Fligh o he Fu u e: An Expe imen al Analysis
o E en -Based Vision o Online Pe cep ion Onboa d
Flapping-Wing Robo s
Raul Tapia,* Ja ie Luna-San ama ia, I an Gu ie ez Rod iguez,
Juan Pablo Rod íguez-Gómez, José Rami o Ma ínez-de Dios, and Anibal Olle o
1. In oduc ion
Flapping-wing obo s, also known as o ni hop e s, a e ae ial
pla o ms ha gene a e li and h us by mimicking he fligh
mechanism o bi ds and insec s. These obo s ha e high maneu-
e abili y and combine glide and flapping fligh modes o
minimize ene gy consump ion. Compa ed o mul i o o and
fixed-wing pla o ms, flapping-wing obo s consume less ene gy
and a e less dange ous in he e en o a collision. In addi ion,
hei wide ange o po en ial applica ions
has mo i a ed a significan esea ch and
de elopmen in e es in ecen yea s.
[1–3]
The design o pe cep ion sys ems o
flapping-wing obo s aces a ious chal-
lenges and limi a ions.
[4]
Fi s , o ni hop e s
ha e s ic es ic ions on payload and
weigh dis ibu ion, which di ec ly a ec
he numbe , size, and shape o senso s,
elec onics, ba e ies, and o he compo-
nen s onboa d. In addi ion, hei agile
mo emen s and flapping s okes p oduce
s ong mechanical ib a ions ha impose
ele an cons ain s on he senso s ha
can be used o online pe cep ion. In addi-
ion, online onboa d p ocessing plays a
c ucial ole in ae ial obo pe cep ion.
I enables he eal- ime decision-making
and con ol equi ed o achie e high
deg ees o sa e y, au onomy, and espon-
si eness h ough au onomous unc ionali-
ies such as obs acle de ec ion, collision
a oidance, and na iga ion, among o he s.
Fas eac i i y is pa icula ly c i ical conside ing he agile maneu-
e s o flapping-wing pla o ms. Howe e , hei cons ained pay-
load capaci y also imposes complex es ic ions on he p ocessing
uni s ha can be used, equi ing low-sized ligh weigh com-
pu e s whose esou ces can be limi ing o some applica ions.
The o ni hop e s’cons ained payload and onboa d compu a-
ional esou ces limi he numbe and ype o senso s ha can be
used and he ype o onboa d pe cep ion p ocessing. Senso s
such as ligh de ec ion and anging (LiDARs), ul asound sen-
so s, ada s, and in a ed came as, which a e widely used in mul-
i o o s, p esen se e al p oblems when used o flapping-wing
obo s, whose payload is in he ange o a ew hund ed g ams.
[5]
On he con a y, ision senso s ha e low size and weigh , p o id-
ing ich in o ma ion abou he en i onmen . Mos exis ing wo ks
ha e p oposed using ision-based pe cep ion o o ni hop e
au onomy.
[4,6]
Some wo ks ha e used schemes based on
adi ional ame-based came as in monocula
[7–9]
o s e eo
[10–14]
configu a ions. O he s ha e p oposed pe cep ion schemes based
on e en came as.
[15,16]
These no el bioinspi ed senso s, which
ope a e asynch onously and pe -pixel independen ly, y o
mimic animal ision sys ems, ei he in speed and ene gy
e ficiency,
[17]
o e ing ad an ages such as low la ency, high
obus ness agains mo ion blu , and high dynamic ange, among
R. Tapia, J. Luna-San ama ia, I. Gu ie ez Rod iguez,
J. P. Rod íguez-Gómez, J. R. Ma ínez-de Dios, A. Olle o
GRVC Robo ics Lab
Uni e si y o Se ille
41092 Se ille, Spain
E-mail: [email p o ec ed]
The ORCID iden ifica ion numbe (s) o he au ho (s) o his a icle
can be ound unde h ps://doi.o g/10.1002/aisy.202401065.
© 2025 The Au ho (s). Ad anced In elligen Sys ems published by Wiley-
VCH GmbH. This is an open access a icle unde he e ms o he C ea i e
Commons A ibu ion License, which pe mi s use, dis ibu ion and
ep oduc ion in any medium, p o ided he o iginal wo k is p ope ly ci ed.
DOI: 10.1002/aisy.202401065
Inspi ed by bi d fligh , flapping-wing obo s ha e gained significan a en ion due
o hei high maneu e abili y and ene gy e ficiency. Howe e , he de elopmen
o hei pe cep ion sys ems aces se e al challenges, mainly ela ed o payload
es ic ions and he e ec s o flapping s okes on senso da a. The limi ed
esou ces o ligh weigh onboa d p ocesso s u he cons ain he online p oc-
essing equi ed o au onomous fligh . E en came as exhibi se e al p ope ies
sui able o o ni hop e pe cep ion, such as low la ency, obus ness o mo ion blu ,
high dynamic ange, and low powe consump ion. This a icle explo es he use o
e en -based ision o online p ocessing onboa d flapping-wing obo s. Fi s , he
sui abili y o e en came as unde fligh condi ions is assessed h ough expe i-
men al es s. Second, he in eg a ion o e en -based ision sys ems onboa d
flapping-wing obo s is analyzed. Finally, he pe o mance, accu acy, and
compu a ional cos o some widely used e en -based ision algo i hms a e
expe imen ally e alua ed when in eg a ed in o flapping-wing obo s flying in indoo
and ou doo scena ios unde di e en condi ions. The esul s confi m he benefi s
and sui abili y o e en -based ision o online pe cep ion onboa d o ni hop e s,
pa ing he way o enhanced au onomy and sa e y in eal-wo ld fligh ope a ions.
RESEARCH ARTICLE
www.ad in ellsys .com
Ad . In ell. Sys . 2025, 2401065 2401065 (1 o 20) © 2025 The Au ho (s). Ad anced In elligen Sys ems published by Wiley-VCH GmbH
o he s.
[18,19]
In a ecen p e ious wo k,
[20]
we compa ed ame-
based and e en came as o flapping-wing obo pe cep ion, con-
cluding ha al hough e en -based echnology has a lowe deg ee
o ma u i y, i sui s he equi emen s o o ni hop e s. Taking ha
wo k as he s a ing poin , we analyze he use o e en -based ision
o online pe cep ion on boa d flapping-wing obo s wi h a
b oade app oach, e alua ing commonly used e en -based ision
algo i hms, sui abili y o indoo and ou doo o ni hop e pe cep-
ion, ease o in eg a ion, and a ailable esou ces and suppo ools.
This a icle b oadly analyzes he sui abili y o e en -based
ision o he online onboa d pe cep ion o flapping-wing obo s.
We in end o answe hese h ee ques ions:
Q1: A e e en came as sui able o online onboa d pe cep ion
conside ing he challenging fligh condi ions o flapping-wing obo s?
Q2: A e e en came as sui able o be in eg a ed (HW&SW) on
flapping-wing obo s conside ing he cons ain s o hese pla o ms?
Q3: A e he pe o mance and compu a ional cos o e en -based
algo i hms easible o enable online onboa d pe cep ion ask o
o ni hop e s?
We aim o answe his ques ion by pe o ming h ee analyses:
1) e alua ion o he sui abili y and obus ness o e en came as in
expe imen s ha mimic he condi ions ha can be ound in
flapping-wing fligh s; 2) e iew o he a ailable e en -based de i-
ces and esou ces and suppo ools ocusing on he HW&SW
in eg abili y on flapping-wing pla o ms; and 3) expe imen al
e alua ion o he pe o mance and compu a ional cos o some
widely used e en -based ision algo i hms when in eg a ed in o
flapping-wing obo s flying in di e en scena ios (including he
pla o m in Figu e 1). To he bes o he au ho s’knowledge,
his is he fi s wo k b oadly analyzing he use o e en ision
o la ge-scale flapping-wing obo s. In addi ion, we p o ide
he da a eco ded onboa d wo di e en flapping-wing obo s
in indoo and ou doo scena ios used in he expe imen al e alu-
a ion o his wo k.
The es o he a icle is s uc u ed as ollows: Sec ion 2 b iefly
summa izes he main wo ks in he opics add essed in he a i-
cle. The expe imen al analysis o he sui abili y o e en came as
o he pe cep ion challenges o flapping-wing fligh is p esen ed
in Sec ion 3. The HW&SW in eg abili y o e en -based ision
onboa d o ni hop e s is analyzed in Sec ion 4. The e alua ion
o widely used e en -based ision algo i hms o o ni hop e s’
onboa d pe cep ion is p esen ed in Sec ion 5. Finally, Sec ion 6
concludes he a icle wi h he findings and main u u e s eps.
2. Rela ed Wo k
2.1. E en Came as
In ecen yea s, e en came as ha e a ac ed inc easing in e es
in fields such as a ificial in elligence, compu e ision, neu o-
mo phics, and obo ics. The o igins o e en came as ace back
o 1991 wi h pionee ing e o s in neu omo phic ision echnol-
ogy and he eme gence o he silicon e ina,
[21]
which mimicked
biological eyes’p ocessing capabili ies. This ounda ion led o he
EU’s CAVIAR p ojec ,
[22]
which ad anced ea ly AER-based e en
ision sys ems. In 2008, he fi s comme cial e en came as
(128 128 esolu ion, 40 μm pixel size; now iniVa ion’s
DVS128) we e in oduced.
[23–26]
By 2019, majo playe s like
Sony and Samsung en e ed he ma ke , signaling g owing com-
me cial in e es in e en -based senso s. Technological ad ance-
men s con inued apidly, wi h he fi s HD e en came as
[27]
(1280 720 esolu ion, 5 μm pixel size) launching in 2021.
In 2022, Me a began showing in e es in his inno a i e echnol-
ogy,
[28]
ollowed by a significan collabo a ion be ween P ophesee
and Qualcomm (h ps://www.p ophesee.ai/2023/02/27) in 2023
and wi h AMD and Lucid Vision Labs (among o he s) (h ps://
www.p ophesee.ai/2024/10/02) in 2024. Founda ion s udies
in oducing no el e en came a designs ha e explo ed hei
ad an ages o e adi ional senso s.
[23–27,29–34]
Howe e , he e
a e ew wo ks ha analyze and compa e ame-based and
e en -based ision. Add essing his gap, Holešo ský e al.
[35,36]
conduc ed an expe imen al analysis using a speed-con olled
spinning disk and a bulle fi ed a a ious eloci ies o compa e
he pe o mance o an ATIS HVGA Gen3 e en came a, a
DVS240 e en came a, and wo high-speed global-shu e cam-
e as. Thei esul s demons a ed he ad an ages o e en cam-
e as, pa icula ly in e ms o bandwid h e ficiency, and also
explo ed he limi a ions o e en -based senso s in pixel la ency
and eadou bandwid h, pa icula ly in highly clu e ed scenes.
Simila ly, Ba ios e al.
[37]
ca ied ou a compa ison employing
a GENIE M640 CCD came a and a non-comme cial e en
CMOS came a connec ed o a Powe link IEEE 61 158 indus ial
ne wo k o con olling a wo-axis plana obo du ing objec
acking. The esul s showcase he e en came a’s capabili y o
enable he obo o ack he a ge wi h g ea e speed, accu acy,
and s abili y, especially unde a ying ligh condi ions. The wo k
o Censi e al.
[38]
p oposed a o mal e alua ion o di e se senso
amilies using a powe -pe o mance cu e. The s udy ocused on
con as ing adi ional CCD/CMOS senso s wi h neu omo phic
ision senso s and e ealed he ask-dependen dominance o di -
e en senso s ac oss a ious sensing powe anges. Cox e al.
[39]
in oduced a heo e ical me hodology o assess he pe o mance
o e en and ame came as. Thei app oach in ol ed employing
sys em-le el models and su oga e pe o mance me ics o a -
ge ecogni ion asks.
Fu he mo e, da a-p ocessing pe spec i es ha e been explo ed
in compa a i e s udies be ween ame and e en came as.
Fa abe e al.
[40]
conduc ed a compa ison be ween ame-based
con olu ional neu al ne wo ks and ame- ee spiking neu al
ne wo ks o objec ecogni ion applica ions. Implemen a ion
examples using VLSI chips and field-p og ammable ga e a ay
(FPGAs) we e p o ided, analyzing di e ences in compu a ional
speed, scalabili y, mul iplexing, and signal ep esen a ion.
Figu e 1. Hyb id: one o he o ni hop e s de eloped by he GRVC Robo ics
Lab employed o eco ding he da ase used in his wo k.
www.ad ancedsciencenews.com www.ad in ellsys .com
Ad . In ell. Sys . 2025, 2401065 2401065 (2 o 20) © 2025 The Au ho (s). Ad anced In elligen Sys ems published by Wiley-VCH GmbH
26404567, 0, Downloaded om h ps://ad anced.onlinelib a y.wiley.com/doi/10.1002/aisy.202401065 by Readcube (Lab i a Inc.), Wiley Online Lib a y on [13/05/2025]. See he Te ms and Condi ions (h ps://onlinelib a y.wiley.com/ e ms-and-condi ions) on Wiley Online Lib a y o ules o use; OA a icles a e go e ned by he applicable C ea i e Commons License
Rebecq e al.
[41,42]
p oposed an image econs uc ion me hod
based on a ecu en neu al ne wo k ained wi h simula ed
e en s, e alua ing and compa ing he quali y o he econs uc ed
images using s anda d compu e ision algo i hms such as
isual-ine ial odome y and objec classifica ion. Fu he , some
s udies shed ligh on he use o e en s and ames o specific
asks (e.g., he wo k om Rod íguez-Gómez e al.
[43]
on isual
s abiliza ion).
2.2. Flapping-Wing Robo Pe cep ion
The de elopmen o online onboa d pe cep ion sys ems o
flapping-wing obo s is a challenging p oblem. T adi ionally,
con ol and guidance me hods o o ni hop e s ha e elied on
ex e nal senso s such as mo ion cap u e sys ems.
[44–46]
In addi-
ion, some s udies ha e pe o med o -boa d p ocessing o
isual senso s.
[6,7,47–51]
Howe e , ecen ad ances in he size
and weigh educ ion o isual senso s and p ocesso s ha e
enabled he in eg a ion o ully onboa d pe cep ion sys ems
o o ni hop e s. One o he fi s me hods p oposed o o ni hop-
e s was he obs acle a oidance me hod o Wag e e al.
[11]
and
Tijmons e al.
[12]
which used a ligh weigh s e eo came a wi h
a CPU module o de ec s a ic obs acles using dispa i y maps.
Mo e ecen ly, Gómez e al.
[15]
p esen ed an o ni hop e guid-
ance me hod based on e en came as. The algo i hm acks line
pa e n e e ences om e en s and eeds a isual se oing con-
olle o guide he obo owa d a goal posi ion. Rod íguez-
Gómez e al.
[16]
p esen ed an e en -based dynamic obs acle
a oidance me hod o o ni hop e s, which quickly de ec s mo -
ing obs acles and igge s e asi e ac ions con olling o ni hop e
ail deflec ions. The use o e en -based equency p ocessing
onboa d a flapping-wing obo is explo ed in Tapia e al.
[52]
These wo ks use di e en ypes o ision senso s ( ame-based,
e en -based, and/o s e eo se ups) bu do no conclude which
senso is he mos sui able o flapping-wing obo s. In a p e i-
ous wo k,
[20]
we compa ed ame and e en came as o o ni-
hop e s, concluding ha al hough e en came as ha e a lowe
le el o ma u i y, hey a e mo e sui able o flapping-wing obo
onboa d pe cep ion han ame-based came as. This a icle ep-
esen s a s ep o wa d and e alua es he use o he comple e
e en -based ision sys em o o ni hop e onboa d pe cep ion
om a comp ehensi e pe spec i e by assessing 1) he ope a-
ional limi s o e en -based ision conside ing e ec s such as
ib a ion le el; 2) he ease o in eg a ion o e en came as on
boa d o ni hop e s; and 3) he capaci y o e en -based sys ems
o pe o m di e en pe cep ion asks on boa d ou flapping-wing
obo s indoo s and ou doo s.
3. Fligh Challenges
This sec ion in ends o answe Q1: A e e en came as sui able o
online onboa d pe cep ion conside ing he challenging fligh condi-
ions o flapping-wing obo s? Fo his pu pose, i is necessa y
o conside wha hese challenging condi ions a e and how hey
a ec onboa d ision sys ems. As s a ed in e . [20], he fligh o
he o ni hop e s imposes subs an ial equi emen s on he
onboa d ision sys em. In sea ch o e ficiency and obus ness,
we need o know he ope a ional limi s o e en -based ision
unde se e al aspec s.
Agile mo emen s, s ong ib a ions, and changes in il angle
caused by flapping s okes lead o sudden changes in he isual
scene cap u ed by he came a. The e o e, he onboa d ision sys-
em equi es a high empo al esolu ion o ensu e ha hese
apid changes can be pe cei ed. The abili y o cap u e da a wi h
a p ecise empo al esolu ion enables a high deg ee o espon-
si eness and eac i e decision-making, which is pa icula ly c i -
ical in dynamic scena ios. Howe e , high empo al esolu ion is
gene ally associa ed wi h gene a ing a la ge amoun o in o ma-
ion du ing fligh . Hence, i is also necessa y o analyze he band-
wid h o he onboa d senso s. In addi ion, o a oid in o ma ion
loss, he ision sys em mus be obus o he mo ion blu ha can
be caused by he apid shi s and ib a ions men ioned ea lie .
Al hough e en came as a e significan ly obus o blu , hey a e
no en i ely insensi i e.
[53]
Mo eo e , ib a ions, as mo ions, o
changes in he il angle can cause ab up fluc ua ions in he
ligh ing cap u ed by he came a. This po en ially leads o in o -
ma ion loss as he came a ails o adap o changing ligh ing
condi ions. Hence, a high dynamic ange is c i ical, pa icula ly
in ou doo scena ios. The analysis o ope a ion in da k ligh ing
condi ions is also ele an o indoo applica ions.
We ha e expe imen ally analyzed he esponse o e en -based
ision sys ems in e ms o 1) empo al esolu ion; 2) bandwid h;
3) mo ion blu ; 4) dynamic ange; and 5) ope a ion in da k con-
di ions. These analyses include expe imen s in es benches
designed o mimic he flapping-wing fligh condi ions.
3.1. Tempo al Resolu ion
O ni hop e ib a ion and agile maneu e s equi e pe cep ion
sys ems capable o e ec i ely cap u ing and analyzing objec s
wi h e y high empo al esolu ion. This is pa icula ly c ucial,
o ins ance, o sense-and-a oid sys ems. Unlike adi ional
came as ha ope a e by cap u ing a fixed numbe o ames
pe second, e en came as cap u e isual in o ma ion asynch o-
nously, hence enabling empo al esolu ions ha a e mainly
limi ed by he came a’s e ac o y pe iod. To illus a e he po en-
iali ies o e en senso s in e ms o empo al esolu ion, we
designed an expe imen in which di e en came as we e used
o de ec a blinking LED, whose equency a ies om 1 Hz
o 1 kHz. The empo al esolu ion o e en -based came as p o-
ides a significan ad an age o de ec ing and acking apid
mo emen s. In he con ex o o ni hop e pe cep ion, i is essen-
ial o conside he combined e ec s o he flapping mo ion and
he de ec ed a ge dynamics. These ac o s can inc ease he ope -
a ional equency demand o he de ec ion sys em. The e o e, he
ope a ional flapping equency o he o ni hop e (yellow a ea in
Figu e 2) should be ega ded as a minimum equi emen o he
onboa d de ec ion sys em. The dynamic ision senso (DVS) o a
DAVIS346, a DVXplo e Mini, and wo ame-based senso s
(wi h ame a es 30 and 40 Hz) we e e alua ed. To a oid any
possible deg ada ion p oduced by mo ion blu , see discussion
in Sec ion 3.3, he LED was placed o co e a su ficien numbe
o pixels, ensu ing i s de ec ion e en i some pixels did no p o-
duce e en s. In he case o he DVS and he DVX, a blink o he
LED is de ec ed as wo consecu i e e en s wi h opposi e pola i ies
www.ad ancedsciencenews.com www.ad in ellsys .com
Ad . In ell. Sys . 2025, 2401065 2401065 (3 o 20) © 2025 The Au ho (s). Ad anced In elligen Sys ems published by Wiley-VCH GmbH
26404567, 0, Downloaded om h ps://ad anced.onlinelib a y.wiley.com/doi/10.1002/aisy.202401065 by Readcube (Lab i a Inc.), Wiley Online Lib a y on [13/05/2025]. See he Te ms and Condi ions (h ps://onlinelib a y.wiley.com/ e ms-and-condi ions) on Wiley Online Lib a y o ules o use; OA a icles a e go e ned by he applicable C ea i e Commons License
(ON and OFF) in he co esponding pixels. Fo he ame cam-
e as, an ab up change in he mean in ensi y o he LED’s pixels
is compu ed.
As shown in Figu e 2, DVS and DVX can co ec ly de ec
blinking LEDs in he en i e ange o equencies analyzed.
In con as , ame-based de ec ion o he blinking equency is
limi ed by he Nyquis equency— ha is, hal o he came a
ame a e: 15 and 20 Hz, espec i ely. A ame came a wi h a
empo al esolu ion simila o ha o he DVS o he DVX would
equi e a ame a e o >2 kHz, which would equi e bigge and
hea ie came as ha a e a beyond he payload capabili ies o
o ni hop e s. Besides, p ocessing ames a high equencies
equi es dedica ed ha dwa e (e.g., GPU, pa alleliza ion) while
inc easing payload and powe consump ion. Resul s o highe
equencies a e no shown because he e en came a’s e ac o y
pe iod would de e io a e he ON/OFF igge de ec ion.
3.2. Bandwid h
The ligh weigh compu e s ha can be moun ed on boa d o ni-
hop e s impose se e e limi a ions in e ms o compu ing powe
(i.e., bi /s ha can be p ocessed). S anda d came as su e om
he bandwid h-la ency ade-o (BW ∝1/Δ ). Con e sely, e en
came as a e no go e ned by a ame a e and, hence, can achie e
a a iable bandwid h while main aining a low la ency. Gi en an
elec onic came a configu a ion (i.e., he cu en biases ha con-
ol bandwid h, con as h eshold, and e ac o y pe iod, among
o he s), he e en gene a ion a e depends mainly on he ela i e
mo ion be ween he came a and he scene. The e o e, i is nec-
essa y o quan i a i ely e alua e he amoun o in o ma ion o be
p ocessed in bo h ame and e en came as. Fo ha pu pose, we
analyzed he numbe o pixels (APS) and e en s (DVS) gene a ed
pe second by he DAVIS346 e en came a in Socce ,Tes bed, and
Hills scena ios om he GRIFFIN Pe cep ion Da ase .
[4]
Bo h
senso s ha e he same esolu ion (346 260), ha e hei pixel
coo dina es aligned, and sha e he same op ics (i.e., same
FoV, AFoV, ocal leng h, dis o ion, e c.). Table 1 shows he num-
be o e en s and pixels gene a ed pe second o h ee di e en
scena ios in he da ase . I is wo h no ing ha he alues a e
lowe o he DVS since i only cap u es changes in b igh ness,
hence educing he amoun o edundan in o ma ion om he
scene (e.g., s a ic backg ound such as open sky). This can also be
in e ed in Figu e 3, whe e he numbe o e en s and pixels pe
millisecond in he sequence Hills Base 1 a e shown. The influ-
ence o he di e en o ni hop e fligh s ages on e en gene a ion
can be easily no iced: 1) launching (low e en gene a ion); 2) flap-
ping (high e en gene a ion); and 3) landing (ab up e en gen-
e a ion). We can conclude ha ame-based senso s su e om
o e sampling du ing mos o he fligh bu also om unde sam-
pling du ing ce ain agg essi e maneu e s (e.g., o ni hop e
landing a =38 s a Figu e 3). In con as , he abili y o e en
came as o gene a e in o ma ion acco ding o he scene dynam-
ics con ibu es o an enhancemen in in o ma ion cap u e
e ficiency.
3.3. Mo ion Blu
Global and olling shu e ame-based came as su e om
mo ion blu , mainly when he exposu e ime is ela i ely long
wi h espec o he dynamic o he isual scene. In con as , e en
came as a e mo e obus agains mo ion blu due o hei asyn-
ch onous na u e, low la ency, and high empo al esolu ion.
O ni hop e flapping s okes p oduce agg essi e il ing mo ions
ha a ec he in o ma ion cap u ed by he came as.
[4]
Al hough
he wo k in e . [54] p esen s a mechanical s abilize o educe il
mo ion, he s ic payload equi emen s o flapping-wing obo s
es ic he in eg a ion o gimbal de ices. Fu he mo e, al hough
some so wa e deblu ing o s abiliza ion app oaches may be
Figu e 2. Measu ed LED blinking equencies wi h wo e en -based cam-
e as (Async.) and wo adi ional came as wi h ame a es o 30 and 40 Hz.
The e ec o aliasing can be no iced wi h blinking equencies beyond
15 and 20 Hz o he ame-based came as. The yellow a ea co esponds
o he ypical ope a ional flapping equency o epo ed o ni hop e s (see
Table 2 in Sec ion 4; excluding hose pla o ms weighing less han 3 g o
being unable o ca y any ision sys em).
Table 1. Mean, median, and maximum numbe o pixels and e en s gene a ed in all he Socce ,Hills, and Tes bed sequences.
[4]
MEPS s ands o million
e en s pe second, MPPS s ands o million pixels pe second.
Socce Hills Tes bed
Mean Median Max Mean Median Max Mean Median Max
APS [MPPS] 3.60 3.60 3.60 3.60 3.60 3.60 3.60 3.60 3.60
DVS [MEPS] 0.42 0.39 8.35 0.39 0.30 8.69 0.91 0.80 4.16
DVS [%]
a)
3.50 3.25 69.58 3.25 2.50 72.42 7.58 6.67 34.67
a)
Assuming a maximum bandwid h o 12 MEPS [ e . 18, Table 1].
www.ad ancedsciencenews.com www.ad in ellsys .com
Ad . In ell. Sys . 2025, 2401065 2401065 (4 o 20) © 2025 The Au ho (s). Ad anced In elligen Sys ems published by Wiley-VCH GmbH
26404567, 0, Downloaded om h ps://ad anced.onlinelib a y.wiley.com/doi/10.1002/aisy.202401065 by Readcube (Lab i a Inc.), Wiley Online Lib a y on [13/05/2025]. See he Te ms and Condi ions (h ps://onlinelib a y.wiley.com/ e ms-and-condi ions) on Wiley Online Lib a y o ules o use; OA a icles a e go e ned by he applicable C ea i e Commons License
easible o low le els o blu ,
[55]
hey in oduce a significan delay
in he p ocessing pipeline ha p e en s eal- ime ope a ion.
Despi e he highe obus ness o mo ion blu when compa ed
o ame-based came as, [ e . 20, Sec ion 5.2], e en came as can
also su e om mo ion blu unde ce ain condi ions. Howe e ,
he e ec obse ed in an e en -based senso when he ela i e
came a-scene mo ion inc eases is qui e di e en om ha expe-
ienced by a con en ional senso . Acco ding o Benosman
e al.
[53]
as he speed inc eases, mo ion blu causes e en s o o m
clus e s a he han sha p edges, esul ing in a spa se mo ion
flow in e en -based sys ems. Addi ionally, he came a does no
gene a e enough e en s o all spa ial loca ions, and as a esul ,
some pixels a e no ac i a ed. This e ec , p oduced by he la en-
cies o he senso when cap u ing he ligh , was e alua ed by
moun ing a DAVIS346 on a pla o m ha mimics he ho izon al
flapping mo ion o o ni hop e s; see Appendix A. The came a
was se poin ing owa d a black can as wi h a whi e ho izon al
line unde uni o m and cons an illumina ion condi ions, and
he came a-can as dis ance was such ha he line was p esen
wi hin he came a FoV du ing he whole expe imen . E en s
a e mainly igge ed by he changes in pixels’in ensi ies caused
by he pla o m’s oscilla o y mo ion. Thus, in hese expe imen s,
e en s a e igge ed a he p ojec ions o he line edges on he
image plane. Du ing he expe imen , he pla o m oscilla es fi s
a 5.0 Hz and hen g adually educes o 2.5 Hz.
To es ima e he mo ion blu , we composed e en images by
accumula ing a fixed numbe o e en s in o ames and d awing
hem as black pixels in a whi e image. Figu e 4 shows he e en
ames ob ained by accumula ing 1000 e en s a di e en
oscilla ion equencies. A highe equencies, mo ion blu
becomes isible as holes (whi e pixels) among he black pixels
ha desc ibe he shape o he line. Ano he e ec can be
obse ed: he hickness o he line inc eases wi h highe equen-
cies. This occu s since he numbe o ac i a ed pixels dec eases
wi h he inc ease in equency (i.e., he line’s speed), as p e i-
ously discussed. We define he pe cen age o igge ed e en s
(PTE) as he numbe o igge ed e en s di ided by he a ea
o he pa ch caused by he line in each e en ame. Tha is,
PTE is an indica o o e en s los by mo ion blu on e en
came as. Figu e 5 shows he expe imen al esul s ob ained.
A 5.0 Hz, PTE is a ound 74.38% and inc eases as he oscilla ion
equency educes, e idencing ha flapping s okes p oduce
mo ion blu in e en came as. This expe imen canno di ec ly
measu e he o al numbe o los da a, as he senso migh lose
one o mo e e en s igge ed a he same pixel, bu i p o ides a
alid app oxima ion.
3.4. Dynamic Range
Due o mechanical ib a ions and sudden il angle changes due
o he flapping s okes, he came as on boa d o ni hop e s can
ha e sudden d as ic changes in ligh ing condi ions, including
b igh ly li scenes, da k scenes, and also cases in which he cam-
e a FoV includes bo h b igh and da k scene pa s a he same
ime. The e o e, dynamic ange is a c i ical aspec o conside
when selec ing he onboa d senso s. In [ e . 20, Sec ion 5.1],
a DVS senso was e alua ed o A Uco de ec ion in e en -
econs uc ed images unde di e en dynamic ange condi ions
and p esen ed an excellen pe o mance o up o 80 dB. In ha
expe imen , he noise and he insu ficien numbe o e en s p e-
en ed he co ec econs uc ion used o he de ec ion wi h
s ong dynamic ange condi ions. To emo e ha influence
and complemen he analysis, we designed a se up consis ing
o a spinning do (a black o a ing disk wi h a whi e do on
i ), which was pa ially illumina ed wi h a s ong ligh sou ce
(mimicking a si ua ion simila o a isual scene pa ially a ec ed
by he sunligh ). The disk was spinning a a cons an eloci y.
The eloci y was low enough o assume ha mo ion blu was
no p oduced. We poin ed ou DAVIS346 came a owa d he disk
oge he wi h o he ame-based came as (In el Realsense D345
and ELP Mini720p; he eina e RS and ELP, espec i ely) o
Figu e 3. Numbe o e en s o pixels pe millisecond gene a ed espec-
i ely by DVS and APS in Hills Base 1 sequence.
[4]
A–C) co espond o
he o ni hop e launching, flapping, and landing, espec i ely. No a ion:
MEPS s ands o million e en s pe second, MPPS s ands o million pixels
pe second.
Figu e 4. E en ames om he mo ion blu expe imen s while he e-
quency o he mo o ized pla o m is a ied om 2.5 (le ) o 5.0 Hz ( igh ).
Each ame was ende ed by accumula ing 1000 e en s. The e en s a e less
sca e ed (in ol ing highe PTE) when educing he equency o he pla -
o m oscilla ions ( om le o igh ).
Figu e 5. Pe cen age o igge ed e en s (PTE, ed) when a ying he e-
quency o he flapping pla o m (blue). The numbe o los e en s
inc eases wi h he equency o he pla o m oscilla ions.
www.ad ancedsciencenews.com www.ad in ellsys .com
Ad . In ell. Sys . 2025, 2401065 2401065 (5 o 20) © 2025 The Au ho (s). Ad anced In elligen Sys ems published by Wiley-VCH GmbH
26404567, 0, Downloaded om h ps://ad anced.onlinelib a y.wiley.com/doi/10.1002/aisy.202401065 by Readcube (Lab i a Inc.), Wiley Online Lib a y on [13/05/2025]. See he Te ms and Condi ions (h ps://onlinelib a y.wiley.com/ e ms-and-condi ions) on Wiley Online Lib a y o ules o use; OA a icles a e go e ned by he applicable C ea i e Commons License
de ec he do . We eco ded a sequence o 30 s o each came a
and hen p ocessed he images using a ci cle de ec ion algo i hm
based on he Hough ans o m. A de ec ion is conside ed when
he numbe o o es ecei ed in a Hough space cell is highe han
a h eshold τ=0.6τ
max
, whe e τ
max
is compu ed by assuming ha
all pixels o he do pe ime e p oduce a o e in he same coo -
dina es. As a esul , he algo i hm was able o ejec alse posi-
i es while main aining an accu a e de ec ion. The inpu images
o he Hough de ec ion we e ob ained using Canny ( o ame-
based came as) o e en images gene a ed by accumula ing
e en s in he empo al windows o 3 ms. Then, we compu ed
he numbe o do de ec ions pe o med in di e en disk sec o s.
Figu e 6 shows a pola his og am wi h he numbe o do de ec-
ions in each disk sec o o each senso ; he disk sec o a ec ed
by he ligh sou ce is shown in g ay colo . The DVS was able o
de ec he do in he illumina ed a ea, and he numbe o
de ec ions was simila o ha in he non-illumina ed sec o .
In con as , all ame-based came as ailed o de ec he do in
he illumina ed a ea due o he senso sa u a ion.
3.5. Da k Ligh ing Condi ions
Pe cep ion using isual senso s becomes pa icula ly challenging
in da k scenes. Al hough, in some cases, inc easing he exposu e
ime o ame-based senso s can mi iga e he p oblem, i migh
also inc emen he mo ion blu and noise le el. We e alua ed he
pe o mance o DAVIS346’s DVS and ac i e pixel senso (APS),
ELP, and RS unde h ee di e en condi ions: pi ch-da k (0 lx),
da k (5 lx), and well-li condi ions (100 lx). All came as we e
se wi h pa allel op ical axes ha poin ed o a pa e n wi h ou
lines. The pa e n was mo ed slowly o gene a e e en s while
minimizing mo ion blu . In addi ion, he h ee di e en ligh ing
condi ions we e applied sequen ially (10 s be ween e e y
change). Line de ec ion was pe o med using he Hough ans-
o m applied o ames and e en images (gene a ed by accumu-
la ing 1000 e en s pe image, see Figu e 7). Fo a ai e alua ion,
he ame-based senso s we e configu ed wi h au oexposu e.
DVS was he only senso capable o de ec ing lines du ing he
whole expe imen . As expec ed, pi ch-da k condi ions hinde line
de ec ion o ame-based came as e en wi h high exposu e
imes. Con e sely, al hough he ab up change in he ligh ing
condi ions gene a es e en s ha lead o alse line de ec ions,
he e en came a was able o de ec all lines in all he es ed ligh -
ing condi ions sa is ac o ily.
3.6. Discussion
E en came as demons a e a s ong po en ial o onboa d pe -
cep ion in flapping-wing obo s by add essing he challenges
posed by hei dynamic fligh . Thei asynch onous na u e allows
hem o cap u e only ele an scene changes, educing da a load
while main aining esponsi eness du ing agile maneu e s.
Al hough mo ion blu can a ec ame-based came as’pe o -
mance unde ex eme condi ions (e.g., ib a ions and agg essi e
flapping s okes), e en came as emain mo e obus . The eli-
abili y and high au onomy o flapping-wing obo s make hem
sui able pla o ms o bo h indoo and ou doo ope a ions.
Hence, a high dynamic ange and good pe o mance unde di -
e en ligh condi ions a e p ope ies o be emphasized. All he
esul s p esen ed in his sec ion highligh he sui abili y o e en
came as o onboa d ision in o ni hop e s. Howe e , he g ow-
ing in e es in R&D o o ni hop e echnology is leading o pla -
o ms wi h highe payload capaci ies, wi h smoo he flapping
s okes, and capable o pe o ming smoo he ajec o ies.
This may enable he in eg a ion o mul i-senso app oaches.
The use o e en came as wi h addi ional senso s can enhance
he pe o mance and eliabili y o pe cep ion sys ems by mi iga -
ing indi idual senso limi a ions, leading o a mo e obus and
adap able pe cep ion sys em h ough da a usion.
4. In eg a ion Challenges
This sec ion aims o answe Q2: A e e en came as sui able o be
in eg a ed (HW&SW) on flapping-wing obo s conside ing he
cons ain s o hese pla o ms? The s ic payload and weigh dis i-
bu ion equi emen s o he flapping-wing obo s se e ely es ic
he in eg a ion o onboa d senso s. The size and weigh o he
came a a e c i ical and can se e ely a ec he fligh abili y and
con ollabili y o hese pla o ms. Recen ad ances in e en -based
echnology poin o an inc easing minia u iza ion o de ices wi h
Figu e 6. Do de ec ions o di e en disk sec o s exp essed as a pola
his og am. Each sec o comp ises a ange o 18%. The heigh o he ba s
ep esen s he numbe o de ec ions p oduced in each angula ange.
The g ay a ea co esponds o he s ongly illumina ed sec o s.
Figu e 7. E en images gene a ed by accumula ing 1000 e en s unde di -
e en ligh ing condi ions ( om le o igh : pi ch-da k, 0 lx; da k, 5 lx; and
well-li condi ions, 100 lx).
www.ad ancedsciencenews.com www.ad in ellsys .com
Ad . In ell. Sys . 2025, 2401065 2401065 (6 o 20) © 2025 The Au ho (s). Ad anced In elligen Sys ems published by Wiley-VCH GmbH
26404567, 0, Downloaded om h ps://ad anced.onlinelib a y.wiley.com/doi/10.1002/aisy.202401065 by Readcube (Lab i a Inc.), Wiley Online Lib a y on [13/05/2025]. See he Te ms and Condi ions (h ps://onlinelib a y.wiley.com/ e ms-and-condi ions) on Wiley Online Lib a y o ules o use; OA a icles a e go e ned by he applicable C ea i e Commons License
inc easing spa ial esolu ion—now o e 1 Mpx.
[34,56]
In addi ion,
he limi ed payload o he o ni hop e p e en s he use o high-
capaci y ba e ies. Hence, low ene gy consump ion becomes a
c i ical aspec o be conside ed. F ame-based came a echnology
has been in de elopmen o yea s. Consequen ly, he e a e a
la ge numbe o algo i hms, da ase s, and benchma ks a ailable.
On he con a y, e en -based ision does no ha e such a long
his o y, and he di ec use o ame-based algo i hms and da a
is no always easible. The e o e, om a so wa e in eg a ion
poin o iew, i is also necessa y o analyze he a ailabili y o
esou ces and suppo ools. In his sec ion, we analyze he in e-
g abili y o e en came as onboa d flapping-wing obo s by
e iewing senso minia u iza ion, esolu ion, ene gy consump-
ion, and he a ailabili y o esou ces and suppo ools.
Di e en la ge-scale flapping-wing ae ial pla o ms ha e been
de eloped in ecen yea s, such as RoBi d,
[57]
Fes o Sma Bi d
(h ps:// es o.com), RoboRa en,
[58–61]
Beihawk.
[62]
O he bioins-
pi ed pla o ms wi h a sho e wingspan a e Ba Bo ,
[63]
Do e,
[64]
Thunde I,
[65]
and UST-Bi d.
[66]
Besides hese pla o ms, o he
small-scale and insec -like solu ions
[67–69]
a e DelFly,
[6,11,70]
Flappe Nimbleþby Flappe D ones (h ps://flappe -d ones.
com), Nano Hummingbi d,
[71]
KUBee le-S,
[72,73]
Me aFly by
BionicBi d (h ps://bionicbi d.com), and Robobee
[45,74–78]
(which
inspi ed he obo in e . [49,50] and Beeþ).
[79]
The GRIFFIN ERC
Ad anced G an p ojec (GRIFFIN (Ac ion 7 882 479): Gene al
complian ae ial Robo ic manipula ion sys em In eg a ing
Fixed and Flapping-wings o INc ease ange and sa e y.
h ps://g i fin-e c-ad anced-g an .eu) aims o de elop no el o ni-
hop e s wi h ad anced pe cep ion and manipula ion capabili ies.
One o he pla o ms de eloped in his p ojec is E-Flap,
[5,80–83]
he
fi s flapping-wing pla o m wi h onboa d p ocessing and pe cep-
ion capabili ies, a ema kable payload, and he abili y o fly a low
speed, enabling he in e ac ion wi h he en i onmen .
[84–86]
In addi ion, he Hyb id obo
[87]
o e s au onomous na iga ion
capabili ies wi h an onboa d au opilo capable o swi ching
be ween fixed-wing and flapping-wing fligh modes. O he
esea ch p ojec s o g ea ele ance include he Po Wings ERC
Ad anced G an
[88,89]
and he DelFly P ojec .
[2,6]
To b idge he
gap be ween he expe imen al se up and eal-wo ld condi ions,
we conduc ed an analysis o examine he easibili y o in eg a ing
he e en came as de ailed in Table 2 in o se e al flapping-wing
obo s (encompassing da a om bo h esea ch li e a u e and com-
me cially a ailable designs) lis ed in Table 3. The main goal is o
p o ide insigh s in o he possibili y o enhancing he pe o mance
o he obo s and expanding hei ope a ional capabili ies h ough
he use o ad anced e en -based pe cep ion echnology.
Table 2. Summa y o he main specifica ions o some e en came as om a ious companies. Da a collec ed om
[18]
and om manu ac u es’da ashee s
and p oduc b ie s.
Came a Senso
Weigh
a)
[g] Volume [mm] Consump ion Max BW [MEPS] La ency [μs] Spa ial Res. [px]
iniVa ion
d)
DVS128 [26] 65 40 60 25 60 mA@5 1 >12 128 128
DVS240 [32] 75 40 60 25 180 mA@5 12 >12 240 180
DAVIS240 [32] 75 56 55 27 180 mA@5 12 >12 240 180
DAVIS346 –100 40 60 25 180 mA@5 12 <1000 346 260
DVXplo e –100 40 60 25 140 mA@5 165 <1000 640 480
DVXplo e Li e –75 40 60 25 140 mA@5 100 <1000 320 240
DVXplo e Mini –21 29 29 32 140 mA@5 450 <1000 640 480
P ophesee EVK1 ATIS [30] 82.5 60 38 50 50–175 mW
b)
–>3 304 240
Gen3 ATIS –82.5 60 38 50 25–87 mW
b)
66 40–200 480 360
Gen3 CD –82.5 60 38 50 36–95 mW
b)
66 40–200 640 480
Gen4 CD [27] 82.5 60 38 50 32–84 mW
b)
1066 20–150 1280 720
EVK2 Gen4 CD [27] 260 102 58 42 7500 mW 1066 220 1280 720
EVK3 Gen3 CD –112 108 76 45 4500 mW 1.6 Gbps 220 640 480
Gen4 CD [27] 112 108 76 45 4500 mW 1.6 Gbps 220 1280 720
GenX320 –112 108 76 45 4500 mW 1.6 Gbps 220 320 320
EVK4 IMX636Es
c)
–40 30 30 36 500 mW 1.6 Gbps 220 1280 720
Samsung DVS-Gen2 [33] no case no case 27–50 mW
b)
300 65–410 640 480
DVS-Gen3 –no case no case 40 mW
b)
600 50 640 480
DVS-Gen4 [34] no case no case 130 mW
b)
1200 150 1280 960
CelePixel
e)
CeleX-IV [156] no case no case –200 >10 768 640
CeleX-V [56] no case no case 400 mW
b)
140 >8 1280 800
Insigh ness
)
Rino3 –15 350 350 20–70 mW
b)
20 125 320 262
a)
Excluding lens;
b)
E en senso powe consump ion;
c)
IMX636ES ealized in collabo a ion be ween Sony and P ophesee;
d)
now pa o SynSense;
e)
now pa o OmniVision,
Will Semiconduc o ;
)
now pa o Sony.
www.ad ancedsciencenews.com www.ad in ellsys .com
Ad . In ell. Sys . 2025, 2401065 2401065 (7 o 20) © 2025 The Au ho (s). Ad anced In elligen Sys ems published by Wiley-VCH GmbH
26404567, 0, Downloaded om h ps://ad anced.onlinelib a y.wiley.com/doi/10.1002/aisy.202401065 by Readcube (Lab i a Inc.), Wiley Online Lib a y on [13/05/2025]. See he Te ms and Condi ions (h ps://onlinelib a y.wiley.com/ e ms-and-condi ions) on Wiley Online Lib a y o ules o use; OA a icles a e go e ned by he applicable C ea i e Commons License
4.1. Minia u iza ion
Simila ly o ame-based ision sys ems, he imp o emen in
e en ision echnology has led o smalle senso s, as e idenced
in Figu e 8. This clea end owa d senso minia u iza ion
endo ses he possibili y o de eloping smalle and ligh e e en
came as. Figu e 9 shows he e olu ion in weigh and olume o
some comme cial de ices manu ac u ed by iniVa ion, one o
he mos ele an e en came a companies. The e is a widesp ead
end among se e al e en came a manu ac u e s owa d achie -
ing le els o minia u iza ion compa able o hose o ame-based
ision senso s. Rele an is o men ion SPECK (h ps://www.syn-
sense.ai/p oduc s/speck-2) by SynSense, he fi s ully e en -based
neu omo phic ision sys em-on-chip (SoC) ha weighing a ew
g ams in eg a es a DVS. To assess he easibili y o in eg a ing
e en came a models in o exis ing flapping-wing obo s, Table 4
Table 3. Summa y o he main cha ac e is ics o some flapping-wing obo s om a ious esea ch ins i u ions and manu ac u e s wi h hei
specifica ions.
Weigh [g] Wingspan [m] Flap. F eq. [Hz] Payload [g] Ba e y [mAh@LiPo cells]
Robobee
[45]
Ha a d U. 0.075 0.035 120 ––
Fou -wings
[157]
UW 0.143 0.056 160 ––
Beeþ
[79]
USC 0.095 0.033 100 ––
DelFly Mic o
[6]
TU Del 3.07 0.10 30 0.4
a)
100-120@1S
DelFly Explo e
[11]
TU Del 20 0.28 10–14 4
a)
250@1S
DelFly Nimble
[70]
TU Del 29 0.33 17 4
a)
250@1S
Flappe Nimbleþ
b)
Flappe D ones 102 0.49 12–20 25 300@2S
N. Hummingbi d
[71]
Ae oVi onmen 19 0.165 –– –
KUBee le-S
[73]
Konkuk U. 15.8 0.20 18 4 160@1S
Me aFly
c)
BionicBi d 9.5 0.29 –– –
Me aBi d
d)
BionicBi d 9.5 0.33 –– –
X-Fly
e)
BionicBi d 12 0.38 –– 60@1S
Ba Bo
[63]
UIUC 93 0.469 10 ––
Do e
[64]
NWPU 220 0.50 12 ––
Thunde I
[65]
NMSU 350 0.70 5.88 –800@3S
UST-Bi d
[66]
USTB 83.2 0.80 4 18 180@3S
USTB-Hawk
[158]
USTB 985 1.78 4.108 192 7000@3S
RoboRa en I
[61]
UMD 285 1.168 4 43.8 370@2S
RoboRa en II
[61]
UMD 301.6 1.330 4 80.4 370@2S
RoboRa en III
[61]
UMD 317 1.330 4 71 370@2S
RoboRa en IV
[61]
UMD 438.1 1.168 4 272.9 370@2S
RoboRa en V
[61]
UMD 438.1 1.168 4 272.9 950@2S
Beihawk
[62]
Beihang U. 1200 1.50 10 ––
E-Flap
[5]
U. Se ille 510 1.50 5.5 520 450@4S
Hyb id
[87]
U. Se ille 930 1.50 3.0 300 450@4S
RoBi d B. Eagle
[57]
U. Twen e
l)
2100 1.76 4 1000 –
RoBi d P. Falcon
[57]
U. Twen e
l)
730 1.12 5.5 100 –
Sma Bi d
)
Fes o 450 2.00 –– 450@2S
BionicOp e
g)
Fes o 175 0.63 15–20 ––
eMo ionBu e flies
h)
Fes o 32 0.50 1–2–7.4@2S
BionicFlyingFox
i)
Fes o 580 2.28 –– –
BionicSwi
j)
Fes o 42 0.68 –– –
BionicBee
k)
Fes o 34 0.24 15–20 –300@1S
a)
Weigh o he ision sys em;
b)
h ps://flappe -d ones.com/wp/nimbleplus;
c)
h ps://www.bionicbi d.com/wo ld/me afly-page;
d)
h ps://www.bionicbi d.com/wo ld/
me abi d-page;
e)
h ps://www.bionicbi d.com/wo ld/x-fly_de ails;
)
h ps://www. es o.com/PDF_Flip/co p/Fes o_Sma Bi d/en;
g)
h ps://www. es o.com/PDF_Flip/co p/
Fes o_BionicOp e /en;
h)
h ps://www. es o.com/PDF_Flip/co p/Fes o_eMo ionBu e flies/en;
i)
h ps://www. es o.com/PDF_Flip/co p/Fes o_BionicFlyingFox/en;
j)
h ps://www. es o.com/PDF_Flip/co p/Fes o_BionicSwi /en;
k)
h ps://www. es o.com/PDF_Flip/co p/Fes o_BionicBee/en.
l)
Spin-o : Clea Fligh Solu ions, acqui ed by
AERIUM Analy ics.
www.ad ancedsciencenews.com www.ad in ellsys .com
Ad . In ell. Sys . 2025, 2401065 2401065 (8 o 20) © 2025 The Au ho (s). Ad anced In elligen Sys ems published by Wiley-VCH GmbH
26404567, 0, Downloaded om h ps://ad anced.onlinelib a y.wiley.com/doi/10.1002/aisy.202401065 by Readcube (Lab i a Inc.), Wiley Online Lib a y on [13/05/2025]. See he Te ms and Condi ions (h ps://onlinelib a y.wiley.com/ e ms-and-condi ions) on Wiley Online Lib a y o ules o use; OA a icles a e go e ned by he applicable C ea i e Commons License
p esen s a compa ison be ween he o iginal weigh s o hese cam-
e as and he payload capaci ies o he o ni hop e s. While his anal-
ysis p o ides a p elimina y insigh in o he po en ial o
inco po a ing e en -based senso s, a success ul in eg a ion would
also equi e conside a ion o addi ional ac o s such as: 1) weigh
dis ibu ion; 2) olume; 3) came a connec o and cable weigh s; o
4) weigh educ ion (e.g., eplacemen o hea y cases by ligh e
cases), among o he s.
4.2. Spa ial Resolu ion
High- esolu ion came as o e mo e de ailed in o ma ion abou
he scene, leading, o example, o fine g ain mapping o mo e
accu a e de ec ion algo i hms. E en came as a e ending
owa d highe senso esolu ions, as illus a ed in Figu e 10.
Howe e , high esolu ion in ol es some d awbacks, such as
inc eased weigh and powe consump ion. The e a e di e en
opinions on he benefi s o using high- esolu ion e en came as
o sol e s anda d compu e ision asks. Highe esolu ions
in ol e highe compu a ional esou ces o p ocess he gene a ed
e en s. In pa icula , Geh ig and Sca amuzza
[90]
poin ed ou ha
low- esolu ion e en came as can achie e be e pe o mance
han high- esolu ion came as while using significan ly less
bandwid h.
4.3. Ene gy Consump ion
O ni hop e s’payload cons ain s also a ec he onboa d ba e -
ies. The e o e, he powe consump ion o he di e en compo-
nen s moun ed onboa d equi es ca e ul conside a ion. Fi s , i
has been expe imen ally demons a ed ha flapping-wing obo s
consume in gliding fligh mode abou 90% less han in flapping
fligh mode.
[91]
Thus, planning he fligh s ages o minimize he
flapping and e ficien ly manage he ene gy is o high ele ance.
In addi ion, he esul s in e . [87] sugges a be e e ficiency o
flapping unde ce ain condi ions. The impo ance o gliding o
sa e ene gy is a majo pa adigm shi wi h espec o mul i o o
pla o ms. E en came as sui pe ec ly in his shi . Whe eas
ame-based came as always gene a e images a a cons an a e,
Figu e 8. E olu ion o pixel wid h o he main e en came a designs om
2008 o 2024. DVX s ands o DVXplo e .
Figu e 9. E olu ion o weigh and olume o some comme cial e en cam-
e as de eloped by iniVa ion.
Table 4. Compa ison o he weigh o e en -based ision came as
(Table 2) and he payload capaci y o flapping-wing obo s (Table 3).
A g een cell indica es ha he came a could be moun ed onboa d
(conside ing payload es ic ions only), while a ed cell indica es he
opposi e.
(21 g) DVX Mini
(40 g) EVK4
(65 g) DVS128
(75 g) DVS240
(75 g) DAVIS240
(75 g) DVX Li e
(82.5 g) EVK1
(100 g) DAVIS346
(100 g) DVX
(112 g) EVK3
(260 g) EVK2
DelFly Mic o (0.4 g)
DelFly Explo e (4 g)
DelFly Nimble (4 g)
KUBee le-S (4 g)
UST-Bi d (18 g)
Flappe Nimble+ (25 g)
RoboRa en I (43.8 g)
RoboRa en III (71 g)
RoboRa en II (80.4 g)
RoBi d P. Falcon (100 g)
USTB-Hawk (192 g)
RoboRa en IV (272.9 g)
RoboRa en V (272.9 g)
Hybi d (300 g)
E-Flap (520 g)
RoBi d B. Eagle (1000 g) Figu e 10. E olu ion o he esolu ion o some main e en came a com-
me cial models om 2008 o 2024. DVX s ands o DVXplo e .
www.ad ancedsciencenews.com www.ad in ellsys .com
Ad . In ell. Sys . 2025, 2401065 2401065 (9 o 20) © 2025 The Au ho (s). Ad anced In elligen Sys ems published by Wiley-VCH GmbH
26404567, 0, Downloaded om h ps://ad anced.onlinelib a y.wiley.com/doi/10.1002/aisy.202401065 by Readcube (Lab i a Inc.), Wiley Online Lib a y on [13/05/2025]. See he Te ms and Condi ions (h ps://onlinelib a y.wiley.com/ e ms-and-condi ions) on Wiley Online Lib a y o ules o use; OA a icles a e go e ned by he applicable C ea i e Commons License
on a ib a ion-isola ed pla o m. The calib a ion esul s o he
DAVIS346 IMU (da is_imu.yaml), he Vec o Na VN200
IMU ( n_imu.yaml), and he Ma ek H743 Mini V3 au opilo
IMU (ma os_imu.yaml) a e p o ided. The o iginal osbag files
(da is_imu.bag, n_imu.bag, and ma os_imu.bag) used o
ob ain he esul s a e also p o ided.
A3. C3 In insic Came a Calib a ion
We used he pinhole came a model (pa ame e s: ocal leng h
u
,
y
and p incipal poin p
u
,p
) wi h he adial- angen ial dis o ion
(pa ame e s: adial k
1
,k
2
and angen ial p
1
,p
2
). Came as we e cal-
ib a ed wi h Kalib (h ps://gi hub.com/e hz-asl/kalib )
[154,155]
using a 6 6 Ap ilG id (specifica ions in ap ilg id.yaml).
Calib a ion esul s a e p o ided as yaml files. The osbag files
(calib a ion_indoo s.bag, calib a ion_ou doo s.bag, and cali-
b a ion_ou doo s_gps.bag) used o calib a ion a e also p o-
ided in case ano he calib a ion me hod is p e e able.
A4. C4 Ex insic Calib a ion
Came a-came a and came a-IMU calib a ions we e also compu ed
wi h Kalib . The alues o
I
T
D
,
V
T
D
,
A
T
D
,and
E
T
D
(see Figu e A3)
a e p o ided in he co esponding yaml files. The es o he ans-
o ma ions can be compu ed as a composi ion o he p o ided
ma ices. The came a-imu empo al shi s we e also compu ed
by Kalib . The same osbag files used o he came a in insic cali-
b a ion can be used i ano he calib a ion me hod is p e e able.
Acknowledgemen s
This wo k was unded by he Eu opean Resea ch Council as pa o he
GRIFFIN ERC Ad anced G an 2017 (Ac ion 788247). Pa ial unding
was ob ained om he Plan Es a al de In es igación Cien ífica y
Técnica y de Inno ación o he Minis e io de Uni e sidades del
Gobie no de Espa˜na (FPU19/04692). The au ho s hank Ma io
He nández o his help wi h he design o he expe imen al se ups and
José Manuel Ca mona o his suppo du ing he flapping-wing obo fligh
expe imen s. The au ho s ex end hei g a i ude o Gee Folke sma om
he Uni e si y o Twen e, Abdessa a Abdelkefi om New Mexico S a e
Uni e si y, Hoang Vu Phan om he École Poly echnique Fédé ale de
Lausanne, Qiang Fu om he Uni e si y o Science and Technology
Beijing, and Ch is ophe de Wag e and Guido de C oon om he
Technical Uni e si y o Del o p o iding aluable de ails and specifica-
ions ega ding hei flapping-wing pla o ms. The au ho s also hank
Guille mo Gallego om he Technical Uni e si y o Be lin o his assis-
ance in ga he ing specifica ions o e en came as. R.T. also hanks
Sa a Ruiz-Mo eno o he collabo a ion and aluable ad ice.
Conflic o In e es
The au ho s decla e no conflic o in e es .
Au ho Con ibu ions
Raul Tapia: concep ualiza ion (lead); in es iga ion (lead); me hodology
(lead); supe ision (lead); isualiza ion (lead); w i ing—o iginal d a
(lead); w i ing— e iew & edi ing (lead). Ja ie Luna-San ama ia:
concep ualiza ion (suppo ing); in es iga ion (suppo ing); me hodology
(suppo ing); w i ing—o iginal d a (suppo ing); w i ing— e iew &
Table A2. Lis o indoo and ou doo sequences.
Sequence Da ase Du a ion [s] Size [GB] DAVIS346 ELP VN200 GPS MOCAP
boa ds_indoo s_1 TORRICELLI 72.398 2.358 ✓✓✓⨯✓
boa ds_indoo s_2 TORRICELLI 60.398 1.911 ✓✓✓⨯✓
human_indoo s_1 TORRICELLI 52.498 1.678 ✓✓✓⨯✓
human_indoo s_2 TORRICELLI 68.096 2.159 ✓✓✓⨯✓
boa ds_ou doo s_1 SAETA 70.995 2.434 ✓✓✓⨯⨯
boa ds_ou doo s_2 SAETA 55.198 1.939 ✓✓✓⨯⨯
gps_ou doo s_1 SAETA 160.092 1.616 ✓⨯⨯✓⨯
gps_ou doo s_2 SAETA 153.487 1.427 ✓⨯⨯✓⨯
Table A3. Lis o ROS opics.
Topic Desc ip ion F eq. [Hz]
/d s/e en s DAVIS346 e en s 30
/d s/image_ aw DAVIS346 g ayscale images 40
/d s/imu DAVIS346 IMU 1000
/elp/image_ aw ELP RGB images 30
/ n/imu Vec o Na VN200 IMU 200
/mocap/pose Op iT ack pose 120
/ma os/imu Au opilo IMU 10
/ma os/gps Au opilo GPS 3
Figu e A3. Re e ence ames used o ex e nal calib a ion: DAVIS346
came a {D} and IMU {I}, ELP came a {E}, Vec o Na VN200 {V}, and
au opilo {A}.
www.ad ancedsciencenews.com www.ad in ellsys .com
Ad . In ell. Sys . 2025, 2401065 2401065 (16 o 20) © 2025 The Au ho (s). Ad anced In elligen Sys ems published by Wiley-VCH GmbH
26404567, 0, Downloaded om h ps://ad anced.onlinelib a y.wiley.com/doi/10.1002/aisy.202401065 by Readcube (Lab i a Inc.), Wiley Online Lib a y on [13/05/2025]. See he Te ms and Condi ions (h ps://onlinelib a y.wiley.com/ e ms-and-condi ions) on Wiley Online Lib a y o ules o use; OA a icles a e go e ned by he applicable C ea i e Commons License
edi ing (suppo ing). I an Gu ie ez Rod iguez: concep ualiza ion (sup-
po ing); in es iga ion (suppo ing); me hodology (suppo ing); w i ing
—o iginal d a (suppo ing); w i ing— e iew & edi ing (suppo ing).
Juan Pablo Rod íguez-Gómez: concep ualiza ion (suppo ing); in es iga-
ion (suppo ing); me hodology (suppo ing); w i ing—o iginal d a (sup-
po ing); w i ing— e iew & edi ing (suppo ing). José Rami o Ma ínez-de
Dios: concep ualiza ion (suppo ing); supe ision (suppo ing); w i ing—
o iginal d a (suppo ing); w i ing— e iew & edi ing (suppo ing). Anibal
Olle o: unding acquisi ion (lead); supe ision (suppo ing); w i ing—
e iew & edi ing (suppo ing).
Da a A ailabili y S a emen
The da a ha suppo he findings o his s udy a e a ailable in he
supplemen a y ma e ial o his a icle.
Keywo ds
e en came as, flapping-wing ae ial obo s, unmanned ae ial sys ems
Recei ed: Decembe 7, 2024
Re ised: Janua y 28, 2025
Published online:
[1] D. Mackenzie, Science 2012,335, 1430.
[2] G. C. H. E. C oon, Sci. Rob. 2020,5, eabd0233.
[3] A. Olle o, M. Tognon, A. Sua ez, D. Lee, A. F anchi, IEEE T ans. Rob.
2022,38, 626.
[4] J. P. Rod íguez-Gómez, R. Tapia, J. L. Paneque, P. G au, A. Gómez
Eguíluz, J. R. Ma ínez-de Dios, A. Olle o, IEEE Rob. Au om. Le .
2021,6, 1066.
[5] R. Zu e ey, J. To mo-Ba be o, M. M. Guzmán, F. J. Maldonado,
E. Sanchez-Laulhe, P. G au, M. Pé ez, J. A. Acos a, A. Olle o, IEEE
Rob. Au om. Le . 2021,6, 3097.
[6] G. C. H. E. C oon, K. M. E. Cle cq, R. Ruijsink, B. D. W. Remes,
C. Wag e , In . J. Mic o Ai Veh. 2009,1, 71.
[7] F. Ga cia Be mudez, R. Fea ing, in IEEE/RSJ In . Con . on In elligen
Robo s and Sys ems S . Louis, MO, Oc obe 2009, pp. 5027–5032.
[8] D. A. Olejnik, B. P. Duis e ho , M. Ka ásek, K. Y. W. Schepe ,
T. Van Dijk, G. C. H. E. C oon, Unmanned Sys . 2020,08, 287.
[9] Y. Jin, Y. Ren, T. Song, Z. Jiang, G. Song, in In . Con . on Compu ing,
Ne wo ks and In e ne o Things, Xiamen, China May 2023,
pp. 278–285.
[10] S. Tijmons, G. C. H. E. C oon, B. Remes, C. Wag e , R. Ruijsink,
E. J. Kampen, Q. Chu, in Ad ances in Ae ospace Guidance,
Na iga ion and Con ol, Sp inge 2013, pp. 463–482.
[11] C. Wag e , S. Tijmons, B. D. W. Remes, G. C. H. E. C oon, in IEEE In .
Con . on Robo ics and Au oma ion Hong Kong, China May 2014,
pp. 4982–4987.
[12] S. Tijmons, G. C. H. E. C oon, B. D. W. Remes, C. De Wag e ,
M. Mulde , IEEE T ans. Rob. 2017,33, 858.
[13] S. Tijmons, C. De Wag e , B. Remes, G. C. H. E. De C oon, Ae ospace
2018,5, 69.
[14] K. Y. Schepe , M. Ka ásek, C. De Wag e , B. D. Remes,
G. C. De C oon, in IEEE In . Con . on Robo ics and Au oma ion,
B isbane, QLD, Aus alia May 2018, pp. 5546–5552.
[15] A. Gómez Eguíluz, J. P. Rod íguez-Gómez, R. Tapia, F. J. Maldonado,
J. A. Acos a, J. R. Ma ínez-de Dios, A. Olle o, in IEEE/RSJ In . Con . on
In elligen Robo s and Sys ems, P ague, Czech Republic Sep embe
2021, pp. 1958–1965.
[16] J. P. Rod íguez-Gómez, R. Tapia, M. M. Guzmán Ga cia,
J. R. Ma ínez-de Dios, A. Olle o, IEEE Rob. Au om. Le . 2022,7, 5413.
[17] J. Ribei o-Gomes, J. Gaspa , A. Be na dino, F on . Rob. AI 2023,10.
[18] G. Gallego, T. Delb ück, G. O cha d, C. Ba olozzi, B. Taba,
A. Censi, S. Leu enegge , A. J. Da ison, J. Con ad , K. Daniilidis,
D. Sca amuzza, IEEE T ans. Pa e n Anal. Mach. In ell. 2022,
44, 154.
[19] B. Chak a a hi, A. A. Ve ma, K. Daniilidis, C. Fe mulle , Y. Yang,
Recen E en Came a Inno a ions: A Su ey, h p://a xi .o g/abs/
2408.13627 (accessed: 2024).
[20] R. Tapia, J. P. Rod íguez-Gómez, J. A. Sanchez-Diaz, F. J. Ga˜nán,
I. G. Rod íguez, J. Luna-San ama ia, J. R. Ma ínez-De Dios,
A. Olle o, in IEEE/RSJ In . Con . on In elligen Robo s and Sys ems
2023, pp. 3025–3032.
[21] M. A. Mahowald, C. Mead, Sci. Am. 1991,264, 76.
[22] R. Se ano-Go a edona, M. Os e , P. Lich s eine , A. Lina es-
Ba anco, R. Paz-Vicen e, F. Gomez-Rod iguez, L. Camunas-Mesa,
R. Be ne , M. Ri as-Pe ez, T. Delb uck, S. C. Liu, R. Douglas,
P. Haflige , G. Jimenez-Mo eno, A. Ci i Ballcels, T. Se ano-
Go a edona, A. J. Acos a-Jimenez, B. Lina es-Ba anco, IEEE
T ans. Neu al Ne wo ks 2009,20, 1417.
[23] P. Lich s eine , T. Delb uck, in IEEE Wo kshop on Cha ge-Coupled
De ices and Ad anced Image Senso s, Ka uizawa, Nagano, Japan
June 2005.
[24] P. Lich s eine , T. Delb uck, in Resea ch in Mic oelec onics and
Elec onics, IEEE Vol. 22005, pp. 202–205.
[25] P. Lich s eine , C. Posch, T. Delb uck, in IEEE In . Solid S a e Ci cui s
Con ., San F ancisco, CA Feb ua y 2006, pp. 2060–2069.
[26] P. Lich s eine , C. Posch, T. Delb uck, IEEE Jou nal o Solid-S a e
Ci cui s 2008,43, 566.
[27] T. Fina eu, A. Niwa, D. Ma olin, K. Tsuchimo o, A. Masche oni,
E. Reynaud, P. Mos a alu, F. B ady, L. Cho a d, F. LeGo ,
H. Takahashi, H. Wakabayashi, Y. Oike, C. Posch, in IEEE In .
Solid-S a e Ci cui s Con . 2020, pp. 112–114.
[28] T. S o egen, H. Da aei, C. Robinson, A. Fix, in IEEE/CVF Win e Con .
on Applica ions o Compu e Vision 2022, pp. 3937–3945.
[29] C. Posch, D. Ma olin, R. Wohlgenann , in IEEE In . Solid-S a e Ci cui s
Con . 2010, pp. 400–401.
[30] C. Posch, D. Ma olin, R. Wohlgenann , IEEE Jou nal o Solid-S a e
Ci cui s 2011,46, 259.
[31] T. Se ano-Go a edona, B. Lina es-Ba anco, IEEE Jou nal o Solid-
S a e Ci cui s 2013,48, 827.
[32] C. B andli, R. Be ne , M. Yang, S. C. Liu, T. Delb uck, IEEE J. Solid-
S a e Ci cui s 2014,49, 2333.
[33] B. Son, Y. Suh, S. Kim, H. Jung, J. S. Kim, C. Shin, K. Pa k, K. Lee,
J. Pa k, J. Woo, Y. Roh, H. Lee, Y. Wang, I. O sianniko , H. Ryu, in
IEEE In . Solid-S a e Ci cui s Con . San F ancisco, CA Feb ua y 2017,
pp. 66–67.
[34] Y. Suh, S. Choi, M. I o, J. Kim, Y. Lee, J. Seo, H. Jung, D. H. Yeo,
S. Namgung, J. Bong, S. Yoo, S. H. Shin, D. Kwon, P. Kang,
S. Kim, H. Na, K. Hwang, C. Shin, J. S. Kim, P. K. J. Pa k, J. Kim,
H. Ryu, Y. Pa k, in IEEE In . Symp. on Ci cui s and Sys ems, Se ille,
Spain Oc obe 2020, pp. 1–5.
[35] O. Holešo ský, V. Hla áˇc, R. Ško ie a, R. Ví ek, in Compu e Vision
Win e Wo kshop, Rogaka Sla ina, Slo enia Feb ua y 2020.
[36] O. Holešo ský, R. Ško ie a, V. Hla áˇc, R. Ví ek, Senso s 2021,
21, 1137.
[37] J. Ba ios-A ilés, T. Iakymchuk, J. Samaniego, L. D. Medus,
A. Rosado-Mu˜noz, Elec onics 2018,7, 304.
[38] A. Censi, E. Muelle , E. F azzoli, S. Soa o, in IEEE In . Con . on
Robo ics and Au oma ion, Sea le, WA, May 2015, pp. 3319–3326.
[39] J. Cox, A. Ashok, N. Mo ley, Uncon . Imaging Adap . Op . 2020,
11508, 63.
www.ad ancedsciencenews.com www.ad in ellsys .com
Ad . In ell. Sys . 2025, 2401065 2401065 (17 o 20) © 2025 The Au ho (s). Ad anced In elligen Sys ems published by Wiley-VCH GmbH
26404567, 0, Downloaded om h ps://ad anced.onlinelib a y.wiley.com/doi/10.1002/aisy.202401065 by Readcube (Lab i a Inc.), Wiley Online Lib a y on [13/05/2025]. See he Te ms and Condi ions (h ps://onlinelib a y.wiley.com/ e ms-and-condi ions) on Wiley Online Lib a y o ules o use; OA a icles a e go e ned by he applicable C ea i e Commons License
[40] C. Fa abe , R. Paz, J. Pe ez-Ca asco, C. Zama e˜no, A. Lina es-
Ba anco, Y. LeCun, E. Culu ciello, T. Se ano-Go a edona,
B. Lina es-Ba anco, F on . Neu osci. 2012,6.
[41] H. Rebecq, R. Ran l, V. Kol un, D. Sca amuzza, in IEEE/CVF Con . on
Compu e Vision and Pa e n Recogni ion, Long Beach, CA June 2019,
pp. 3852.
[42] H. Rebecq, R. Ran l, V. Kol un, D. Sca amuzza, IEEE T ans. Pa e n
Anal. Mach. In ell. 2021,43, 1964.
[43] J. P. Rod íguez-Gómez, J. R. Ma ínez-de Dios, A. Olle o, G. Gallego,
IEEE Rob. Au om. Le . 2024,9, 8802.
[44] F. Y. Hsiao, H. K. Hsu, C. L. Chen, L. J. Yang, J. F. Shen, J. Appl. Sci.
Eng. 2012,15, 213.
[45] K. Y. Ma, P. Chi a a ananon, S. B. Fulle , R. J. Wood, Science 2013,
340, 603.
[46] F. J. Maldonado, J. Á. Acos a, J. To mo-Ba be o, P. G au,
M. M. Guzmán, A. Olle o, in IEEE/RSJ In . Con . on In elligen
Robo s and Sys ems, Las Vegas, Ne ada Oc obe 2020,
pp. 1385–1390.
[47] G. C. H. E. C oon, E. Wee d , C. Wag e , B. D. W. Remes, in IEEE In .
Con . on Robo ics and Biomime ics, Tianjin, China Decembe 2010,
pp. 1606–1611.
[48] G. C. H. E. C oon, E. de Wee d , C. De Wag e , B. D. W. Remes,
R. Ruijsink, IEEE T ans. Rob. 2012,28, 529.
[49] P. E. J. Duhamel, N. O. Pé ez-A ancibia, G. L. Ba ows, R. J. Wood, in
IEEE In . Con . on Robo ics and Au oma ion, Sain Paul, MN May 2012,
pp. 4228–4235.
[50] P. E. J. Duhamel, N. O. Pé ez-A ancibia, G. L. Ba ows, R. J. Wood,
IEEE/ASME T ans. Mecha on. 2013,18, 556.
[51] S. Ryu, U. Kwon, H. J. Kim, in IEEE/RSJ In . Con . on In elligen Robo s
and Sys ems, Daejeon, Ko ea (Sou h) Oc obe 2016, pp. 5645–5650.
[52] R. Tapia, J. R. Ma ínez-de Dios, A. Olle o, IEEE T ans. Pa e n Anal.
Mach. In ell. 2024,46, 9630.
[53] R. Benosman, C. Cle cq, X. Lago ce, S. H. Ieng, C. Ba olozzi, IEEE
T ans. Neu al Ne wo ks Lea n. Sys . 2014,25, 407.
[54] E. Pan, X. Liang, W. Xu, IEEE Sens. J. 2020,20, 8017.
[55] P. Zhang, H. Liu, Z. Ge, C. Wang, E. Y. Lam, IEEE T ans. Image P oc.
2024,33, 2318.
[56] S. Chen, M. Guo, in IEEE/CVF Con . on Compu e Vision and Pa e n
Recogni ion Wo kshops, Long Beach, CA June 2019, pp. 1682–1683.
[57] G. A. Folke sma, W. S aa man, N. Nijenhuis, C. H. Venne ,
S. S amigioli, IEEE Rob. Au om. Mag. 2017,24, 22.
[58] J. Ge des, A. Holness, A. Pe ez-Rosado, L. Robe s, A. G eisinge ,
E. Ba ne , J. Kempny, D. Lingam, C. H. Yeh, H. A. B uck,
S. K. Gup a, So Rob. 2014,1, 275.
[59] A. Pe ez-Rosado, H. A. B uck, S. K. Gup a, J. Mech. Rob. 2016,8.
[60] A. E. Holness, H. A. B uck, S. K. Gup a, In . J. Mic o Ai Veh. 2018,
10, 50.
[61] H. A. B uck, S. K. Gup a, Biomime ics 2023,8, 485.
[62] Z. Jiao, L. Wang, L. Zhao, W. Jiang, Ae osp. Sci. Technol. 2021,116,
106870.
[63] A. Ramezani, X. Shi, S. J. Chung, S. Hu chinson, in IEEE In . Con .
on Robo ics and Au oma ion, S ockholm, Sweden May 2016,
pp. 3219–3226.
[64] W. Yang, L. Wang, B. Song, In . J. Mic o Ai Veh. 2018,10, 70.
[65] M. Hassanalian, A. Abdelkefi, M. Wei, S. Ziaei-Rad, Ac a Mech. 2017,
228, 1097.
[66] H. Huang, W. He, J. Wang, L. Zhang, Q. Fu, IEEE/ASME T ans.
Mecha on. 2022,27, 5484.
[67] N. F anceschini, F. Ru fie , J. Se es, Cu . Biol. 2007,17, 329.
[68] E. W. Hawkes, D. Len ink, J. R. Soc. In e ace 2016,13, 20160730.
[69] F. Ru fie , Science 2018,361, 1073.
[70] M. Ka ásek, F. T. Muij es, C. De Wag e , B. D. W. Remes,
G. C. H. E. de C oon, Science 2018,361, 1089.
[71] M. Keennon, K. Klingebiel, H. Won, in AIAA Ae ospace Sciences
Mee ing Including he New Ho izons Fo um and Ae ospace
Exposi ion, Nash ille, TN Janua y 2012.
[72] H. V. Phan, S. Au ecianus, T. Kang, H. C. Pa k, In . J. Mic o Ai Veh.
2019,11.
[73] H. V. Phan, S. Au ecianus, T. K. L. Au, T. Kang, H. C. Pa k, IEEE Rob.
Au om. Le . 2020,5, 5059.
[74] R. J. Wood, IEEE T ans. Rob. 2008,24, 341.
[75] S. B. Fulle , A. Sands, A. Hagge y, M. Ka pelson, R. J. Wood, in IEEE
In . Con . on Robo ics and Au oma ion, Ka ls uhe, Ge many May 2013,
pp. 1374–1380.
[76] K. Y. Ma, S. M. Fel on, R. J. Wood, in IEEE/RSJ In . Con . on In elligen
Robo s and Sys ems, Vilamou a-Alga e, Po ugal Oc obe 2012,
pp. 1133–1140.
[77] E. F. Helbling, S. B. Fulle , R. J. Wood, in IEEE In . Con . on Robo ics
and Au oma ion, Hong Kong, China May 2014, pp. 5516–5522.
[78] R. Malka, A. L. Desbiens, Y. Chen, R. J. Wood, in IEEE/RSJ In . Con .
on In elligen Robo s and Sys ems, Chicago, IL Sep embe 2014,
pp. 2879–2885.
[79] X. Yang, Y. Chen, L. Chang, A. A. Calde ón, N. O. Pé ez-A ancibia,
IEEE Rob. Au om. Le . 2019,4, 4270.
[80] M. Guzmán, C. R. Páez, F. J. Maldonado, R. Zu e ey, J. To mo-
Ba be o, J. Acos a, A. Olle o, in IEEE/RSJ In . Con . on In elligen
Robo s and Sys ems 2021, pp. 6358–6365.
[81] L. Cal en e, J. Á. Acos a, A. Olle o, in Ae ial Robo ic Sys ems Physically
In e ac ing wi h he En i onmen , Biog ad na Mo u, C oa ia Oc obe
2021, pp. 1–6.
[82] V. Pe ez-Sanchez, A. E. Gomez-Tamm, E. Sa as ano, B. C. A ue,
A. Olle o, Appl. Sci. 2021,11, 2930.
[83] E. Sa as ano, V. Pe ez-Sanchez, B. A ue, A. Olle o, IEEE Rob. Au om.
Le . 2022,7, 8076.
[84] A. Gómez Eguíluz, J. P. Rod íguez-Gómez, J. L. Paneque, P. G au,
J. R. Ma ínez-de Dios, A. Olle o, in P oc. o he Wo kshop on
Resea ch, Educa ion and De elopmen o Unmanned Ae ial Sys ems,
C anfield, UK No embe 2019, pp. 335–343.
[85] R. Zu e ey, J. To mo-Ba be o, D. Feliu-Talegón, S. R. Nekoo,
J. Á. Acos a, A. Olle o, Na . Commun. 2022,13, 7713.
[86] S. R. Nekoo, D. Feliu-Talegon, R. Tapia, A. C. Sa ue, J. R. Ma ínezde
Dios, A. Olle o, Robo ica 2023,41, 3022.
[87] D. Gayango, R. Salmo al, H. Rome o, J. M. Ca mona, A. Sua ez,
A. Olle o, IEEE Rob. Au om. Le . 2023,8, 4243.
[88] R. Rashad, F. Cali ano, A. J. an de Scha , S. S amigioli, IMA J.
Ma h. Con ol In . 2020,37, 1400.
[89] F. Cali ano, R. Rashad, A. Dijkshoo n, L. G. Koe kamp, R. Sneep,
A. B ugnoli, S. S amigioli, Annu. Re . Con ol 2021,51, 37.
[90] D. Geh ig, D. Sca amuzza, A e High-Resolu ion E en Came as Really
Needed? h p://a xi .o g/abs/2203.14672 (accessed: 2022).
[91] R. Tapia, A. C. Sa ue, S. R. Nekoo, J. R. Ma ínez-de Dios, A. Olle o, in
IEEE In . Con . on Robo ics and Au oma ion Wo kshops, London,
Uni ed Kingdom May 2023.
[92] F. Clade a, A. Bisulco, D. Kepple, V. Isle , D. D. Lee, in IEEE In . Con .
on Image P ocessing, Abu Dhabi, Uni ed A ab Emi a es Oc obe 2020,
pp. 3084–3088.
[93] D. R. Kepple, D. Lee, C. P epsius, V. Isle , I. M. Pa k, D. D. Lee, in
Eu opean Con . on Compu e Vision (Eds: A. Vedaldi, H. Bischo ,
T. B ox, J. M. F ahm), Glasgow, UK Augus 2020, pp. 500–516.
[94] A. Bisulco, F. Clade a, V. Isle , D. D. Lee, in IEEE In . Con . on Robo ics
and Au oma ion,Xi’an, China May 2021, p. 14098.
[95] Z. Wang, F. Clade a, A. Bisulco, D. Lee, C. J. Taylo , K. Daniilidis,
M. A. Hsieh, D. D. Lee, V. Isle , IEEE Rob. Au om. Le . 2022,7,
8737.
[96] K. Kodama, Y. Sa o, Y. Yo ikado, R. Be ne , K. Mizoguchi, T. Miyazaki,
M. Tsukamo o, Y. Ma oba, H. Shinozaki, A. Niwa, T. Yamaguchi,
www.ad ancedsciencenews.com www.ad in ellsys .com
Ad . In ell. Sys . 2025, 2401065 2401065 (18 o 20) © 2025 The Au ho (s). Ad anced In elligen Sys ems published by Wiley-VCH GmbH
26404567, 0, Downloaded om h ps://ad anced.onlinelib a y.wiley.com/doi/10.1002/aisy.202401065 by Readcube (Lab i a Inc.), Wiley Online Lib a y on [13/05/2025]. See he Te ms and Condi ions (h ps://onlinelib a y.wiley.com/ e ms-and-condi ions) on Wiley Online Lib a y o ules o use; OA a icles a e go e ned by he applicable C ea i e Commons License
C. B andli, H. Wakabayashi, Y. Oike, in IEEE In . Solid-S a e Ci cui s
Con . San F ancisco, CA, Feb ua y 2023, pp. 92–94.
[97] A. Niwa, F. Mochizuki, R. Be ne , T. Ma uya ma, T. Te ano,
K. Takamiya, Y. Kimu a, K. Mizoguchi, T. Miyazaki, S. Kaizu,
H. Takahashi, A. Suzuki, C. B andli, H. Wakabayashi, Y. Oike, in
IEEE In . Solid-S a e Ci cui s Con ., San F ancisco, CA, Feb ua y
2023, pp. 4–6.
[98] R. Tapia, A. Gómez Eguíluz, J. R. Ma ínez-de Dios, A. Olle o, in IEEE
In . Con . on Robo ics and Au oma ion Wo kshops, Vi ual Con e ence,
May 2020, pp. 1–3.
[99] R. Tapia, J. R. Ma ínez-de Dios, A. Gómez Eguíluz, A. Olle o, Au on.
Rob. 2022,46, 879.
[100] E. Mueggle , H. Rebecq, G. Gallego, T. Delb uck, D. Sca amuzza,
In . J. Rob. Res. 2017,36, 142.
[101] A. Mi okhin, C. Ye, C. Fe mülle , Y. Aloimonos, T. Delb uck, in
IEEE/RSJ In . Con . on In elligen Robo s and Sys ems, Macau,
China No embe 2019, pp. 6105–6112.
[102] L. Bu ne , A. Mi okhin, C. Fe mülle , Y. Aloimonos, EVIMO2: An
e en Came a Da ase o Mo ion Segmen a ion, Op ical Flow,
S uc u e om Mo ion, and Visual Ine ial Odome y in Indoo
Scenes wi h Monocula o S e eo Algo i hms, h p://a xi .o g/
abs/2205.03467 (accessed: 2022).
[103] S. Klenk, J. Chui, N. Demmel, D. C eme s, in IEEE/RSJ In . Con . on
In elligen Robo s and Sys ems, P ague, Czech Republic Sep embe
2021, pp. 8601–8608.
[104] J. Hidalgo-Ca ió, G. Gallego, D. Sca amuzza, in IEEE/CVF Con . on
Compu e Vision and Pa e n Recogni ion, New O leans, LA June
2022, pp. 5771–5780.
[105] M. Geh ig, W. Aa en s, D. Geh ig, D. Sca amuzza, IEEE Rob. Au om.
Le . 2021,6, 4947.
[106] L. Gao, Y. Liang, J. Yang, S. Wu, C. Wang, J. Chen, L. Kneip, IEEE Rob.
Au om. Le . 2022,7, 8217.
[107] P. Chen, W. Guan, F. Huang, Y. Zhong, W. Wen, IEEE T ans. In ell.
Veh. 2024,9, 407.
[108] P. Chen, W. Guan, P. Lu, IEEE Rob. Au om. Le . 2023,8, 3661.
[109] F. Ba anco, C. Fe mulle , Y. Aloimonos, T. Delb uck, F on .
Neu osci. 2016,10.
[110] B. Rueckaue , T. Delb uck, F on . Neu osci. 2016,10.
[111] A. Z. Zhu, D. Thaku , T. Özaslan, B. P omme , V. Kuma ,
K. Daniilidis, IEEE Rob. Au om. Le . 2018,3, 2032.
[112] J. Delme ico, T. Cieslewski, H. Rebecq, M. Faessle , D. Sca amuzza,
inIEEE In . Con . on Robo ics and Au oma ion, Mon eal, QC, Canada
May 2019, pp. 6713–6719.
[113] K. Chaney, F. Clade a, Z. Wang, A. Bisulco, M. A. Hsieh, C. Ko pela,
V. Kuma , C. J. Taylo , K. Daniilidis, in IEEE/CVF Con . on Compu e
Vision and Pa e n Recogni ion Wo kshops, Vancou e , BC, Canada
June 2023, pp. 4016–4023.
[114] C. Schee linck, H. Rebecq, T. S o egen, N. Ba nes, R. Mahony,
D. Sca amuzza, in IEEE/CVF Con . on Compu e Vision and
Pa e n Recogni ion Wo kshops, Long Beach, CA June 2019,
pp. 1684–1693.
[115] T. Se ano-Go a edona, B. Lina es-Ba anco, F on . Neu osci.
2015,9.
[116] S. A sha , T. J. Hamil on, J. Tapson, A. an Schaik, G. Cohen, F on .
Neu osci. 2019,12.
[117] A. Ami , B. Taba, D. Be g, T. Melano, J. McKins y, C. Di Nol o,
T. Nayak, A. And eopoulos, G. Ga eau, M. Mendoza, J. Kusni z,
M. Debole, S. Esse , T. Delb uck, M. Flickne , D. Modha, IEEE/
CVF Con . on Compu e Vision and Pa e n Recogni ion, Honolulu,
HI July 2017, pp. 7388–7397.
[118] C. Bo e i, P. Bich, F. Pa eschi, L. P ono, R. Ro a i, G. Se i, in IEEE/
CVF Con . on Compu e Vision and Pa e n Recogni ion Wo kshops,
Vancou e , BC, Canada June 2023, pp. 4065–4070.
[119] H. Rebecq, D. Geh ig, D. Sca amuzza, in Con . on Robo Lea ning,
Zü ich, Swi ze land Oc obe 2018, pp. 969–982.
[120] Y. Hu, S. C. Liu, T. Delb uck, in IEEE/CVF Con . on Compu e Vision
and Pa e n Recogni ion Wo kshops, Nash ille, TN June 2021,
pp. 1312–1321.
[121] D. Joube , A. Ma ci eau, N. Ralph, A. Jolley, A. an Schaik,
G. Cohen, F on . Neu osci. 2021,15.
[122] A. Ziegle , D. Teigland, J. Tebbe, T. Gossa d, A. Zell, in IEEE In .
Con . on Robo ics and Au oma ion, London, UK May 2023,
pp. 11669–11675.
[123] V. Vasco, A. Glo e , C. Ba olozzi, in IEEE/RSJ In . Con . on In elligen
Robo s and Sys ems, Daejeon, Ko ea (Sou h), Oc obe 2016,
pp. 4144–4149.
[124] E. Mueggle , C. Ba olozzi, D. Sca amuzza, in B i ish Machine Vision
Con ., London, HK, Sep embe 2017, pp. 1–11.
[125] I. Alzuga ay, M. Chli, IEEE Rob. Au om. Le . 2018,3, 3177.
[126] C. Ha is, M. S ephens, in Al ey Vision Con . Manches e , UK
Sep embe 1988, pp. 23.1–23.6.
[127] C. B ¨
andli, J. S ubel, S. Kelle , D. Sca amuzza, T. Delb uck, in IEEE
In . Con . on E en -based Con ol, Communica ion, and Signal
P ocessing, K akow, Poland June 2016, pp. 1–7.
[128] D. Re e e Valei as, X. Clady, S. H. Ieng, R. Benosman, IEEE T ans.
Neu al Ne wo ks Lea n. Sys . 2019,30, 1218.
[129] D. Geh ig, H. Rebecq, G. Gallego, D. Sca amuzza, In . J. Compu .
Vision 2020,128, 601.
[130] J. Y. Bougue , In el Co po a ion, Technical Repo 5, 2001.
[131] J. P. Rod íguez-Gomez, A. Gómez Eguíluz, J. R. Ma ínez-de Dios,
A. Olle o, in IEEE In . Con . on Robo ics and Au oma ion, Pa is,
F ance. Augus 2020, pp. 8518–8524.
[132] A. Lukežiˇc, T. Vojíˇ , L. Čeho in Zajc, J. Ma as, M. K is an, In . J.
Compu . Vision 2018,126, 671.
[133] J. F. Hen iques, R. Casei o, P. Ma ins, J. Ba is a, in Eu opean Con .
on Compu e Vision (Eds: A. Fi zgibbon, S. Lazebnik, P. Pe ona,
Y. Sa o, C. Schmid), Fi enze, I aly Oc obe 2012, pp. 702–715.
[134] D. S. Bolme, J. R. Be e idge, B. A. D ape , Y. M. Lui, IEEE/CVF Con .
on Compu e Vision and Pa e n Recogni ion, San F ancisco, CA, USA
June 2010, pp. 2544–2550.
[135] H. Kim, S. Leu enegge , A. J. Da ison, in Eu opean Con . on
Compu e Vision (Eds: B. Leibe, J. Ma as, N. Sebe, M. Welling),
Ams e dam, The Ne he lands Oc obe 2016, pp. 349–364.
[136] H. Rebecq, T. Ho s schae e , G. Gallego, D. Sca amuzza, IEEE Rob.
Au om. Le . 2017,2, 593.
[137] A. Z. Zhu, N. A anaso , K. Daniilidis, in IEEE/CVF Con . on
Compu e Vision and Pa e n Recogni ion, Honolulu, HI July 2017,
pp. 5816–5824.
[138] H. Rebecq, T. Ho s schae e , D. Sca amuzza, in B i ish Machine
Vision Con . 2017, pp. 1–8.
[139] A. R. Vidal, H. Rebecq, T. Ho s schae e , D. Sca amuzza, IEEE Rob.
Au om. Le . 2018,3, 994.
[140] F. Mahlknech , D. Geh ig, J. Nash, F. M. Rockenbaue , B. Mo ell,
J. Delaune, D. Sca amuzza, IEEE Rob. Au om. Le . 2022,
7, 8651.
[141] X. Liu, H. Xue, X. Gao, H. Liu, B. Chen, S. S. Ge, IEEE T ans. Ins um.
Meas. 2023,72,1.
[142] A. Gup a, P. Sha ma, D. Ghosh, D. Ghose, S. K. Mu hukuma , in
IEEE In . Con . on Elec onics, Compu ing and Communica ion
Technologies, Bangalo e, India July 2023, pp. 1–6.
[143] W. Guan, P. Chen, Y. Xie, P. Lu, IEEE T ans. Au om. Sci. Eng. 2024,
21, 6277.
[144] S. Klenk, M. Mo ze , L. Koes le , D. C eme s, in In . Con . on 3D
Vision 2024, pp. 739–749.
[145] W. Guan, P. Chen, H. Zhao, Y. Wang, P. Lu, Ad . In ell. Sys . 2024,6,
2400243.
www.ad ancedsciencenews.com www.ad in ellsys .com
Ad . In ell. Sys . 2025, 2401065 2401065 (19 o 20) © 2025 The Au ho (s). Ad anced In elligen Sys ems published by Wiley-VCH GmbH
26404567, 0, Downloaded om h ps://ad anced.onlinelib a y.wiley.com/doi/10.1002/aisy.202401065 by Readcube (Lab i a Inc.), Wiley Online Lib a y on [13/05/2025]. See he Te ms and Condi ions (h ps://onlinelib a y.wiley.com/ e ms-and-condi ions) on Wiley Online Lib a y o ules o use; OA a icles a e go e ned by he applicable C ea i e Commons License
[146] M. S. Lee, J. H. Jung, Y. J. Kim, C. G. Pa k, IEEE Rob. Au om. Le .
2024,9, 1003.
[147] J. Lu, H. Feng, W. Liu, B. Hu, in IEEE Ad anced In o ma ion
Technology, Elec onic and Au oma ion Con ol Con . 2024, Vol. 7,
pp. 1510–1515.
[148] R. Pelle i o, M. Cannici, D. Geh ig, J. Belhadj, O. Dubois-Ma a,
M. Casasco, D. Sca amuzza, Deep isual odome y wi h e en s
and ames, in IEEE/RSJ In e na ional Con e ence on In elligen
Robo s and Sys ems, Abu Dhabi, Uni ed A ab Emi a es Oc obe
2024, pp. 8966–8973.
[149] T. Qin, P. Li, S. Shen, IEEE T ans. Rob. 2018,34, 1004.
[150] A. Tomy, A. Paigwa , K. S. Mann, A. Renzaglia, C. Laugie , in IEEE
In . Con . on Robo ics and Au oma ion, Philadelphia, PA, USA May
2022, pp. 933–939.
[151] D. Geh ig, D. Sca amuzza, Na u e 2024,629, 1034.
[152] F. Kaup, S. Hacke , E. Men zendo , C. Meu isch,
D. Haushee , Ne Sys Lab, Technical Repo Ne Sys-TR-2018-01
2018.
[153] U. Iqbal, T. Da ies, P. Pe ez, Senso s 2024,24, 4830.
[154] P. Fu gale, J. Rehde , R. Siegwa , in IEEE/RSJ In . Con . on In elligen
Robo s and Sys ems 2013, pp. 1280–1286.
[155] J. Rehde , J. Nikolic, T. Schneide , T. Hinzmann, R. Siegwa , in IEEE
In . Con . on Robo ics and Au oma ion, S ockholm, Sweden May
2016, pp. 4304–4311.
[156] M. Guo, J. Huang, S. Chen, in IEEE In . Symp. on Ci cui s and
Sys ems, Bal imo e, MD May 2017,p.1.
[157] S. B. Fulle , IEEE Rob. Au om. Le . 2019,4, 570.
[158] X. Wu, W. He, Q. Wang, T. Meng, X. He, Q. Fu, IEEE T ans. Ind.
Elec on. 2023,70, 8215.
[159] P. V. C. Hough, US3 069 654A 1962.
www.ad ancedsciencenews.com www.ad in ellsys .com
Ad . In ell. Sys . 2025, 2401065 2401065 (20 o 20) © 2025 The Au ho (s). Ad anced In elligen Sys ems published by Wiley-VCH GmbH
26404567, 0, Downloaded om h ps://ad anced.onlinelib a y.wiley.com/doi/10.1002/aisy.202401065 by Readcube (Lab i a Inc.), Wiley Online Lib a y on [13/05/2025]. See he Te ms and Condi ions (h ps://onlinelib a y.wiley.com/ e ms-and-condi ions) on Wiley Online Lib a y o ules o use; OA a icles a e go e ned by he applicable C ea i e Commons License
