing and he ime when
he esul ing CAM is accessed.
C. CAM Scena io Gene a ion
The basis o he scena io gene a ion a e 375 in e sec ion
acciden s om he GIDAS da abase. This includes cases whe e
he e e ence ehicle had he igh o way and cases whe e i
had o yield.
Acco ding o he baseline gene a ion app oach B (c . Sec-
ion III-B), hese o iginal cases a e modi ied o i he in ended
use. The e e ence ehicle is eplaced by he SuT, while he
beha io o he con lic ing ehicle is no changed. In ac , i
Fig. 4. T acks o di e en ehicles communica ed ia CAMs a a na ow
oad sec ion, no allowing high de ia ions om he e e ence line. Re e ence
line dashed.
Fig. 5. His og am o he a e age la e al de ia ion o each ehicle om he
e e ence line. Absolu e alues mi o ed a 0 m. Fi ed Gaussian wi h s anda d
de ia ion o 1.41 m.
is ollowing a p ede ined ajec o y coming om he o iginal
case. Fo he case gene a ion p ocess, each o iginal case is
a ied h ee imes ega ding he iming when he con lic ing
ehicle eaches he con lic poin ela i e o he SuT. The di -
e ence in iming is sampled om a uni o m dis ibu ion om
−3 s o 3 s. The dis ibu ion is chosen uni o m, as i is assumed
ha he ini ial iming o app oaching he in e sec ion o bo h
pa icipan s be o e eac ing o each o he is independen o
each o he . Sampling he iming ensu es ha he gene a ed
baseline includes ac ual con lic s as well as close encoun e s.
In addi ion, o each o he esul ing 1,125 cases, h ee
posi ion e o s a e sampled om he dis ibu ion de i ed in
Sec ion IV-B. The sampled e o is kep cons an h oughou
he du a ion o he indi idual cases, as he da a analysis
e ealed sys ema ic e o s o sho segmen s o d i ing. The
la e al and longi udinal e o s a e sampled independen ly o
each o hese cases. This esul s in up o 3,375 cases ha
a e simula ed o he di e en SuTs. The esul ing de ailed
baseline app oach is B2P [1].
D. Sys ems unde es
The goal o his expe imen is o assess he impac s o
coope a i e awa eness in ela ion o a non-connec ed ADF.
Two di e en ea men sys ems a e compa ed o one common
baseline:
•Baseline ADF (BADF): he ADF uses he own pe cep ion
only.
•Ideal coope a i e awa eness ADF (ICA-ADF): he
ADF ecei es g ound u h posi ion o he con lic ing
objec wi hou any la encies. I.e., he age o in o ma ion
is always 0.
•Real coope a i e awa eness ADF (RCA-ADF): he
ADF ecei es co up ed posi ions o dynamic objec s in
case hey a e no isible o i s own senso s. The ADF is
awa e o he age o in o ma ion om he CAM imes amp
and p edic s he objec ’s mo emen .
The baseline sys em o he expe imen is he Baseline ADF
(BADF) om he Hi-D i e p ojec [2]. The ele an senso s o
he o e all senso se up o his expe imen a e a on acing
long- ange came a and LiDAR, a long- ange on ada , and
mid- ange co ne ada s a he wo on co ne s. This gi es
he BADF he abili y o de ec dynamic objec s om 150 m
dis ance, as long as hey a e no isually obs uc ed. Based
on he pe cei ed objec s, decision-making is pe o med. As
long as no con lic s a e de ec ed, he BADF app oaches he
in e sec ion wi h he speed limi as a ge speed, when i is on
a majo oad. D i ing on a mino oad, he BADF app oaches
he s op line a 10 km h−1. I in his case a con lic is de ec ed,
i adjus s he desi ed speed a he s op line o 0 km h−1. The
BADF ies o lea e a gap o 2 s in case i needs o yield.
In case a collision is p edic ed, independen o he oad, an
au oma ed eme gency b aking sys em (AEB) is ac i a ed based
on he ime- o-collision (TTC).
Two di e en ea men sys ems a e compa ed o his base-
line. The gene al d i ing s a egy is he same as o he BADF,
jus he objec in o ma ion used o decision-making di e s.
The i s ea men sys em is an ADF wi h ideal coope a i e
awa eness (ICA-ADF), meaning ha is has access o he
g ound u h in o ma ion o all dynamic objec s a any ime.
This ea men ep esen s he maximum impac s ha could be
achie ed u ilizing coope a i e awa eness.
The o he ea men sys em is an ADF which can ecei e
CAMs om isually obs uc ed objec s (RCA-ADF). The
CAMs con ain he co up ed objec s a e acco ding o Sec-
ion IV-C o he ime a message was igge ed. The messages
can only be accessed by he ADF as soon as he cons an
la ency has passed. Based on he ansmi ed imes amp, i
p edic s he objec mo emen using a cons an eloci y model.
As long as he SuT canno pe cei e he objec wi h he own
senso s, i akes he ansmi ed in o ma ion including co a i-
ances om he CAM o he decision-making. Conside ing he
co a iances is c ucial as he example in Fig. 6 shows.
Based on he ansmi ed posi ion and he co a iance alues
in longi udinal and la e al posi ion, one de ined poin is
calcula ed, which equi es he ea lies eac ion o he SuT (c .
Sec ion III-C). This is achie ed by sol ing he op imiza ion
p oblem (5). ξis se o one, meaning he co a iance ellipse
is no scaled, o co e he 95 % mos p obable posi ions. A
sa e y gap o pPETmin = 0.5sis used, co e ing also e y
close misses in p edic ion, whe e a sa ely designed ADF would
s ill b ake.
As soon as he SuT can pe cei e he objec wi h i s
own senso s, i is deciding based on his pe cep ion, as he
Visual obs uc ion
SuT
(A) Objec g ound u h
(B) CAM objec
posi ion
(C) Objec posi ion
conside ed by SuT
E o model om
CAM da a analysis
T ansmi ed
co a iance ellipsis
Fig. 6. Visualiza ion o objec posi ions - no a scale o clea e isualiza ion.
The posi ion ansmi ed ia he CAM (B) is calcula ed based on he g ound
u h (A) and he e o sampled om he e o model de i ed om da a. Based
on (B) and he ansmi ed co a iance, he SuT conside s he poin equi ing
he ea lies eac ion (C).
ansmi ed co a iances in he CAMs a e a he high, leading
o a highe us in he own pe cep ion.
V. EXPERIMENT RESULTS AND LIMITATIONS
The scena ios gene a ed based on Sec ion IV-C we e sim-
ula ed wi h he h ee sys ems de ined in Sec ion IV-D. In
addi ion, as a baseline, he gene a ed cases we e simula ed
inco po a ing he o iginal eac ions o he human d i e s (Man-
ual) in he ini ial acciden da a used o baseline gene a ion.
A. Expe imen Resul s
Fo he h ee condi ions ha don’ conside V2X e o s
(Manual, BADF, ICA-ADF), he 1,125 scena ios esul ing
om he iming a ia ion acco ding o he selec ed app oach
B2P we e simula ed. Fo he RCA-ADF, o each o hese
scena ios, h ee e o s we e sampled based on he model
de i ed in Sec ion IV-B, esul ing in 3,375 simula ed scena ios.
The ocus o he e alua ion o he conduc ed simula ions is
on he esul ing c ash a es. E en hough a di e en numbe
o scena ios we e simula ed o di e en SuTs, he indi idual
c ash a es a e s ill compa able as hey a e ela i e o he
numbe o simula ed scena ios.
The esul ing c ash a es pe SuT a e isualized in Fig 7. Fo
he Manual baseline, 48.1 % o he simula ed scena ios esul
in c ashes. This indica es ha he baseline includes collisions
as well as nea misses, esul ing in bo h si ua ions whe e he
SuTs need o eac o a oid collisions and si ua ions whe e
no eac ion is needed when he SuT has he igh in o ma ion
a ailable.
The BADF educes he c ash a e o 18.4 %. This shows
ha also non-connec ed ADFs ha e he possibili y o inc ease
sa e y a u ban in e sec ion con lic s. S ill, he e a e some
si ua ions in which he BADF canno a oid a con lic . This
is especially he case when an objec is dis ega ding he igh
o way and isual obs uc ions a e p esen .
The ICA-ADF achie es o o e come his sho age o he
BADF by educing he esul ing c ash a e o 5.3 % as i
is expec ed o coope a i e awa eness. Also as app oaching
Fig. 7. Expe imen esul s including he human-d i en baseline as e e ence
(Manual), he baseline ADF wi hou V2X (BADF), he ideal coope a i e
awa eness ADF (ICA-ADF), and he eal coope a i e awa eness ADF (RCA-
ADF) using CAMs wi h e o s.
ehicles which a e isually obs uc ed can be de ec ed ea ly
enough o a oid con lic s.
When ealis ic e o s a e added, he c ash a e esul s o
he RCA-ADF o 6.7 %. E en hough his SuT is designed
o eac as sa e as possible o he ansmi ed posi ion and
co a iances (c . Sec ion IV-D), he e is an inc ease in c ash
a e compa ed o he ideal da a ansmission. In addi ion, alse-
posi i e ac i a ions o he AEB migh inc ease, leading o
po en ial new con lic s wi h ollowing a ic which would
be a oidable. S ill, using he ealis ic CAMs ou pe o ms no
using connec i i y a all.
B. Expe imen Limi a ions
The expe imen esul s we e gene a ed looking a one con-
lic ype only. In addi ion, he da a conside ed o build up
he CAM e o model is limi ed. Conside ing mo e da a and
mo e in luencing ac o s such as loca ion and su ounding in-
as uc u e in luencing GNSS signals could enhance alidi y.
The SuT conside ed in he model-in- he-loop simula ions is
hea ily abs ac ed, as i is usual o p ospec i e sa e y impac
assessmen . Especially he ajec o y planning conside ing he
co a iance p o ided wi hin he CAM migh ha e a majo
in luence on he esul s. Wi hin his wo k, one easonable
app oach was assumed, no necessa ily ep esen a i e o all
possible app oaches.
VI. CONCLUSION
The p esen ed wo k de eloped an ex ended me hodology
o conside coope a i e awa eness, including inaccu acies, in
sa e y impac assessmen . The me hod was demons a ed based
on he p o ision o ETSI CAMs by human-d i en ehicles a
isually obs uc ed u ban in e sec ions. Based on da a om
he V2AIX da ase , he posi ioning e o in cu en CAMs
was modeled using a Gaussian dis ibu ion wi h a s anda d
de ia ion o 1.41 m wi h a sys ema ic e o samples o each
simula ed case.
Di e en SuTs we e in es iga ed, co e ing no connec i i y,
ideal coope a i e awa eness, and ac ual coope a i e awa eness
including posi ioning e o s. The esul s e ealed a clea sa e y
gain o ideal coope a i e awa eness compa ed o he non-
connec ed ADF. The sa e y dec eased compa ed o he ideal
coope a i e awa eness, when e o s o he CAMs we e added,
while he c ash a e was s ill lowe compa ed o he non-
connec ed ADF.
Fu u e wo k will ocus on de ailed modeling o he usion o
he SuT’s own pe cep ion wi h he ecei ed V2X in o ma ion
and modeling o he a ic pa icipan s which migh be exposed
o dange due o alse-posi i e eac ions o he SuT.
REFERENCES
[1] F. Fah enk og, A. Das, D. Sande , J. B¨
a gman, M. U ban, M. Pohl,
C. Glasmache , and e al., “Deli e able D2.1 - p ospec i e sa e y
assessmen amewo k - ins uc ion,” V4SAFETY, Tech. Rep., 2024.
[2] L. Va e , E. Ai oniemi, A. Bjo a n, F. Fah enk og, C. Felchon, S. In-
namaa, E. Leh onen, S. Rahmani, and H. Webe , “Deli e able D4.5 -
e ec s e alua ion me hods,” Hi-D i e P ojec , Tech. Rep., 2023.
[3] F. A. Schiegg, I. Lla se , D. Bischo , and G. Volk, “Collec i e pe cep-
ion: A sa e y pe spec i e,” Senso s (Swi ze land), ol. 21, pp. 1–24, 1
2021.
[4] ETSI, “ETSI EN 302 637-2 - in elligen anspo sys ems (ITS);
ehicula communica ions; basic se o applica ions; pa 2: Speci ica ion
o coope a i e awa eness basic se ice,” Tech. Rep., 2014.
[5] A. Bazzi, B. M. Masini, and A. Zanella, “Coope a i e awa eness in he
in e ne o ehicles o sa e y enhancemen ,” EAI Endo sed T ansac ions
on In e ne o Things, ol. 3, no. 9, 1 2017.
[6] M. Ka oui, V. Be g, and S. May a gue, “Assessmen o V2X commu-
nica ions o enhanced ulne able oad use s sa e y,” in IEEE Vehicula
Technology Con e ence, ol. 2022-June. Ins i u e o Elec ical and
Elec onics Enginee s Inc., 2022.
[7] M. Shan, K. Na ula, Y. F. Wong, S. Wo all, M. Khan, P. Alexande ,
and E. Nebo , “Demons a ions o coope a i e pe cep ion: Sa e y and
obus ness in connec ed and au oma ed ehicle ope a ions,” Senso s
(Swi ze land), ol. 21, pp. 1–31, 1 2021.
[8] L. Scho ies, N. Dah inge , U. Pi am, A. Rau , S. Nikolaou,
I. G agkopoulos, I. Tse sinas, and M. Panou, “Sa e y pe o mance as-
sessmen ia i ual simula ion o V2X wa ning igge s o cyclis s wi h
models c ea ed om eal-wo ld es ing,” Sus ainabili y (Swi ze land),
ol. 16, 1 2024.
[9] Bundesans al ¨
u S aßen- und Ve keh swesen and Fo schungs e eini-
gung Au omobil echnik, “GIDAS (Ge man In-Dep h Acciden S udy),”
h ps://www.gidas.o g/, accessed: 2025-04-09.
[10] M. Baude , A. Fes ag, T. Kubja ko, and H. G. Schweige , “Da a accu acy
in ehicle- o-x coope a i e awa eness messages: An expe imen al s udy
o he i s comme cial deploymen o C-ITS in eu ope,” Vehicula
Communica ions, ol. 47, 6 2024.
[11] B. S. Ba i, D. Pu hal, and K. Yelama hi, “Da ase s in ehicula com-
munica ion sys ems: A e iew o cu en ends and u u e p ospec s,”
SN Compu e Science, ol. 6, 3 2025.
[12] G. Kueppe s, J.-P. Busch, L. Reihe , and L. Ecks ein, “V2aix: A
mul i-modal eal-wo ld da ase o e si i s 2x messages in public oad
a ic,” 2024. [Online]. A ailable: h ps://a xi .o g/abs/2403.10221
[13] C. Gelle , B. Haas, A. Kloeke , J. He mens, B. Lampe, T. Beemelmanns,
and L. Ecks ein, “CARLOS: An open, modula , and scalable simula ion
amewo k o he de elopmen and es ing o so wa e o C-ITS,”
in IEEE In elligen Vehicles Symposium, P oceedings. Ins i u e o
Elec ical and Elec onics Enginee s Inc., 2024, pp. 3100–3106.
[14] A. Piazzoni, J. Che ian, R. Vijay, L. P. Chau, and J. Dauwels, “CoPEM:
Coope a i e pe cep ion e o models o au onomous d i ing,” in IEEE
Con e ence on In elligen T anspo a ion Sys ems, P oceedings, ITSC,
ol. 2022-Oc obe . Ins i u e o Elec ical and Elec onics Enginee s
Inc., 2022, pp. 3934–3939.
[15] M. Schol es, L. Wes ho en, L. R. Tu ne , K. Lo o, M. Schuldes,
H. Webe , N. Wagene , C. Neu oh , M. H. Bollmann, F. Ko ke, J. Hille ,
M. Hoss, J. Bock, and L. Ecks ein, “6-laye model o a s uc u ed
desc ip ion and ca ego iza ion o u ban a ic and en i onmen ,” IEEE
Access, ol. 9, pp. 59 131–59 147, 2021.
[16] Z. Liu, Z. Liu, Z. Meng, X. Yang, L. Pu, and L. Zhang, “Implemen a ion
and pe o mance measu emen o a V2X communica ion sys em o
ehicle and pedes ian sa e y,” In e na ional Jou nal o Dis ibu ed
Senso Ne wo ks, ol. 12, 09 2016.
