Player is loading...
LightAI4Comfort
Wednesday, Aug. 24, 2022 · 12:13 p.m. · 11m 25s
This project aims at developing an AI-driven lighting control system adaptable to different office spaces. This system is conceived to manage the solar admission through electric blinds and the electric lighting in order to maximize the use of daylight, while avoiding direct sunlight as a potential source of glare. Unlike other systems, our setup does not include any sensor nor camera inside the space, thus enforcing privacy preservation for the user(s) and reducing infrastructure costs.
Embed
Transcriptions
Note: this content has been automatically generated.
00:00:00
hello everybody thank you for being here and body
00:00:04
excited with um my teammates or material john to present
00:00:08
the the project green been working on the no the last name nine days late a. i. for comfort
00:00:15
so him modern times we spend a ninety percent of our time indoors
00:00:21
this is a lot of time especially because uh as humans
00:00:24
we appeal to lead a a experiences they life especially daylight
00:00:29
so we spent time uh i tones by special enough this is norma at least a norma in normal times
00:00:35
and um the problems that a delight is goodbye will sell like a lot
00:00:40
of people complaining about that and uh yeah some service as shown that that
00:00:45
uh forty people uh uh forty percent of people i struggle of uh uh uh to working poor lighting conditions
00:00:51
uh every day and uh maybe you are you have experience these conditions this is said quite common
00:00:57
uh to not be able to read a document because it's too it's too dark
00:01:02
complaining because they you are too hot you only have to pay we know
00:01:05
there's some light comes into a into the room and get close to your body
00:01:10
and then maybe you out to try to adjust adapt your
00:01:14
all your blind system your you'll you'll like into indoor lighting
00:01:18
and uh and then maybe you cannot see your uh i
00:01:20
just uh your computer screen is everybody commons a a situation
00:01:25
yet uh they are not fully uh sold a
00:01:29
yet and what we propose it's uh to use uh
00:01:32
uh yeah i do have a automation about the um the lighting control so to
00:01:37
have optimal condition to be able to perform a commonality starts at the your best
00:01:43
how we do that so we have a this predicting
00:01:45
model then uh adopts the bank traditions and the light intensity
00:01:50
so to achieve their right uh those of light you need to perform this uh off it's a task
00:01:56
and to do that to be use a a data that
00:01:58
come from um uh outside so the um the whether whether they
00:02:02
thought that they can be from our weather station like we have with the idea of from a an external uh a. p. i.
00:02:09
and then re input some basic uh room features that describe uh the space we work
00:02:14
on so we are able to predict uh the light inside that that specific uh rubber
00:02:20
yeah we do believe there's a drawing potential a market that for this
00:02:24
especially like after copy times uh a lot of companies that struggle
00:02:29
having like a people back to work though so this a phenomenon of
00:02:34
quite greeting like people leaving their their job because they are not satisfied
00:02:38
so i think like management is a lot thinking about
00:02:42
how we uh have that that's working condition in our offices
00:02:46
and uh the good thing is that a lot of um office buildings like n. sync that traditional ones
00:02:53
then can benefit for the solutions by uh either because they
00:02:56
are i mean newly built though because they are being about graffiti
00:03:00
and so we just need the elected by lines and the mobile looming areas
00:03:04
to be able to um to uh yeah profit from uh from our system
00:03:09
so just a bit behind the this scene so what what we of what we do what we propose with our technology
00:03:15
so um and we we train ah i'm actually bought
00:03:20
model based on on physical simulation light in the indoor spaces
00:03:25
and uh so this is our predicting more that that is um uh based on as a as a as i said on
00:03:31
a external data like a that is the non intrusive so
00:03:35
we don't have a sensor inside the room that's the biggest difference
00:03:38
with uh some uh competitors so we don't have cameras that we
00:03:41
don't have add anything that affects privacy of the of the workers so
00:03:46
uh we uh we use the done now a lightweight solution
00:03:49
that can be deployed uh like always in every office space
00:03:53
and we we do use some um uh yeah solution
00:03:57
like additional programming that uh allows us to have a
00:04:00
like customisable graphical user interface that adapts to the to the customer needs
00:04:06
uh so our our team there is it's quite interdisciplinary yes you
00:04:10
can see from a from a dislike this was very important to solo
00:04:14
uh the the problems that we had uh in uh for the specific task that uh
00:04:20
a related to buildings to a building controls but also to the physics of low lighting and of course
00:04:25
to mature earning and um and uh um yeah and
00:04:30
develop in the programming for developing uh our our solution
00:04:34
uh just a bit this story of our project because there was an early prototype we based our project on
00:04:39
um and so this prototype that was installed uh he uh a needy up and also in the infrared work
00:04:45
as shown that users with a satisfied with the with the with the system and we also sing
00:04:50
some energy savings using the system but the main drawback was that uh the moral high cost for commission
00:04:57
so that was the main idea behind participate in this uh this challenge how we can sort of this of this problem
00:05:03
so uh the idea was to use the genetics model i generalised what model that
00:05:07
can be adapted to different conditions in different uh office spaces in the easiest way possible
00:05:12
uh of course we want to do one thing what good was in the in the previous project
00:05:17
and also a lot though especially wanted to be able to um to reduce the the cost for commissioning
00:05:24
by you having a model that is a simpler and can be adapted to different uh uh conditions
00:05:30
and so what we did uh during this i. c. c. was indeed uh
00:05:34
uh using a parametric a three d. model to to train
00:05:38
uh uh the the maturity model so we can have different conditions
00:05:42
in our in our model and the real so we're able to input a data from uh
00:05:47
a swiss um uh open a a. p. i. so that we can
00:05:51
a joystick idea apply this this model in different um their location in switzerland
00:05:56
and also we worked a lot uh on the new user interface that was adapted to the to this new model
00:06:03
and also uh we had a future for the uh
00:06:06
for the users that they are able to have a personalise
00:06:09
you mean assuage so it's not everything automatic yeah they are able to like
00:06:14
to have a person alive uh lighting a experience in at their at their desk
00:06:20
uh we we uh ambition for this price additional
00:06:23
level of our commissioning like from lightweight uh wants
00:06:27
to the uh uh lower they have lower cost and that can be deployed in like a generic standard
00:06:32
uh office buildings that like we have many of them and then we can work on more customised
00:06:37
solutions having a refined model that can adapt or
00:06:41
for a building to a special architectural features for example
00:06:45
uh so what about the the next steps we plan to use uh as we did also before
00:06:49
like the the ups uh as the one of the the most waiters possibly extending toward their uh offices
00:06:55
and use it to contact with potential partners manufacturers and uh installers
00:07:00
and then i able in this way to reach out to potential clients that we believe are large building owners
00:07:06
uh from we reach a lot common uh a lot companies uh went from
00:07:11
their office spaces or if they did uh actually on the buildings directed is uh
00:07:16
this company is then they could benefit there um uh their clients
00:07:20
idiot um their um their employees with a a good lighting conditions
00:07:24
uh think here we have the demo for the for the videos so we'll switch thank you
00:07:31
it uh just one first
00:07:35
and it's you
00:07:42
which
00:07:45
oh
00:07:49
oh i see it
00:07:57
uh_huh
00:08:02
uh_huh
00:08:09
e. lou was uh_huh
00:08:15
e. o. t. mm mm mm e.
00:08:22
o. g. e. e. huh ooh oh
00:08:32
uh_huh e. e. o.
00:08:38
uh_huh who e. leno rue e.
00:08:45
uh_huh uh_huh oh
00:08:48
uh_huh oh
00:08:53
e. o. e. o. uh
00:09:01
yeah we ah uh_huh
00:09:05
mm e. um a
00:09:17
so we have a yeah a philosopher to locks which is uh consider very good the
00:09:23
um it's about four hundred locks which is the normal standard for a like a high quality um
00:09:30
a precise uh work uh there for international standards
00:09:37
and the yeah we have a next video stream or a a closely the user interface
00:09:44
e. e. e. visual
00:09:50
s. e. uh_huh uh_huh e. uh_huh uh_huh yeah
00:09:57
uh_huh uh_huh uh_huh uh_huh uh_huh uh_huh uh_huh
00:10:04
mm hum you know is in
00:10:09
uh_huh uh_huh uh_huh uh_huh uh_huh you're mm
00:10:17
uh_huh uh_huh uh_huh lame
00:10:23
uh_huh uh_huh uh_huh were uh_huh
00:10:28
yes uh_huh uh_huh uh_huh
00:10:31
uh_huh uh_huh uh_huh uh_huh uh_huh
00:10:37
uh_huh uh_huh uh_huh yeah uh_huh uh_huh uh_huh
00:10:44
sure justin last slide for a a
00:10:54
yeah that too and the slides and for the big and uh huh okay i was load okay okay yeah but
00:11:04
uh thank you for for your attention and um yeah
00:11:07
a um your creep will be able to to present the
00:11:11
uh how i'm aware work and uh for which we have a lot a lot
Conference Program
LightAI4Comfort
Aug. 24, 2022 · 12:13 p.m.
119 views
LaternaMagica
Aug. 24, 2022 · 1:08 p.m.
118 views
SolveHR: Natural language inference for professional search
Aug. 24, 2022 · 1:26 p.m.
Recommended talks
VERISWISS Pitch
Javier Gutiérrez Fernández
Sept. 17, 2018 · 2:42 p.m.
157 views
Discoverying Life patterns
Fausto Giunchiglia, from University of Trento, Italy
June 8, 2015 · 4:07 p.m.
278 views