Driving Innovation at the Edge: Chuck Gershman on AI, Silicon and Safer Autonomous Vehicles

[Music]

hi friends welcome again to another

episode of beyond the clouds EDG to Edge

transformation and transformation has to

happen at every level I've been meeting

some amazing entrepreneurs and Chuck uh

I met him at a silicon Catalyst

presentation for the last six months it

was really amazing to hear the story of

all autonomous Imaging for once AI

doesn't start for artificial

intelligence actually it does a little

bit we we use AIML in our system the AI

has a double on Tandra in this

particular case on purpose that's

perfect so uh Chuck tell us how did you

end up being the Chuck of that we know

now and how did you end up being in this

company which is so

amazing well I've been in the tech

industry for three plus decades um I've

had a very extensive career on the

semiconductor uh side um I've also

worked in iot and networking both on the

digital and analog side of the business

so a long long career at the chip level

um at the subsystem level and even at

the system level in some of the

companies so seen a little bit of

everything and uh tell us a little bit

about all image and what do you guys do

so AI uh we is ay that um that I

co-founded with my partner Gan patii

roughly 5 years ago at the time Gan had

been working on Advanced uh imaging

technology in the thermal domain he and

I had worked on a project for the US Air

Force that was quite sophisticated uh

the beauty of that Air Force project was

it was a proof of concept and the

company that we worked for was able to

retain all the intellectual property

rights and that company's intellectual

property rights eventually came to us to

start start a new company so what got us

motivated and what we do at Ali is we

basically build a thermal night vision

system uh and then we couple that with

perception software specifically AIML

software and the purpose of that is

basically for safety so we build systems

that go into automobiles trucks cars

buses that are able to Independent of

the driver uh visualize the entire scene

and make uh real-time decisions on being

able to avoid collisions accidents

pedestrians animals Etc the system is

designed in such a way that the driver

of course could take action first but if

the driver fails to respond the system

can take over an ample time to make a

safety decision uh and the goal of the

company is simply to save

lives and honestly when you told us

about the data uh how many pedestrians

get hurt especially when it's dark that

was quite uh quite alarming uh can you

tell us a little bit about that yeah so

the statistics are not at all comforting

in any way shape or form annually 1.4

million people are killed in automobile

accidents it is a very very significant

um number however when you drill down on

the numbers and 1.4 is a big number half

of the people who are killed in car

accidents weren't actually in the car

they were pedestrians they were on

bicycles they were on motorcycles they

were on scooters they were walking down

the street um that's one out of every

two people killed and then if you drill

down even further you'll find that 76%

of those people actually were killed at

night and this incident rate occurs not

only for deaths but for just for any

kind of incident so basically you know

um automobile incidents at night is more

than three qus of all incidents that

occur especially when it comes to

fatalities uh it's quite staggering can

you combine this technology with many

other technologies that are existing in

the market

and if so how does it work in

tandem yeah so um in know we don't

Envision um nor do we Advocate uh that

the use of our particular sensor and

software suite should be used

independent of other sensors and

software that are already existing in

the vehicles or coming to market the

basic premise is that we fill a gap in

the overall sensor Suite with the

uniqueness of what we do uh we build a

system based on thermal imaging and then

computer vision based on thermal imaging

so there's still going to be visual

cameras in the cars and they're still

going to do what they do but they

degrade at night just like humans

degrade at night they degrade in foul

weather so whether it's raining snowing

sleep or if it's just Smoky or foggy um

cameras will degrade thermal cameras

their degradation in these conditions is

a couple orders of magnitude less

degradation we supplement these visual

cameras we supplement Radars we

supplement uh Imus we supplement all the

other sensors to make a more robust

overall system uh and again going back

to the statistics the nighttime

operation of these vehicles is

definitely in question just like humans

degrade at night in terms of their

visualization so do the current sensor

Suites that are in the cars and they

just need to be better you said you've

done some work with the Air Force have

you been able to do some testing of your

systems on the road and uh with humans

and this

point well um subsequent to our Air

Force work which was done before we

founded the company we have done work

with the dod uh a as Al U mainly for

what we call off-road so we've developed

systems that are not on pavement the

sensor systems obviously work fine but

the software systems need to be slightly

modified to work whether it's onroad or

off-road so yes we do that um and we've

already done that with the dod um but in

terms of actual testing so the national

highway traffic safety administration

which is part of the Department of

Transportation actually is is well aware

of the statistics that I just

articulated um and it has become quite

an issue so we've seen that the US

Congress under the infrastructure act

has actually created line items to

address um pedestrian safety uh as well

as the European Union has now created

regulatory respon so in the United

States we now have regulations out there

that are going to be mandatory by the

end of the year that says You must be

able to stop for a pedestrian at night

at high speed um and that's going to be

a minimum requirement to sell a vehicle

in the United States by 2028 we see

similar regulations in the in the

European Union so these problems are not

lost on governments now about a year ago

Nitsa published a preliminary um set of

recommendations which ultimately is

going to become the final regulations so

we took those

specifications uh we went out to Detroit

and rented time on a test track we took

a we got a test vehicle we took our our

full system we mounted it on the system

we hired a test driver and we hired a

third-party test company and we ran

through the suite of nit a test now you

said with real people we asked our

interns whether they wanted to volunteer

to run in front of the cars at 45 miles

an hour we couldn't get anybody to agree

so we Ed what's called electronic

dummies um so these are mannequins that

are on tracks they're heated um to give

a thermal signature and you drive the

vehicle down the track you hit a laser

that triggers a certain speed and time

and that triggers the dummy to run in

front of the vehicle such the time it

such that right when the car gets there

so that was the test that we did we did

this um 12 times times under different

configurations we never struck the dummy

now the same test that we were testing

against has been tested against 21

production vehicles and under the

conditions that we tested against almost

every one of those Vehicles ran over the

dummy at one point in time or more the

vehicles that are out there are not

currently capable of passing these tests