Bob Friday, Chief AI Officer, Juniper Networks 

Juniper AIOps and Location

Summits AI & ML
Bob Friday Headshot
The screenshot of a presentation slide is titled, “Mist Auto Placement/Rotation With AI/ML.” There are four bullet points underneath on the left that read: Orchestration of RTT-FTM, VBLE, patent pending AI/ML algorithms for location estimation and directionality<2m 90%; Patent pending location algorithm for estimating LOS & NLOS APS; Ability to auto-detect errors and inconsistencies in existing deployments AND auto-correct; Ability to auto place and orient all APs on a map for new deployments, simplifying Day 1 even more. There are charts on the right side under the heading, “Auto placement — Location Estimation.” The charts show Bias Removal and Correction of Flip Ambiguity.

The latest from Mist AI, and how it can simplify Day 1 operations.

IT professionals, don’t miss this video from Mobility Field Day 2022 as Juniper’s Bob Friday explains how Marvis is becoming AP (access point) aware. The Mist AI™ solution can now automatically place and orient APs post-install against desired plans, and also highlight any misplaced APs for quick correction. What does that mean for you? Substantially simplified AP deployment from Day 1 and reduced overall WLAN installation costs.

Show more

You’ll learn

  • How Mist uses AI and ML for auto placement and location estimation 

  • How to correct deployments for brownfields 

  • How to place APs for greenfield deployments 

Who is this for?

Network Professionals Business Leaders


Bob Friday Headshot
Bob Friday
Chief AI Officer, Juniper Networks 

Guest speakers

Sunalini Sankhavaram Headshot
Sunalini Sankhavaram
Sr Director, Product Management​, Juniper Networks 


0:11 so indoor location has been one of these

0:13 passions there for 20 years the mission

0:15 to make indoor location on par

0:19 make a must have in the industry right

0:20 some i've been working on since my

0:21 aerospace adventure and what we're doing

0:24 here right now we're starting to make

0:26 marvelous

0:27 location aware and why is that important

0:30 it's important for things like coverage

0:33 right we want marvis to be started aware

0:34 of what happens when people start to

0:37 move their access points right and what

0:40 we're announcing today is really

0:43 the beginnings of identifying ap

0:46 location right

0:47 and this really brings in two pieces

0:49 into the marvelous adventure i mean

0:51 right now when you take the missed

0:52 network and deploy it if you look at the

0:54 problems we have with location one of

0:56 them is really getting the whole

0:58 deployment process easier now we made

0:59 that easier with taking echo how designs

1:02 make it very easier to port these

1:04 designs

1:05 into miss marvis that's part of the

1:07 deployment process

1:09 you know we made it easier when

1:10 deployers are actually deploying these

1:12 access points right with the mobile app

1:14 making sure they actually take pictures

1:16 locate where these aps are getting

1:18 deployed where we're taking the next

1:20 step is really bringing

1:22 auto ap placement to the party right

1:24 this is verifying that aps are actually

1:27 where they are supposed to be positioned

1:30 so this is probably the next big

1:31 announcement is really this is all based

1:34 on the fine time measurement of locating

1:36 making sure these access points are

1:37 relatively located where they're

1:38 supposed to be and why is this important

1:40 to marvis well i'm just interruption

1:42 there it's fine time measurement plus a

1:44 lot of aiml happening in the back end

1:47 and that's probably the you know the

1:49 other the other algorithm i talked about

1:51 is this

1:52 graph algorithm

1:54 is making sure we can actually locate

1:56 these ap's really so there's a

1:58 quadrilateral chain algorithm that

2:00 actually once you know the relative

2:01 location to the aps

2:03 there's a bunch of fancy map to make

2:05 sure those aps fit that data

2:08 right you can kind of visualize right

2:10 once i know the relative

2:12 distance between aps

2:14 the question is how do you make sure you

2:15 find the right geometry of where those

2:18 ap codes and that's the magic

2:21 just a simple question

2:24 is this so that if i if my installer

2:26 puts the ap

2:27 at a location that's not

2:29 where it is on the map it's going to

2:31 adjust the map or is it going to tell

2:34 the installer you need to move this one

2:36 three meters south because we'll get

2:38 better results

2:40 it it is going to tell

2:42 that to solve the problem of where ap is

2:44 lying about their location so where the

2:47 placement is on the map is right the

2:50 data doesn't match right so yeah so

2:51 keith let's it's basically the the plan

2:53 versus the actual right the age-old

2:55 problem so

2:56 we take we already make it easy because

2:58 we ingest that map with ap placements

3:00 today in the missed portal right

3:02 but that plan says oh mr installer put

3:04 the ap here and the installer goes

3:06 inside

3:08 says ah it's here or here it doesn't

3:09 matter or in some cases i've heard it's

3:11 opposite the building also right so

3:13 that's where they install and we've had

3:15 multiple customers report hey i'm

3:17 troubled shooting a wireless issue it's

3:19 on this ap 12 but this ap 12 is not

3:22 where it's on the map it's somewhere

3:23 else can you help me fix that right

3:24 different floors different areas so just

3:26 a quick one on on the analysis of what's

3:28 behind the scenes to make this happen

3:31 yeah so this is basically fine time

3:33 management all we all know about doll 11

3:35 mc this is basically leveraging the fine

3:37 time management it has never become an

3:38 industry standard in our clients you

3:40 know we are working with

3:42 customers like zebra that support it

3:44 it's usually not supported most of our

3:46 phones but really leveraging that to

3:47 actually find the distance between these

3:49 access points and that's and with the ai

3:51 magic that's what let's locate these aps

3:54 accurately

3:55 uh the other thing we're starting to

3:56 bring latin long into location and this

3:58 is two reasons for doing this one is for

4:00 connectivity for marvis another is for

4:03 our indoor location services

4:05 with that i think when you actually see

4:07 the demo you'll start to see the magic

4:09 of this of making sure these aps

4:11 actually get put where they belong

4:13 there's probably the guest star the big

4:15 retail customer that basically had the

4:16 problem of

4:17 complaining about coverage there was an

4:19 access point on the map when he went to

4:21 the back room the access point wasn't

4:23 there right these are the type of

4:24 problems you deal with in

4:26 real deployments and the the two key

4:28 things i want to talk about right where

4:29 does ai come into play so you heard

4:31 about rtt you know find uh finding

4:33 measurement yes we are definitely

4:34 adopting that but that doesn't help

4:36 solve non-linear side problems and i

4:38 have to solve in a healthcare deployment

4:40 in a school network in a university i

4:43 have to be able to solve for

4:46 verifying the location of an ap install

4:48 which is not line of sight and that's

4:50 where the ai ml engine comes into play

4:51 to say okay what can i do in terms of in

4:53 addition to rtt finding measurement

4:57 where can i solve for the nylon sides

4:58 that's number one number two we want to

5:00 solve this not just for greenfield

5:02 deployment or installer going on site

5:03 and putting an ap and how they can

5:04 automatically place apis on a floor plan

5:07 we also want to solve this for

5:08 brownfield deployments for customers

5:10 who've been deployed with us for the

5:11 last two or three or four years we want

5:13 to give them the benefit to get a view

5:15 of the actual install versus what was

5:18 the plan installed on the icahoe import

5:20 and our takeaway from here is

5:23 that fine time measurement thing

5:25 and probably the other thing is there's

5:27 two

5:28 key algorithms here one is getting the

5:29 eps placed correctly the other key

5:32 algorithm is dealing with none line of

5:33 sight

5:34 you know the good thing about fine time

5:36 management it works great line of sight

5:38 questions how do you handle none line of

5:40 sight and that's the problem the problem

5:41 you can deal with

5:43 a question with the ap placement

5:45 is there an assumption that all the aps

5:47 will be ceiling mounted or what if you

5:49 like mount aps on a wall or set it on a

5:51 table or a cabinet would that have any

5:53 impact initially we're looking at the

5:55 ceiling that's 90 percent of the

5:56 population and then we'll involve this

5:58 with the antenna model to also look at

5:59 tilt as well as as xyz height and

6:02 probably the other key thing i didn't

6:03 mention is this also takes care of

6:05 rotation

6:06 in addition to ap it's actually

6:08 doing the directionality making sure the

6:10 ap is pointed in the right direction

6:12 that's for a virtual ble

6:14 location use case

6:16 could you tell this vertical then yes

6:18 yeah we have the vba arrays for that

6:20 exactly

6:21 yeah so really quickly from a a

6:23 demonstration perspective

6:25 for brownfield right what you'll see

6:28 here is these are ap's on a map customer

6:30 goes and clicks and this is going to

6:32 show up in the next couple of months on

6:33 your dashboards customer goes and clicks

6:36 auto locate auto place the back end

6:38 system run this magic and automatically

6:40 will then place the aps in the right

6:42 placements on the floor plan so urine

6:44 will do anything except say auto locate

6:46 auto place right and now let's get to so

6:49 i just add absolutely on the so on the

6:51 orientation troy um it doesn't actually

6:54 so much matter right we have a

6:56 we have a accelerometer in the ap so we

6:58 know the orientation of the aps but it

7:00 doesn't so much matter the orientation

7:02 for refine time measurement

7:04 for connectivity or for locating them

7:07 it's more for location not for

7:08 connectivity correct yeah from a

7:10 from a ap

7:12 placement it doesn't so much matter

7:14 so you saw the magic we do for

7:16 brownfield right so imagine you heard

7:17 about our large customers and they all

7:19 want to get a better visibility of hey i

7:22 know i gave a plan i know the installer

7:24 is a good installer but we know it

7:26 should happen

7:27 the brownfield mechanism no anchors

7:29 needed we would auto detect anchors we

7:31 will then place the grid for you and

7:33 place the aps in the maps again marv is

7:35 coming into play order remediation all

7:37 you say is authorize the action

7:39 marvelous takes care of the rest now

7:41 comes the green field i'm deploying

7:43 large networks today multi-floor

7:45 buildings huge campuses or thousands of

7:48 branch locations what can i do to save

7:50 time during the install process right

7:53 today

7:54 uh installer goes in the maps already in

7:57 the system

7:59 they either have the ap placements as

8:01 per the ikaho import or they haven't

8:03 done that import installers are going in

8:05 and pulling ap's from the sidebar and

8:07 placing them on the map three ap s4ap is

8:09 not a big deal just start talking about

8:11 10 20 30 ap you have to do four or five

8:13 sites a night that becomes a nightmare

8:15 so what can we do from a green field

8:18 install perspective to save the

8:19 installers time and therefore the end

8:21 customer money because it's much faster

8:24 now to get the network set up and get

8:26 the real

8:28 location of apes on the map again

8:30 no need for any adjacent technology

8:32 we're leveraging what we already have

8:33 and our ai location engine to deliver

8:36 this so as the installer goes about

8:38 their work they essentially give us the

8:40 measurement of just the first word

8:41 they've deployed anywhere in the ceiling

8:43 that they have and then that's it once

8:46 the installer gives us the measurement

8:48 for because we need some source of truth

8:50 and while the installer is doing their

8:51 job you're gonna give us hey here's the

8:53 x y for this location by just putting it

8:55 on the floor plan that's it once they're

8:58 done that they go again and say auto

9:00 just like you do rm optimize now you now

9:02 go and do auto place auto locate

9:05 and there you see it

9:07 you take the action the ai engine in the

9:09 back end then runs its calculations gets

9:11 all the information about all the aps

9:14 and

9:14 magic happens literally

9:17 there you go

9:19 it's as easy as that how much time do

9:21 you need can you do it as soon as you're

9:23 done installing the last ap that is

9:25 correct the the goal that we have in the

9:27 testing you've done so far is within 10

9:28 to 20 minutes it should be all done so

9:30 the goal is to like when the install is

9:32 done make sure that they're click on

9:33 auto place and it happens and then they

9:35 can go home yeah

9:37 you don't have to come back the next day

9:38 you don't have to do a you know a site

9:40 survey to validate

9:41 ai is doing the job for you yes or well

9:44 after you do that auto

9:47 placement do you provide

9:49 a deviation

9:50 say

9:51 i did upload a map using echo how

9:54 and i have the ap placements there

9:56 but when you run this and it's all

9:58 changed will you give me that deviation

10:00 so in the brown field we will be giving

10:02 you a view also of this is where the ap

10:05 actually so right now you saw marvis

10:06 take care of it and just move the aps

10:08 but we're also looking at again is to

10:10 give you the evidence to say and let you

10:12 authorize that change because you may

10:13 say no let the aps be where they're at

10:16 don't accept the change or you may say

10:19 let me send a tech on site to make that

10:20 change happen right so that the reality

10:22 matches what has been built we will be

10:25 able to give you a view of here's the ap

10:27 placement that the location engine found

10:29 which is where this is where the ap is

10:31 at as per the plan would you like to

10:33 accept the new placement or do you like

10:35 to trigger a tech workflow to go and

10:37 move the ap okay

10:39 is this ga today no this is in the works

10:42 we're going to be getting into betas i

10:44 was in the next couple of months and you

10:45 but you will see this in the i would say

10:47 early for time but we are super excited

10:49 about this we've we've been quiet about

10:51 this because we've been running our own

10:53 simulation validations large networks

10:55 small networks domes multi-floor you all

10:58 know about our nvidia deployment uh

11:00 we've actually tried this out there done

11:02 simulations to see how good we are we

11:04 are now confident we can ship this very

11:06 soon so yes still include in the

11:09 marvelous license

11:10 this would be part wi-fi storage part

11:12 marvelous correct

11:14 once you have this data you also then

11:17 use the ai information to

11:20 say you need another ap here

11:22 bingo so you talk you heard bob talk

11:25 about the coverage whole aspect now that

11:27 we'll get the actuality placements mars

11:30 and already know there's a coverage

11:31 whole problem but it'll show you based

11:33 on what was imported that coverage whole

11:35 efficacy becomes much higher now because

11:37 we have the actual location of the ap so

11:38 absolutely yes short answer yes yes

11:43 um so the workflow flow for that is the

11:45 um is it auto placing them um

11:48 directly or do you have to place those

11:50 aps on the floor first and then auto

11:52 placement occurs auto placement directly

11:57 into the organization i hit auto place

11:59 and they're on the floor there you go

12:01 perfect yeah yes sir as a follow-up

12:03 keith's question instead of adding aps

12:05 could it suggest moving aps for better

12:07 placement

12:08 of those on the map so that we already

12:10 do as part of our coverage hole where we

12:12 will say assume that the reality matches

12:14 what was the plan and our current marvin

12:17 uh marvis coverage whole action will

12:18 tell you if an ap placement is

12:20 recommended or an ad is recommended that

12:22 that is taking care of today

12:24 so i know i'm over time thank you so

12:26 much okay one last question just for you

12:28 okay i just have a quick question i

12:30 missed how many truths that you actually

12:32 need to key in and then i missed what

12:34 the accuracy is is it like a meter two

12:37 meters two meters ninety percent of what

12:38 the current stimulation is is showing we

12:40 have actually gotten less than one meter

12:42 but i'm not gonna say that publicly oh i

12:44 just did it

12:48 what i stand by and and three truths did

12:50 you say in terms of for the green field

12:53 deployments where we have no history and

12:54 no data to collect we look at these

12:56 three anchors yeah but that is like

12:57 quality installed installing we have the

12:59 mobile app yeah they just point that

13:00 mobile app gets the xy that it's as

13:02 simple as that

13:03 okay and this helps us also solve the

13:05 problem of i don't need

13:07 i know i'm not supposed to say this and

13:08 bob said to me don't say it

13:10 but i don't need an outside signal to

13:12 tell me the location of my anchor right

13:15 because often times you get in

13:16 deployments where you don't have an

13:17 outside satellite signal what do you do

13:19 there right so therefore in this

13:21 situation i'm handling both

13:23 the outside deployments which have

13:25 satellite signal which i don't need or

13:27 internal deployments only right again

13:30 think of

13:31 of hospitals and patient rooms and think

13:33 of higher ed and hallways you don't have

13:36 underside to a window that gives you

13:38 some signal

13:39 all right with that over to the next

13:42 session any last questions for bob

13:44 before we transition

13:45 tom is here go ahead tom

13:47 well anybody have any last questions for

13:49 bob

13:50 yeah i'm unliking after this oh then i

13:52 do have one question if i may

13:54 going way back on one of your earlier

13:56 slides when you showed the tools there

13:58 was a bunch of circles that said new

14:01 is that what marvelous release seven is

14:04 yes so if you look in the chat in the

14:06 data science toolbox i mean the team is

14:08 like this particle chain thing that's a

14:10 new algorithm for solving the geometry

14:13 of auto ap placement so that's a new

14:15 algorithm that we've added into the

14:16 toolboxes and so if we define release

14:18 seven that's what it is

14:20 yes i would add that to release that

14:22 that's great okay

14:24 well we had what you'd mentioned earlier

14:27 about adding lat long

14:29 how does that interface with what you

14:32 just mentioned is that

14:33 that that's for our location there's two

14:36 stories here one is connectivity like

14:38 marvis which is more around network

14:39 operations the other part is for indoor

14:41 location services

14:43 a lot of our mobile app customers want

14:45 latin long to be returned in digit xy

14:48 and that ties back to the boris client

14:51 android zebra windows because that is

14:52 essentially a combination of location

14:54 and network so that's how the lat long

14:56 piece comes across across the board

14:58 would we just set light long on those

15:00 three anchors

15:02 no we would actually get the latch long

15:04 from the client

15:05 and then and do all the calculation the

15:06 back end so

15:08 all right with that thank you so much

15:10 and overall to the next session actually

15:12 going to take a real quick break because

15:14 we've been up here for an hour and a

15:15 half and we want to make sure that

15:16 everybody has a chance to stretch their

15:17 legs

Show more