Előadás

AI in Smart City Transportation

Vida Rolland

BME TMIT

Beyond Artificial Intelligence Research: a New Innovation and Industrial Domain

2025. november 7. 13:00Babeș-Bolyai University, Central Building, Aula Magna · Kolozsvár · Románia

Átirat · 48:43

Thank you. Um so I would like to to welcome everyone. Um uh just a few words about

Átirat · 48:52

myself. So I will talk in uh now about uh AI in uh smart city transportation.

Átirat · 48:58

Uh but before starting the talk just uh one short word about the fact that I uh

Átirat · 49:05

finished my uh studies here in this building 30 years ago. So it's uh

Átirat · 49:12

actually very good to be back uh and I uh have some very nice uh memories from

Átirat · 49:17

uh from those years. Uh so since then uh uh as it was mentioned I'm involved in

Átirat · 49:26

smart city related activities within it e at the university and what is behind

Átirat · 49:31

the smart cities actually the sensing infrastructure sensors how do we know what happens in the city how these

Átirat · 49:38

sensors are operating uh how they communicate uh uh so this is one of the

Átirat · 49:46

parts of my of my research so you see some of my involvement ment uh uh yeah within it e uh so then let's get started

Átirat · 49:57

uh with a very brief introduction about smart cities I guess that everyone uh

Átirat · 50:02

heard about these so this is a very common term although there's no um no uh

Átirat · 50:09

universally accepted definition of what a smart city is but we can say basically

Átirat · 50:15

that um is a city that builds on an integrated digital sensing and

Átirat · 50:23

communication infrastructure uh that will handle together all the different urban services uh that we

Átirat · 50:31

have. Uh it should be of course sustainable. It should be energy efficient. It should involve the people

Átirat · 50:38

actively. So a smart city should not be a showcase of technology just for the

Átirat · 50:44

sake of technology but it should be really developed for the people who live in that city and through the active

Átirat · 50:50

involvement of those people. Um and so obviously the final goal should be to

Átirat · 50:57

improve the quality of life and uh to have a livable city. And so I uh

Átirat · 51:02

highlighted there in in blue one very important aspect that I did not mention until now. A smart city should be

Átirat · 51:10

adaptive should adapt to what happens now in the city. What is the current

Átirat · 51:16

situation and the finer the time scale the city can adapt to the current

Átirat · 51:22

situation, the smarter the city will be. And so we will see some examples of such

Átirat · 51:28

adaptivity. Um so you will probably hear uh in more

Átirat · 51:35

detail about uh AI in uh car manufacturing in in smart vehicles uh

Átirat · 51:41

self-driving cars. So I will not enter that much into the details here. But

Átirat · 51:46

when you uh speak about smart transportation AI in smart transportation within the cities

Átirat · 51:52

obviously the first thing to think about is uh self-driving cars autonomous cars.

Átirat · 51:58

Uh so but basically these are cars where uh they rely on AI based perception and

Átirat · 52:05

decision making uh to navigate in a quite complex urban environment uh

Átirat · 52:11

without human intervention or limited uh intervention from from humans. So AI is

Átirat · 52:18

there for the perception. So what we mean by the perception we have all kinds of sensors installed in the car. So we

Átirat · 52:26

have cameras, we have lighters, we have radars, ultrasonic uh sensors uh and

Átirat · 52:32

they are there to provide object detection to to understand the scene

Átirat · 52:38

where the car is located. So the the environment around them and u to uh help

Átirat · 52:46

basically navigate in this uh in this um um environment.

Átirat · 52:52

So that's the perception. Then you need AI also for the prediction. What do you

Átirat · 52:58

predict? What will happen next around you? What will the other cars do around you? What will the passenger the the the

Átirat · 53:06

pedestrians do? Will a pedestrian that you see that you detected will uh it cross the street or not? Uh and so on.

Átirat · 53:15

And then obviously you will need also AI for planning and decision making. Okay.

Átirat · 53:21

to take all kinds of decisions. Which is the best route to take based on the current situation in the city? Uh when

Átirat · 53:29

should I change the lane? When should I slow down or speed up to avoid an accident or to be more efficient to

Átirat · 53:36

cross through an intersection more rapidly and so on. So this was so

Átirat · 53:41

self-driving cars. This is a uh long dream of humanity. uh and we thought

Átirat · 53:47

about it as a science fiction for many years and today we can say that it is really a reality even if we do not see

Átirat · 53:54

these cars here in Kluj maybe or in Budapest but they are there in several other cities if you if you go to San

Átirat · 54:02

Francisco there are hundreds of Whimos u u taxis that are self-driving uh robot

Átirat · 54:10

taxis that you can uh just um actually jump in and and uh u

Átirat · 54:17

Yeah, use them. So, uh it is a reality. We are also working uh in this field. Uh

Átirat · 54:25

at BME, there were two uh recent spin-off uh that were formed uh that are

Átirat · 54:31

involved in this driveby cloud is working on creating a digital twin um

Átirat · 54:38

for uh the environment. So uh modeling basically the environment all kinds of

Átirat · 54:45

different scenarios where the car could be uh um involved uh and so without

Átirat · 54:52

really testing the car in these physical environments. Okay, you can uh actually

Átirat · 54:58

model this and uh IMOS drive is the other uh company uh spin-off that is um

Átirat · 55:05

working also on software related to uh self-driving cars.

Átirat · 55:11

Then uh you could uh use or you should use AI in uh smart intersections. Uh so

Átirat · 55:19

traditionally an intersection is regulated by a traffic light. Okay, that

Átirat · 55:24

uh has a timing. How much time should you have a green light and how much time you should have a red light in a given

Átirat · 55:30

uh lane in a given direction. That is based on static data from the past.

Átirat · 55:35

Okay. Usually which lane has higher traffic. how so in the past uh what

Átirat · 55:42

happened uh but I would be smarter if I can be adaptive as I said so uh I have

Átirat · 55:50

all the sensors I have cameras I have embedded sensors in the road that can detect how many cars are there at the

Átirat · 55:56

intersection now how many cars are waiting how many people are in those cars that are waiting I mean cars are

Átirat · 56:03

intelligent they have sensors even inside so they know how many people are sitting in the car the car can

Átirat · 56:09

communicate with a traffic light can say that hey I'm waiting here I'm a car with five passengers give me the green light

Átirat · 56:16

okay that's easy to say obviously that's a difficult uh uh um problem to solve

Átirat · 56:22

but then you have um current data and then you have uh to react to solve the

Átirat · 56:30

current problem but that's still not that much of AI okay AI when it will

Átirat · 56:36

intervene could be an integrated and predictive solution. Okay, you do not

Átirat · 56:42

just analyze the given intersection but you also analyze what happens around

Átirat · 56:48

what happens in the other intersection uh yeah in the neighborhood uh what is

Átirat · 56:53

the traffic light in the city um how the traffic is predicted to evolve and

Átirat · 57:00

basically you will want to prevent congestion not to solve the problem once

Átirat · 57:06

it happened. you want to prevent to occure. Okay. So, uh that would be the

Átirat · 57:12

goal and we have also at BME there are uh research groups working on different

Átirat · 57:18

uh projects in this uh field. So, um there's one project which is called CTI

Átirat · 57:25

that is a Hungarian financed uh project by the Hungarian government uh which

Átirat · 57:31

involves BME and two smaller companies. There is a pilot uh deployment in pitch.

Átirat · 57:37

Uh you can see there the picture. So it it shows the the pilot deployment. So

Átirat · 57:44

with about uh eight or nine traffic lights that are uh controlled uh through

Átirat · 57:50

uh through this uh uh smart AI based software and uh there's a group that is

Átirat · 57:56

involved also in a in an international project which is an EU project together

Átirat · 58:01

with Japan. uh so there's BME there's a company from Turkey and there's a

Átirat · 58:06

research institute from Japan who are working on this uh project uh with a pilot uh that is deployed in Istanbul

Átirat · 58:16

uh okay so what's next uh what about

Átirat · 58:21

AIdriven congestion zone pricing so you know that uh many cities are having

Átirat · 58:27

these problems that they have uh peak hour congestion questions in the city

Átirat · 58:32

center and they try to uh to to regulate this by introducing some pricing scheme.

Átirat · 58:39

So you have to pay if you want to go by car with your personal car to the city center you have to pay for it. But

Átirat · 58:45

that's again that's not smart if you have a fixed amount that everyone should

Átirat · 58:51

pay. Smarter is an adaptive system. So what they introduced in Singapore for example, they were the first one to to

Átirat · 58:58

start this electric um um road pricing um um adaptive solution that uh the

Átirat · 59:07

price depends on uh the time of the day on the location that you are on the

Átirat · 59:13

congestion level of the of the city. So um obviously uh the price will go up in

Átirat · 59:21

the peak hour. I mean if you can shift your travel if it's not mandatory for you to travel at the peak hour if you

Átirat · 59:28

can shift it with uh with one hour let's say you will have to pay less but also

Átirat · 59:34

it's depends on the location. So there are several roads going through the city

Átirat · 59:40

towards the city center and some of them are more crowded than others. So uh if they have adaptive prices and the the

Átirat · 59:48

road that is more crowded will be more expensive then uh this might shape a bit

Átirat · 59:54

the traffic. So people might try to uh to uh select other uh ways to get to the

Átirat · 1:00:01

city center. Uh and so there's again um

Átirat · 1:00:06

um project that just started this year uh cooperation with a um with a Canadian

Átirat · 1:00:13

uh AI uh company uh that we are working together with them on uh this issue

Átirat · 1:00:20

dynamic congestion pricing. Uh then what about AIdriven search

Átirat · 1:00:26

pricing? uh search pricing is something similar. Basically uh we can see it

Átirat · 1:00:33

applied in several areas. It means that the price for any given service is not

Átirat · 1:00:38

fixed but it depends on offer and demand. Uh and that can be used in

Átirat · 1:00:43

different um applications. So for example, Yuber uses this uh uh um this I

Átirat · 1:00:50

guess that many of you uh have used already Yuber in the past. Um

Átirat · 1:00:57

I don't Yeah. So um so what's the idea

Átirat · 1:01:02

here is that uh when you ask for a ride in Yuber you will get an estimation of

Átirat · 1:01:09

the price and that will differ depending on the demand and offer. So obviously on

Átirat · 1:01:16

a Saturday night it will be more expensive than on the Sunday morning

Átirat · 1:01:21

because there are more people wanting to to catch a yuber to go home or to go to

Átirat · 1:01:26

a restaurant or whatever. But that dynamic pricing is not just towards the

Átirat · 1:01:32

the users but it's also towards the drivers. So when you uh ask for a ride,

Átirat · 1:01:38

the application will uh propose that ride to several different Yuber drivers.

Átirat · 1:01:46

Okay. And uh interestingly uh they will be asked to to take that

Átirat · 1:01:53

ride but not for the same price. Some will be asked will you take this guy to

Átirat · 1:01:59

this destination for $5 and the other will be uh

Átirat · 1:02:05

proposed $7 let's say. So it's not the same price that the application proposes

Átirat · 1:02:10

to the uh drivers um because

Átirat · 1:02:16

it's pro so no one knows how the application works how the AI engine

Átirat · 1:02:21

behind it works. It it's probably based on the drivers

Átirat · 1:02:28

uh previous behavior. So if you accept cheap drives then you might get proposed

Átirat · 1:02:34

even cheaper drives by the application. If you refuse cheaper drives then you

Átirat · 1:02:40

might uh get more uh expensive drives which are better for you. So you will be

Átirat · 1:02:45

paid more but it's not for sure. Nobody knows how this works. So the yuber

Átirat · 1:02:51

drivers try to uh reverse engineer actually what happens. What happens if I

Átirat · 1:02:57

refuse this and if I continue to refuse offers for an hour? What the application

Átirat · 1:03:03

will propose? And there all kinds of uh videos on the internet that you can uh

Átirat · 1:03:08

see about this that the drivers trying to reverse engineer what happens behind.

Átirat · 1:03:14

But what can be seen is that it it's it's probably uh here Yuber uses AI to maximize its

Átirat · 1:03:24

own trophy uh its own profit and not specifically to for the good of the

Átirat · 1:03:29

city. Okay. Uh hopefully nobody will uh

Átirat · 1:03:34

tell this to you but that I'm speaking badly about them here and I will uh get some issues with them later on. But

Átirat · 1:03:42

let's see uh uh search pricing in a different setup for parking. I don't

Átirat · 1:03:50

know if you heard about uh apps like just park. Uh this is a UK based uh

Átirat · 1:03:56

solution where the idea is that people can can rent their own private parking

Átirat · 1:04:02

space. So for example, you can rent your own driveway. So the the the the path uh

Átirat · 1:04:12

that is just before your garage. Okay. That uh normally you would not like

Átirat · 1:04:18

people other people to park their car in front of your house in front of your garage because then you will not be able

Átirat · 1:04:25

to come out from your house. But uh if you can sell this parking place um

Átirat · 1:04:34

when you will not use your car when you are not at home whatever then you can do

Átirat · 1:04:39

it through this application and the question is how much should I ask for my

Átirat · 1:04:45

uh private parking spot and that's where AI can can help um so there's an AI

Átirat · 1:04:53

based search u um pricing solution that

Átirat · 1:04:58

will propose you a price based on the expected demand and based on past

Átirat · 1:05:05

behaviors as well. So uh the the owner will will get a hint how much could he

Átirat · 1:05:13

ask for this uh for that uh place then he can decide what will be the final uh

Átirat · 1:05:19

price that he will ask for it. Okay. So he might comply with this suggestion or not. Um so when talking about smart

Átirat · 1:05:27

parking I also put here uh some uh words about uh traditional traditional let's

Átirat · 1:05:34

say uh smart parking solution uh that has nothing to do with AI but it has to

Átirat · 1:05:41

do with smart cities that we can find this in many many uh cities already around the the world. So uh the the

Átirat · 1:05:49

smart parking is that uh it is equipped with some kind of uh sensors that detect

Átirat · 1:05:55

whether that that spot is free or not and that can be then fed to an application that will uh show you where

Átirat · 1:06:01

are the free uh parking spots around the city. So we were in such a project uh uh

Átirat · 1:06:08

together with Ericson and Vodafon to develop a pilot and then this was taken over by a startup uh that is smart

Átirat · 1:06:15

links. So that uh actually operates today uh uh in the city of Budapest with

Átirat · 1:06:21

uh 6,000 sensors that are deployed in in two different districts.

Átirat · 1:06:26

Okay. And then this is my last slide. Uh there

Átirat · 1:06:32

are lots of other possible ways where you could uh use uh AI related to smart

Átirat · 1:06:40

transportation making smart making uh transportation within the city more efficient. So one thing is uh

Átirat · 1:06:48

matchmaking in in uh um right sharing applications. So right sharing means

Átirat · 1:06:55

that you would not like people to travel alone in their cars. We have the big congestions, the big uh traffic jams in

Átirat · 1:07:02

the city because 90% of the cars are just with their driver. Okay? Nobody

Átirat · 1:07:08

else sits in the car. And ride sharing apps uh try to uh uh get people uh more

Átirat · 1:07:16

people in the same car. Okay. um then you can obviously decrease the

Átirat · 1:07:23

number of of uh of uh cars and you will decrease the pollution level and uh and

Átirat · 1:07:28

the congestion level as well. The question is uh how to do the matchmaking. Okay. So you have people

Átirat · 1:07:35

who are uh with their own cars, you have other people who would like to get in those cars. All these people have

Átirat · 1:07:42

different destinations um yeah within the city. And then you should try to do the matchmaking in the

Átirat · 1:07:49

um in an efficient way to see which car takes which passenger, how long should

Átirat · 1:07:54

be the detour for this and uh and so on so forth. Um and I also mentioned here

Átirat · 1:08:00

that I mean uh when you think about right sharing there are several apps

Átirat · 1:08:05

that are doing right sharing. Uh ways for example I don't know if you heard

Átirat · 1:08:11

about ways that's also um um transportation application that also has

Átirat · 1:08:16

a uh a ways car poolool uh part that that is on uh um providing right

Átirat · 1:08:23

sharing. Yuber is usually known as a right sharing app which is not true.

Átirat · 1:08:29

Yuber is not a right sharing app. It's not right hailing app. It's basically a cheap taxi. But anyhow I would not like

Átirat · 1:08:36

to insist too much on Uber. Okay. Public transportation obviously that is also a

Átirat · 1:08:42

very important uh part as well. Uh how to optimize how to make the scheduling

Átirat · 1:08:48

adaptive. Okay. The the scheduling, how often should a given bus, metro, whatever come uh on a given line or the

Átirat · 1:08:57

trajectory, what should it be? What is the optimal trajectory to to serve the best the people and uh that

Átirat · 1:09:06

the smarter would be if that would be adaptive that would uh be um um u made

Átirat · 1:09:13

taking into account the current um uh needs for of of people. How many

Átirat · 1:09:20

people are waiting now in the in the different bus stations or or uh what are

Átirat · 1:09:26

their current mobility needs? uh that can change basically. You cannot rely just on static patterns uh from the past

Átirat · 1:09:34

because many other uh many events can happen in the city be that um rallies uh

Átirat · 1:09:40

sporting events uh visits of uh of uh um

Átirat · 1:09:46

presidents or whatever that will uh generate uh big uh changes in the um

Átirat · 1:09:52

traffic in the city. So uh there you can have AI based predictions that can be

Átirat · 1:09:58

used uh to uh have such adaptivity uh taken into account. And then the last

Átirat · 1:10:04

thing that I would like to mention is um vehicleto grid AI based optimized

Átirat · 1:10:10

vehicle-to- grid operation. Uh what vehicle to grid means is that we have electric cars that have batteries that

Átirat · 1:10:18

could be used for transporting not just people but transporting electricity.

Átirat · 1:10:23

What's the meaning? You have your electric car at home. Uh you can charge it in the evening during the cheap u

Átirat · 1:10:33

price period or you can uh charge it with uh free u uh free electricity

Átirat · 1:10:41

through your solar panels and then you take that electricity to the city

Átirat · 1:10:46

center. You go to your work the next day. uh you consume part of your uh um

Átirat · 1:10:54

battery charge for the travel but you have extra usually that you can uh

Átirat · 1:10:59

discharge in the city uh where there's peak demand uh uh so uh how this should

Átirat · 1:11:07

be optimized uh because you need to predict in advance what will be the

Átirat · 1:11:13

electricity need the next day in the city uh how many people will travel with

Átirat · 1:11:19

electric cars, how how much batteries spare battery uh will they have to be

Átirat · 1:11:25

discharged in the system and optimizing this it's a quite difficult issue. Okay,

Átirat · 1:11:31

so that's it. Thanks a lot for your attention.

Átirat · 1:11:42

Thank you very much Roland. And for this session before the break we have uh uh

Átirat · 1:11:48

presentation from Stefan Matei uh here from Kluj uh Robert Bosch Engineering

Átirat · 1:11:53

Center please and let's see where is your

Átirat · 1:12:00

presentation.

Átirat · 1:12:14

So Stefan, floor is yours. Thank you very much. Let's try the slide

Témák

MI-támogatott

Intelligens közlekedés

Összefoglaló

MI-támogatott

Vida Rolland az okosvárost adaptív, integrált érzékelési és kommunikációs infrastruktúrára épülő rendszerként értelmezi, majd azt vizsgálja, hol lép be az MI a városi közlekedésbe. Példái az önvezető járművek percepciójától és döntéshozatalától az adaptív jelzőlámpákon, dinamikus torlódási és szolgáltatási árazáson, okosparkoláson és ride-sharingen át a közösségi közlekedés optimalizálásáig terjednek. A végén a vehicle-to-grid működés MI-alapú optimalizálását is bemutatja, ahol az elektromos autók akkumulátorai a városi energiarendszer részeivé válhatnak.

Videó navigáció

Fejezetek

MI-támogatott