Transcriptions
Note: this content has been automatically generated.
00:00:00
preceding up the word cars i ceasing to exist the way we know today grades uh becomes far with more
00:00:06
automation and m. m. m. battery no stock you already know the computer networks are gonna be destructive itself
00:00:12
itself it doesn't work well it can leave a vehicle destruction it's
00:00:15
part and that's why mission critical software requires high reliability
00:00:21
my research it is here to provide liability protection against crashes and delay falls
00:00:28
via built deployed and tested smart be reliably distributions that's or expertise is
00:00:34
now we don't apply our solutions in other areas like autonomous cars and a data centres
00:00:41
so what i'm gonna do is explain the challenges involved in in doing so
00:00:45
and from your side if you like it i would like to get he's so
00:00:49
that i can continue with my customer discovery process that's going on right now
00:00:56
well uh we went up like autonomous cars i don't understand the problem by profiling the software
00:01:03
we don't have access to a car software the best we can do it with good
00:01:06
so for what we did is we landed softly for a long time and realise
00:01:11
that the normal response times on one millisecond that's how it's designed to be but
00:01:16
every now and then it has a peak of two fifty three hundred millisecond
00:01:20
if this were a car on hiding the us that is doing a emergency stopping application then it would mean that
00:01:27
the car would be inclined more for five to six meters that's quite scary you wouldn't dissident are like that
00:01:34
with our but in the software we can go from blind modified to six meters go blind mode
00:01:38
of under one meter more importantly we did use the fork rake by several orders of magnitude
00:01:45
the basic idea is very simple if something isn't reliable you replicate
00:01:50
now people have been doing for thirty years for discourse passive replication bit primaries in standby is
00:01:55
this was very powerful crash false but does not target delay for it's it's important in real time setting
00:02:01
what happens here is your primary that's doing the control and the backup that
00:02:05
detective the primaries not working correctly and comes back up a bit
00:02:09
to take over the control the other alternative is active didn't sit there
00:02:14
and you have two or more primaries controlling the car simultaneously
00:02:18
you're the problem that arises bird brain you could have these controllers giving competing
00:02:23
set points and driving the car in competing directions that's not receive
00:02:28
the way to solve it is by consensus rating these
00:02:31
replicated controllers agree and then move forward but consensus is not possible to do in real time or it will be like
00:02:39
engine that's where i think you toss lights we have found a class of systems
00:02:44
where we can do like with agreement is of a full blown consensus