vtt/episode_039_small.vtt

WEBVTT

00:00.000 --> 00:02.800
 The following is a conversation with Colin Engel.

00:02.800 --> 00:05.840
 He's the CEO and cofounder of iRobot,

00:05.840 --> 00:09.680
 a robotics company that for 29 years has been creating robots

00:09.680 --> 00:12.600
 that operate successfully in the real world.

00:12.600 --> 00:15.600
 Not as a demo or on a scale of dozens,

00:15.600 --> 00:18.880
 but on a scale of thousands and millions.

00:18.880 --> 00:24.320
 As of this year, iRobot has sold more than 25 million robots

00:24.320 --> 00:28.200
 to consumers, including the Roomba vacuum cleaning robot,

00:28.200 --> 00:30.000
 the Bravo floor mopping robot,

00:30.000 --> 00:34.000
 and soon the Terra lawn mowing robot.

00:34.000 --> 00:37.680
 29 million robots successfully operating autonomously

00:37.680 --> 00:39.680
 in real people's homes.

00:39.680 --> 00:42.120
 To me, it's an incredible accomplishment

00:42.120 --> 00:45.120
 of science, engineering, logistics,

00:45.120 --> 00:48.800
 and all kinds of general entrepreneurial innovation.

00:48.800 --> 00:51.360
 Most robotics companies fail.

00:51.360 --> 00:56.880
 iRobot has survived and succeeded for 29 years.

00:56.880 --> 01:00.160
 I spent all day at iRobot, including a long tour

01:00.160 --> 01:03.520
 and conversation with Colin about the history of iRobot,

01:03.520 --> 01:06.720
 and then sat down for this podcast conversation

01:06.720 --> 01:08.560
 that would have been much longer

01:08.560 --> 01:10.760
 if I didn't spend all day learning about

01:10.760 --> 01:13.960
 and playing with the various robots in the company's history.

01:13.960 --> 01:17.480
 I'll release the video of the tour separately.

01:17.480 --> 01:20.720
 Colin, iRobot, its founding team,

01:20.720 --> 01:23.200
 its current team, and its mission

01:23.200 --> 01:26.200
 has been and continues to be an inspiration to me

01:26.200 --> 01:28.880
 and thousands of engineers who are working hard

01:28.880 --> 01:33.000
 to create AI systems that help real people.

01:33.000 --> 01:35.640
 This is the Artificial Intelligence Podcast.

01:35.640 --> 01:38.000
 If you enjoy it, subscribe on YouTube,

01:38.000 --> 01:41.240
 give it five stars on iTunes, support it on Patreon,

01:41.240 --> 01:43.280
 or simply connect with me on Twitter

01:43.280 --> 01:47.120
 at Lex Freedman, spelled F R I D M A N.

01:47.120 --> 01:51.080
 And now, here's my conversation with Colin Engel.

01:52.120 --> 01:55.120
 In his 1942 short story, Run Around,

01:55.120 --> 01:58.520
 from his iRobot collection, Asimov,

01:59.400 --> 02:02.840
 proposed the three laws of robotics in order,

02:02.840 --> 02:06.800
 don't harm humans, obey orders, protect yourself.

02:06.800 --> 02:07.640
 So two questions.

02:07.640 --> 02:11.640
 First, does the Roomba follow these three laws?

02:11.640 --> 02:14.760
 And also, more seriously,

02:14.760 --> 02:17.120
 what role do you hope to see robots take

02:17.120 --> 02:20.280
 in modern society and in the future world?

02:20.280 --> 02:25.280
 So the three laws are very thought provoking

02:25.720 --> 02:30.720
 and require such a profound understanding

02:31.360 --> 02:36.280
 of the world a robot lives in,

02:36.280 --> 02:38.360
 the ramifications of its action

02:38.360 --> 02:40.040
 and its own sense of self,

02:40.040 --> 02:45.040
 that it's not a relevant bar,

02:46.640 --> 02:48.360
 at least it won't be a relevant bar

02:48.360 --> 02:50.160
 for decades to come.

02:50.160 --> 02:54.560
 And so, if Roomba follows the three laws,

02:54.560 --> 02:56.840
 and I believe it does,

02:58.040 --> 03:00.920
 it is designed to help humans not hurt them,

03:00.920 --> 03:03.120
 it's designed to be inherently safe,

03:03.120 --> 03:05.960
 and we design it to last a long time.

03:07.200 --> 03:11.600
 It's not through any AI or intent on the robot's part.

03:11.600 --> 03:14.960
 It's because following the three laws

03:14.960 --> 03:18.200
 is aligned with being a good robot product.

03:19.600 --> 03:23.120
 So I guess it does,

03:23.120 --> 03:27.240
 but not by explicit design.

03:27.240 --> 03:28.800
 So then the bigger picture,

03:28.800 --> 03:31.560
 what role do you hope to see robotics,

03:31.560 --> 03:36.560
 robots take in what's currently mostly a world of humans?

03:37.360 --> 03:42.360
 We need robots to help us continue

03:42.360 --> 03:44.960
 to improve our standard of living.

03:46.160 --> 03:51.160
 We need robots because the average age of humanity

03:52.840 --> 03:55.040
 is increasing very quickly,

03:55.040 --> 03:59.760
 and simply the number of people young enough

03:59.760 --> 04:02.480
 and spry enough to care for the elder

04:03.880 --> 04:07.880
 growing demographic is inadequate.

04:08.800 --> 04:11.440
 And so what is the role of robots?

04:11.440 --> 04:15.000
 Today, the role is to make our lives a little easier,

04:15.000 --> 04:17.400
 a little cleaner, maybe a little healthier.

04:18.520 --> 04:22.160
 But in time, robots are gonna be the difference

04:22.160 --> 04:25.720
 between real gut wrenching declines

04:25.720 --> 04:28.360
 in our ability to live independently

04:28.360 --> 04:30.280
 and maintain our standard of living,

04:30.280 --> 04:34.920
 and a future that is the bright one

04:34.920 --> 04:37.520
 where we have more control of our lives,

04:37.520 --> 04:40.560
 can spend more of our time focused

04:40.560 --> 04:43.280
 on activities we choose.

04:44.680 --> 04:48.840
 And I'm so honored and excited to be

04:48.840 --> 04:50.520
 playing a role in that journey.

04:50.520 --> 04:51.800
 So you've given me a tour,

04:51.800 --> 04:54.280
 you showed me some of the long histories now,

04:54.280 --> 04:57.280
 29 years that iRobot has been at it,

04:57.280 --> 04:59.280
 creating some incredible robots.

04:59.280 --> 05:01.280
 You showed me PacBot,

05:01.280 --> 05:03.200
 you showed me a bunch of other stuff

05:03.200 --> 05:04.600
 that led up to Roomba,

05:04.600 --> 05:08.760
 that led to Brava and Tara.

05:08.760 --> 05:13.760
 So let's skip that incredible history

05:14.080 --> 05:15.080
 in the interest of time,

05:15.080 --> 05:16.160
 because we already talked about it,

05:16.160 --> 05:18.080
 I'll show this incredible footage.

05:18.080 --> 05:22.680
 You mentioned elderly and robotics and society.

05:22.680 --> 05:26.320
 I think the home is a fascinating place for robots to be.

05:26.320 --> 05:29.840
 So where do you see robots in the home?

05:29.840 --> 05:31.680
 Currently, I would say, once again,

05:31.680 --> 05:34.560
 probably most homes in the world don't have a robot.

05:34.560 --> 05:36.200
 So how do you see that changing?

05:36.200 --> 05:39.880
 Where do you think is the big initial value add

05:39.880 --> 05:41.960
 that robots can do?

05:41.960 --> 05:46.960
 So iRobot has sort of over the years narrowed in on the home,

05:48.160 --> 05:53.160
 the consumer's home as the place where we want to innovate

05:53.160 --> 05:58.160
 and deliver tools that will help a home be

06:00.840 --> 06:04.280
 a more automatically maintained place,

06:04.280 --> 06:06.840
 a healthier place, a safer place

06:06.840 --> 06:11.520
 and perhaps even a more efficient place to be.

06:11.520 --> 06:16.520
 And today vacuum we mop, soon we'll be mowing your lawn,

06:17.000 --> 06:21.520
 but where things are going is,

06:22.760 --> 06:27.120
 when do we get to the point where the home,

06:27.120 --> 06:29.160
 not just the robots that live in your home,

06:29.160 --> 06:32.200
 but the home itself becomes part of a system

06:32.200 --> 06:36.000
 that maintains itself and plays an active role

06:36.000 --> 06:40.800
 in caring for and helping the people live in that home.

06:40.800 --> 06:43.240
 And I see everything that we're doing

06:43.240 --> 06:46.200
 as steps along the path toward that future.

06:46.200 --> 06:47.760
 So what are the steps?

06:47.760 --> 06:51.760
 So if we can summarize some of the history of Roomba,

06:53.280 --> 06:55.560
 you've mentioned, and maybe you can elaborate on it,

06:55.560 --> 06:57.280
 but you mentioned that the early days

06:57.280 --> 07:02.280
 were really taking a robot from something that works

07:02.360 --> 07:04.920
 either in the lab or something that works in the field

07:04.920 --> 07:09.920
 that helps soldiers do the difficult work they do

07:10.240 --> 07:12.680
 to actually be in the hands of consumers

07:12.680 --> 07:15.680
 and tens of thousands, hundreds of thousands of robots

07:15.680 --> 07:18.520
 that don't break down over how much people love them

07:18.520 --> 07:21.480
 over months of very extensive use.

07:21.480 --> 07:22.880
 So that was the big first step.

07:22.880 --> 07:26.040
 And then the second big step was the ability

07:26.040 --> 07:29.920
 to sense the environment, to build a map, to localize,

07:29.920 --> 07:32.600
 to be able to build a picture of the home

07:32.600 --> 07:34.640
 that the human can then attach labels to

07:34.640 --> 07:38.440
 in terms of giving some semantic knowledge

07:38.440 --> 07:40.920
 to the robot about its environment.

07:40.920 --> 07:44.880
 Okay, so that's like a huge, two big, huge steps.

07:46.280 --> 07:47.560
 Maybe you can comment on them,

07:47.560 --> 07:51.080
 but also what is the next step

07:51.080 --> 07:54.760
 of making a robot part of the home?

07:54.760 --> 07:55.600
 Sure.

07:55.600 --> 08:00.600
 So the goal is to make a home that takes care of itself,

08:01.320 --> 08:03.720
 takes care of the people in the home

08:03.720 --> 08:07.880
 and gives a user an experience of just living their life

08:07.880 --> 08:10.880
 and the home is somehow doing the right thing,

08:10.880 --> 08:14.160
 turning on and off lights when you leave,

08:14.160 --> 08:17.280
 cleaning up the environment.

08:17.280 --> 08:22.280
 And we went from robots that were right in the lab,

08:22.280 --> 08:26.760
 but were both too expensive and not sufficiently capable

08:26.760 --> 08:30.480
 to ever do an acceptable job of anything

08:30.480 --> 08:34.080
 other than being a toy or a curio in your home

08:34.080 --> 08:39.080
 to something that was both affordable

08:38.880 --> 08:42.440
 and sufficiently effective to drive,

08:42.440 --> 08:45.360
 be above threshold and drive purchase intent.

08:47.520 --> 08:50.000
 Now we've disrupted the intent of the work

08:50.000 --> 08:54.400
 and now we've disrupted the entire vacuuming industry.

08:55.480 --> 08:59.840
 The number one selling vacuums, for example, in the US

08:59.840 --> 09:02.480
 are Roombas, so not robot vacuums,

09:02.480 --> 09:05.160
 but vacuums and that's really crazy and weird.

09:05.160 --> 09:08.080
 We need to pause that, I mean, that's incredible.

09:08.080 --> 09:10.520
 That's incredible that a robot

09:10.520 --> 09:15.520
 is the number one selling thing that does something.

09:15.560 --> 09:16.400
 Yep.

09:16.400 --> 09:18.240
 Something as essential as vacuuming.

09:18.240 --> 09:20.120
 So we're... Congratulations.

09:20.120 --> 09:20.960
 Thank you.

09:20.960 --> 09:22.440
 It's still kind of fun to say,

09:22.440 --> 09:26.600
 but just because this was a crazy idea

09:26.600 --> 09:30.920
 that just started in a room here,

09:30.920 --> 09:33.800
 we're like, do you think we can do this?

09:35.000 --> 09:36.160
 Hey, let's give it a try.

09:38.000 --> 09:40.440
 But now the robots are starting

09:40.440 --> 09:42.840
 to understand their environment.

09:42.840 --> 09:45.280
 And if you think about the next step,

09:45.280 --> 09:47.880
 there's two dimensions.

09:48.720 --> 09:53.040
 I've been working so hard since the beginning of iRobot

09:53.040 --> 09:55.080
 to make robots are autonomous,

09:55.080 --> 09:59.120
 that they're smart enough and understand their task enough

09:59.120 --> 10:02.480
 that they can just go do it without human involvement.

10:04.160 --> 10:07.440
 Now what I'm really excited and working on

10:07.440 --> 10:09.520
 is how do I make them less autonomous?

10:09.520 --> 10:14.520
 Meaning that the robot is supposed to be your partner,

10:15.080 --> 10:20.080
 not this automaton that just goes and does what a robot does.

10:20.240 --> 10:22.440
 And so that if you tell it,

10:22.440 --> 10:27.160
 hey, I just dropped some flour by the fridge in the kitchen.

10:27.160 --> 10:29.000
 Can you deal with it?

10:29.000 --> 10:31.840
 Wouldn't be awesome if the right thing just happened

10:31.840 --> 10:35.240
 based on that utterance.

10:35.240 --> 10:37.960
 And to some extent that's less autonomous

10:37.960 --> 10:40.120
 because it's actually listening to you,

10:40.120 --> 10:44.400
 understanding the context and intent of the sentence,

10:44.400 --> 10:49.400
 mapping it against its understanding of the home it lives in

10:50.040 --> 10:52.680
 and knowing what to do.

10:52.680 --> 10:56.360
 And so that's an area of research.

10:56.360 --> 10:59.360
 It's an area where we're starting to roll out features.

10:59.360 --> 11:02.880
 You can now tell your robot to clean up the kitchen

11:02.880 --> 11:05.880
 and it knows what the kitchen is and can do that.

11:05.880 --> 11:09.360
 And that's sort of 1.0 of where we're going.

11:10.440 --> 11:12.480
 The other cool thing is that

11:12.480 --> 11:14.640
 we're starting to know where stuff is.

11:14.640 --> 11:16.040
 And why is that important?

11:16.040 --> 11:21.040
 Well, robots are supposed to have arms, right?

11:21.520 --> 11:24.200
 Data had an arm, Rosie had an arm,

11:24.200 --> 11:25.240
 Robbie the robot had an arm.

11:25.240 --> 11:26.680
 I mean, robots are, you know,

11:26.680 --> 11:29.560
 they are physical things that move around in an environment

11:29.560 --> 11:31.280
 they're supposed to like do work.

11:31.280 --> 11:34.120
 And if you think about it,

11:34.120 --> 11:35.440
 if a robot doesn't know anything,

11:35.440 --> 11:38.760
 where anything is, why should it have an arm?

11:38.760 --> 11:43.760
 But with this new dawn of home understanding

11:44.320 --> 11:47.680
 that we're starting to go enjoy,

11:47.680 --> 11:49.320
 I know where the kitchen is.

11:49.320 --> 11:52.040
 I might in the future know where the refrigerator is.

11:52.040 --> 11:55.280
 I might, if I had an arm, be able to find the handle,

11:55.280 --> 11:58.440
 open it and even get myself a beer.

11:58.440 --> 12:01.920
 Obviously that's one of the true dreams of robotics

12:01.920 --> 12:03.520
 is to have robots bringing us a beer

12:03.520 --> 12:05.280
 while we watch television.

12:05.280 --> 12:10.280
 But I think that that new category of tasks

12:11.000 --> 12:14.240
 where physical manipulation, robot arms

12:14.240 --> 12:19.240
 is just a popery of new opportunity and excitement.

12:20.200 --> 12:23.800
 And you see humans as a crucial part of that.

12:23.800 --> 12:26.280
 So you kind of mentioned that.

12:26.280 --> 12:28.960
 And I personally find that a really compelling idea.

12:28.960 --> 12:33.960
 I think full autonomy can only take us so far,

12:33.960 --> 12:35.360
 especially in the home.

12:35.360 --> 12:38.920
 So you see humans as helping the robot understand

12:38.920 --> 12:42.600
 or give deeper meaning to the spatial information.

12:43.600 --> 12:45.680
 Right, it's a partnership.

12:46.800 --> 12:51.800
 The robot is supposed to operate according to descriptors

12:52.800 --> 12:55.620
 that you would use to describe your own home.

12:57.040 --> 13:02.040
 The robot is supposed to in lieu of better direction

13:02.040 --> 13:03.840
 kind of go about its routine,

13:03.840 --> 13:07.520
 which ought to be basically right

13:07.520 --> 13:12.200
 and lead to a home maintained

13:12.200 --> 13:14.840
 in a way that it's learned you like,

13:14.840 --> 13:19.840
 but also be perpetually ready to take direction

13:21.440 --> 13:26.440
 that would activate a different set of behaviors or actions

13:26.880 --> 13:28.880
 to meet a current need

13:28.880 --> 13:32.320
 to the extent it could actually perform that task.

13:32.320 --> 13:33.560
 So I gotta ask you,

13:33.560 --> 13:36.960
 I think this is a fundamental and a fascinating question

13:36.960 --> 13:39.760
 because iRobot has been a successful company

13:39.760 --> 13:42.320
 and a rare successful robotics company.

13:42.320 --> 13:46.720
 So Anki, Gibo, Mayfield Robotics with a Robot Curry,

13:46.720 --> 13:49.160
 SciFi Works, Rethink Robotics.

13:49.160 --> 13:51.320
 These were robotics companies that were founded

13:51.320 --> 13:52.920
 and run by brilliant people.

13:54.000 --> 13:58.760
 But all very unfortunately, at least for us roboticists,

13:58.760 --> 14:02.120
 and all went out of business recently.

14:02.120 --> 14:05.160
 So why do you think they didn't last longer?

14:05.160 --> 14:07.000
 Why do you think it is so hard

14:07.000 --> 14:10.680
 to keep a robotics company alive?

14:10.680 --> 14:14.160
 You know, I say this only partially in jest

14:14.160 --> 14:16.760
 that back in the day before Roomba,

14:17.840 --> 14:22.840
 you know, I was a high tech entrepreneur building robots,

14:23.920 --> 14:26.440
 but it wasn't until I became a vacuum cleaner salesman

14:26.440 --> 14:29.420
 that we had any success.

14:29.420 --> 14:34.240
 So I mean, the point is technology alone

14:34.240 --> 14:37.680
 doesn't equal a successful business.

14:37.680 --> 14:42.680
 We need to go and find the compelling need

14:43.600 --> 14:45.920
 where the robot that we're creating

14:47.640 --> 14:52.640
 can deliver clearly more value

14:52.800 --> 14:55.440
 to the end user than it costs.

14:55.440 --> 14:59.040
 And this is not a marginal thing

14:59.040 --> 15:01.800
 where you're looking at the skin, like, it's close.

15:01.800 --> 15:04.380
 Maybe we can hold our breath and make it work.

15:04.380 --> 15:09.380
 It's clearly more value than the cost of the robot

15:11.560 --> 15:13.920
 to bring, you know, in the store.

15:13.920 --> 15:17.360
 And I think that the challenge has been finding

15:17.360 --> 15:22.360
 those businesses where that's true

15:22.360 --> 15:26.400
 that's true in a sustainable fashion.

15:28.240 --> 15:33.240
 You know, when you get into entertainment style things,

15:34.600 --> 15:38.240
 you could be the cat's meow one year,

15:38.240 --> 15:42.440
 but 85% of toys, regardless of their merit,

15:43.400 --> 15:45.640
 fail to make it to their second season.

15:45.640 --> 15:47.720
 It's just super hard to do so.

15:47.720 --> 15:52.720
 And so that that's just a tough business.

15:53.800 --> 15:58.800
 And there's been a lot of experimentation around

15:59.200 --> 16:02.640
 what is the right type of social companion?

16:02.640 --> 16:05.840
 What is the right robot in the home

16:05.840 --> 16:10.840
 that is doing something other than tasks people do every week

16:14.560 --> 16:16.400
 that they'd rather not do?

16:16.400 --> 16:20.880
 And I'm not sure we've got it all figured out right.

16:20.880 --> 16:22.960
 And so that you get brilliant roboticists

16:22.960 --> 16:25.680
 with super interesting robots

16:25.680 --> 16:29.800
 that ultimately don't quite have

16:29.800 --> 16:34.040
 that magical user experience and thus the,

16:36.480 --> 16:40.560
 that value benefit equation remains ambiguous.

16:40.560 --> 16:43.960
 So you as somebody who dreams of robots,

16:43.960 --> 16:45.720
 you know, changing the world,

16:45.720 --> 16:46.880
 what's your estimate?

16:47.840 --> 16:52.840
 Why, how big is the space of applications

16:53.200 --> 16:55.800
 that fit the criteria that you just described

16:55.800 --> 16:58.080
 where you can really demonstrate

16:58.080 --> 17:00.520
 an obvious significant value

17:00.520 --> 17:04.620
 over the alternative non robotic solution?

17:05.720 --> 17:08.680
 Well, I think that we're just about none of the way

17:10.000 --> 17:13.360
 to achieving the potential of robotics at home.

17:13.360 --> 17:18.360
 But we have to do it in a really eyes wide open,

17:20.400 --> 17:21.920
 honest fashion.

17:21.920 --> 17:25.400
 And so another way to put that is the potential is infinite

17:25.400 --> 17:27.040
 because we did take a few steps

17:27.040 --> 17:29.640
 but you're saying those steps are just very initial steps.

17:29.640 --> 17:32.560
 So the Roomba is a hugely successful product

17:32.560 --> 17:34.440
 but you're saying that's just the very, very beginning.

17:34.440 --> 17:36.520
 That's just the very, very beginning.

17:36.520 --> 17:37.960
 It's the foot in the door.

17:37.960 --> 17:40.840
 And, you know, I think I was lucky

17:40.840 --> 17:45.120
 that in the early days of robotics,

17:45.120 --> 17:48.360
 people would ask me, when are you gonna clean my floor?

17:48.360 --> 17:52.240
 It was something that I grew up saying,

17:53.560 --> 17:54.800
 I got all these really good ideas

17:54.800 --> 17:58.080
 but everyone seems to want their floor clean.

17:58.080 --> 18:02.480
 And so maybe we should do that.

18:02.480 --> 18:03.320
 Yeah, your good ideas.

18:03.320 --> 18:05.840
 Earn the right to do the next thing after that.

18:05.840 --> 18:10.160
 So the good ideas have to match with the desire of the people

18:10.160 --> 18:13.280
 and then the actual cost has to like the business,

18:13.280 --> 18:16.640
 the financial aspect has to all match together.

18:16.640 --> 18:20.160
 Yeah, during our partnership

18:20.160 --> 18:22.200
 back a number of years ago with Johnson Wax,

18:22.200 --> 18:27.200
 they would explain to me that they would go into homes

18:29.480 --> 18:32.520
 and just watch how people lived

18:32.520 --> 18:35.560
 and try to figure out what were they doing

18:35.560 --> 18:39.960
 that they really didn't really like to do

18:39.960 --> 18:42.440
 but they had to do it frequently enough

18:42.440 --> 18:47.440
 that it was top of mind and understood as a burden.

18:51.720 --> 18:53.000
 Hey, let's make a product

18:54.480 --> 18:58.480
 or come up with a solution to make that pain point

18:58.480 --> 19:02.040
 less challenging.

19:02.040 --> 19:07.040
 And sometimes we do certain burdens so often as a society

19:07.400 --> 19:09.400
 that we actually don't even realize,

19:09.400 --> 19:10.680
 like it's actually hard to see

19:10.680 --> 19:13.200
 that that burden is something that could be removed.

19:13.200 --> 19:15.760
 So it does require just going into the home

19:15.760 --> 19:19.560
 and staring at, wait, how do I actually live life?

19:19.560 --> 19:21.080
 What are the pain points?

19:21.080 --> 19:26.080
 Yeah, and it getting those insights is a lot harder

19:26.400 --> 19:29.360
 than it would seem it should be in retrospect.

19:29.360 --> 19:33.120
 So how hard on that point,

19:33.120 --> 19:37.480
 I mean, one of the big challenges of robotics

19:37.480 --> 19:40.240
 is driving the cost to something,

19:40.240 --> 19:42.240
 driving the cost down to something

19:42.240 --> 19:45.680
 that consumers, people would afford.

19:45.680 --> 19:48.880
 So people would be less likely to buy a Roomba

19:48.880 --> 19:52.280
 if it costs $500,000, right?

19:52.280 --> 19:55.840
 Which is probably sort of what a Roomba would cost

19:55.840 --> 19:58.040
 several decades ago.

19:58.040 --> 20:02.200
 So how do you drive, which I mentioned is very difficult.

20:02.200 --> 20:04.200
 How do you drive the cost of a Roomba

20:04.200 --> 20:07.920
 or a robot down such that people would want to buy it?

20:07.920 --> 20:09.720
 When I started building robots,

20:09.720 --> 20:12.240
 the cost of the robot had a lot to do

20:12.240 --> 20:15.480
 with the amount of time it took to build it.

20:15.480 --> 20:18.400
 And so that we would build our robots out of aluminum,

20:18.400 --> 20:21.160
 I would go spend my time in the machine shop

20:21.160 --> 20:22.840
 on the milling machine,

20:23.840 --> 20:28.000
 cutting out the parts and so forth.

20:28.000 --> 20:29.720
 And then when we got into the toy industry,

20:29.720 --> 20:34.520
 I realized that if we were building at scale,

20:34.520 --> 20:36.000
 I could determine the cost of the Roomba

20:36.000 --> 20:38.920
 instead of adding up all the hours to mill out the parts,

20:38.920 --> 20:40.360
 but by weighing it.

20:42.080 --> 20:44.200
 And that's liberating.

20:44.200 --> 20:49.200
 You can say, wow, the world has just changed

20:49.560 --> 20:53.160
 as I think about construction in a different way.

20:53.160 --> 20:56.880
 The 3D CAD tools that are available to us today,

20:56.880 --> 21:01.680
 the operating at scale where I can do tooling

21:01.680 --> 21:06.680
 and injection mold, an arbitrarily complicated part,

21:07.080 --> 21:09.800
 and the cost is going to be basically

21:09.800 --> 21:12.560
 the weight of the plastic in that part

21:13.920 --> 21:16.360
 is incredibly exciting and liberating

21:16.360 --> 21:18.560
 and opens up all sorts of opportunities.

21:18.560 --> 21:23.560
 And for the sensing part of it, where we are today

21:23.560 --> 21:28.560
 is instead of trying to build skin,

21:29.280 --> 21:31.440
 which is like really hard for a long time.

21:31.440 --> 21:36.440
 I spent creating strategies and ideas

21:38.240 --> 21:42.720
 around how could we duplicate the skin on the human body

21:42.720 --> 21:45.360
 because it's such an amazing sensor.

21:47.880 --> 21:49.600
 Instead of going down that path,

21:49.600 --> 21:53.000
 why don't we focus on vision?

21:54.000 --> 21:59.000
 And how many of the problems that face a robot

22:00.080 --> 22:04.880
 trying to do real work could be solved

22:04.880 --> 22:08.480
 with a cheap camera and a big ass computer?

22:09.480 --> 22:12.440
 And Moore's Law continues to work.

22:12.440 --> 22:16.520
 The cell phone industry, the mobile industry

22:16.520 --> 22:18.800
 is giving us better and better tools

22:18.800 --> 22:21.120
 that can run on these embedded computers.

22:21.120 --> 22:26.120
 And I think we passed an important moment,

22:26.600 --> 22:31.600
 maybe two years ago, where you could put

22:32.760 --> 22:37.520
 machine vision capable processors on robots

22:37.520 --> 22:39.640
 at consumer price points.

22:39.640 --> 22:43.040
 And I was waiting for it to happen.

22:43.040 --> 22:46.600
 We avoided putting lasers on our robots

22:46.600 --> 22:51.600
 to do navigation and instead spent years researching

22:51.840 --> 22:54.640
 how to do vision based navigation

22:54.640 --> 22:58.440
 because you could just see it

22:58.440 --> 23:01.640
 where these technology trends were going

23:01.640 --> 23:05.880
 and between injection molded plastic

23:05.880 --> 23:08.040
 and a camera with a computer

23:08.040 --> 23:10.840
 capable of running machine learning

23:10.840 --> 23:12.560
 and visual object recognition,

23:12.560 --> 23:15.560
 I could build an incredibly affordable,

23:15.560 --> 23:18.760
 incredibly capable robot

23:18.760 --> 23:21.240
 and that's gonna be the future.

23:21.240 --> 23:23.440
 So you know, on that point with a small tangent

23:23.440 --> 23:25.000
 but I think an important one,

23:25.000 --> 23:27.600
 another industry in which I would say

23:27.600 --> 23:31.880
 the only other industry in which there is automation

23:31.880 --> 23:34.840
 actually touching people's lives today

23:34.840 --> 23:36.520
 is autonomous vehicles.

23:37.640 --> 23:42.360
 What the vision is described of using computer vision

23:42.360 --> 23:44.520
 and using cheap camera sensors,

23:44.520 --> 23:48.320
 there's a debate on that of LiDAR versus computer vision

23:48.320 --> 23:53.320
 and sort of the Elon Musk famously said

23:53.320 --> 23:58.320
 that LiDAR is a crutch that really in the long term,

23:58.440 --> 24:00.880
 camera only is the right solution

24:00.880 --> 24:03.520
 which echoes some of the ideas you're expressing.

24:03.520 --> 24:05.120
 Of course, the domain

24:05.120 --> 24:07.720
 in terms of its safety criticality is different.

24:07.720 --> 24:10.720
 But what do you think about that approach

24:10.720 --> 24:13.480
 in the autonomous vehicle space?

24:13.480 --> 24:15.200
 And in general, do you see a connection

24:15.200 --> 24:18.560
 between the incredible real world challenges

24:18.560 --> 24:20.800
 you have to solve in the home with Roomba

24:20.800 --> 24:23.000
 and I saw a demonstration of some of them,

24:23.000 --> 24:27.920
 corner cases literally and autonomous vehicles.

24:27.920 --> 24:31.720
 So there's absolutely a tremendous overlap

24:31.720 --> 24:35.520
 between both the problems, you know,

24:35.520 --> 24:38.680
 a robot vacuum and autonomous vehicle are trying to solve

24:38.680 --> 24:41.880
 and the tools and the types of sensors

24:41.880 --> 24:46.720
 that are being applied in the pursuit of the solutions.

24:48.040 --> 24:53.040
 In my world, my environment is actually much harder

24:54.680 --> 24:57.320
 than the environment in automobile travels.

24:57.320 --> 25:02.320
 We don't have roads, we have t shirts, we have steps,

25:02.720 --> 25:07.120
 we have a near infinite number of patterns and colors

25:07.120 --> 25:10.200
 and surface textures on the floor.

25:10.200 --> 25:12.560
 Especially from a visual perspective.

25:12.560 --> 25:14.760
 Yeah, visually it's really tough.

25:14.760 --> 25:18.880
 Is an infinitely variable.

25:18.880 --> 25:22.560
 On the other hand, safety is way easier on the inside.

25:22.560 --> 25:26.440
 My robots, they're not very heavy,

25:26.440 --> 25:28.280
 they're not very fast.

25:28.280 --> 25:31.480
 If they bump into your foot, you think it's funny.

25:32.720 --> 25:36.920
 And, you know, and autonomous vehicles

25:36.920 --> 25:39.400
 kind of have the inverse problem.

25:39.400 --> 25:44.400
 And so that for me saying vision is the future,

25:45.920 --> 25:47.800
 I can say that without reservation.

25:49.480 --> 25:54.480
 For autonomous vehicles, I think I believe what Elon's saying

25:55.360 --> 25:59.040
 about the future is ultimately gonna be vision.

25:59.040 --> 26:01.000
 Maybe if we put a cheap lighter on there

26:01.000 --> 26:03.280
 as a backup sensor, it might not be the worst idea

26:03.280 --> 26:04.120
 in the world.

26:04.120 --> 26:04.960
 So the stakes are much higher.

26:04.960 --> 26:05.800
 The stakes are much higher.

26:05.800 --> 26:08.200
 You have to be much more careful thinking through

26:08.200 --> 26:10.720
 how far away that future is.

26:10.720 --> 26:14.240
 Right, but I think that the primary

26:16.080 --> 26:19.320
 environmental understanding sensor

26:19.320 --> 26:21.760
 is going to be a visual system.

26:21.760 --> 26:23.000
 Visual system.

26:23.000 --> 26:25.560
 So on that point, well, let me ask,

26:25.560 --> 26:29.440
 do you hope there's an iRobot robot in every home

26:29.440 --> 26:30.920
 in the world one day?

26:31.880 --> 26:34.880
 I expect there to be at least one iRobot robot

26:34.880 --> 26:36.440
 in every home.

26:36.440 --> 26:41.160
 You know, we've sold 25 million robots.

26:41.160 --> 26:44.600
 So we're in about 10% of US homes,

26:44.600 --> 26:45.760
 which is a great start.

26:47.120 --> 26:51.080
 But I think that when we think about the numbers

26:51.080 --> 26:55.840
 of things that robots can do, you know,

26:55.840 --> 26:58.560
 today I can vacuum your floor, mop your floor,

26:58.560 --> 27:01.240
 cut your lawn or soon we'll be able to cut your lawn.

27:01.240 --> 27:06.240
 But there are more things that we could do in the home.

27:06.720 --> 27:11.520
 And I hope that we continue using the techniques

27:11.520 --> 27:14.480
 I described around exploiting computer vision

27:14.480 --> 27:18.680
 and low cost manufacturing that we'll be able

27:18.680 --> 27:22.680
 to create these solutions at affordable price points.

27:22.680 --> 27:25.600
 So let me ask, on that point of a robot in every home,

27:25.600 --> 27:26.880
 that's my dream as well.

27:26.880 --> 27:28.400
 I'd love to see that.

27:28.400 --> 27:31.880
 I think the possibilities there are indeed

27:31.880 --> 27:34.560
 infinite positive possibilities.

27:34.560 --> 27:39.560
 But in our current culture, no thanks to science fiction

27:39.800 --> 27:44.720
 and so on, there's a serious kind of hesitation

27:44.720 --> 27:47.120
 and anxiety, concern about robots

27:47.120 --> 27:50.080
 and also a concern about privacy.

27:51.480 --> 27:54.040
 And it's a fascinating question to me

27:54.040 --> 27:59.040
 why that concern is amongst a certain group of people

27:59.600 --> 28:02.880
 is as intense as it is.

28:02.880 --> 28:04.280
 So you have to think about it

28:04.280 --> 28:05.520
 because it's a serious concern,

28:05.520 --> 28:08.040
 but I wonder how you address it best.

28:08.040 --> 28:09.840
 So from a perspective of a vision sensor,

28:09.840 --> 28:14.200
 so robots that move about the home and sense the world,

28:14.200 --> 28:19.200
 how do you alleviate people's privacy concerns?

28:19.720 --> 28:22.880
 How do you make sure that they can trust iRobot

28:22.880 --> 28:25.360
 and the robots that they share their home with?

28:26.640 --> 28:28.160
 I think that's a great question.

28:28.160 --> 28:33.160
 And we've really leaned way forward on this

28:33.760 --> 28:38.760
 because given our vision as to the role the company

28:38.880 --> 28:40.720
 intends to play in the home,

28:43.000 --> 28:45.480
 really for us, make or break is,

28:45.480 --> 28:50.480
 can our approach be trusted to protecting the data

28:50.480 --> 28:53.560
 and the privacy of the people who have our robots?

28:53.560 --> 28:56.920
 And so we've gone out publicly

28:56.920 --> 29:00.440
 with a privacy manifesto stating we'll never sell your data.

29:00.440 --> 29:05.440
 We've adopted GDPR, not just where GDPR is required,

29:05.520 --> 29:10.520
 but globally, we have ensured that images

29:16.640 --> 29:18.080
 don't leave the robot.

29:18.080 --> 29:22.120
 So processing data from the visual sensors

29:22.120 --> 29:23.680
 happens locally on the robot

29:23.680 --> 29:28.680
 and only semantic knowledge of the home

29:29.520 --> 29:32.720
 with the consumer's consent is sent up.

29:32.720 --> 29:34.480
 We show you what we know

29:34.480 --> 29:39.480
 and are trying to go use data as an enabler

29:41.560 --> 29:44.120
 for the performance of the robots

29:44.120 --> 29:49.120
 with the informed consent and understanding

29:50.040 --> 29:52.440
 of the people who own those robots.

29:52.440 --> 29:56.880
 And we take it very seriously.

29:56.880 --> 30:01.880
 And ultimately, we think that by showing a customer

30:01.880 --> 30:06.880
 that if you let us build a semantic map of your home

30:07.360 --> 30:09.000
 and know where the rooms are,

30:09.000 --> 30:11.760
 well, then you can say clean the kitchen.

30:11.760 --> 30:13.720
 If you don't want the robot to do that,

30:13.720 --> 30:14.560
 don't make the map,

30:14.560 --> 30:17.000
 it'll do its best job cleaning your home,

30:17.000 --> 30:18.640
 but it won't be able to do that.

30:18.640 --> 30:20.280
 And if you ever want us to forget

30:20.280 --> 30:22.080
 that we know that it's your kitchen,

30:22.080 --> 30:26.680
 you can have confidence that we will do that for you.

30:26.680 --> 30:31.680
 So we're trying to go and be a sort of a data 2.0

30:34.520 --> 30:37.680
 perspective company where we treat the data

30:37.680 --> 30:40.800
 that the robots have of the consumer's home

30:40.800 --> 30:43.200
 as if it were the consumer's data

30:43.200 --> 30:47.360
 and that they have rights to it.

30:47.360 --> 30:50.960
 So we think by being the good guys on this front,

30:50.960 --> 30:53.840
 we can build the trust and thus be entrusted

30:55.120 --> 31:00.120
 to enable robots to do more things that are thoughtful.

31:00.200 --> 31:04.520
 You think people's worries will diminish over time?

31:04.520 --> 31:06.840
 As a society, broadly speaking,

31:06.840 --> 31:09.320
 do you think you can win over trust,

31:09.320 --> 31:10.640
 not just for the company,

31:10.640 --> 31:14.880
 but just the comfort of people have with AI in their home

31:14.880 --> 31:17.040
 enriching their lives in some way?

31:17.040 --> 31:19.560
 I think we're an interesting place today

31:19.560 --> 31:22.400
 where it's less about winning them over

31:22.400 --> 31:26.240
 and more about finding a way to talk about privacy

31:26.240 --> 31:28.840
 in a way that more people can understand.

31:28.840 --> 31:30.920
 I would tell you that today,

31:30.920 --> 31:33.320
 when there's a privacy breach,

31:33.320 --> 31:37.040
 people get very upset and then go to the store

31:37.040 --> 31:38.320
 and buy the cheapest thing,

31:38.320 --> 31:41.000
 paying no attention to whether or not the products

31:41.000 --> 31:44.640
 that they're buying honor privacy standards or not.

31:44.640 --> 31:48.600
 In fact, if I put on the package of my Roomba,

31:50.080 --> 31:53.640
 the privacy commitments that we have,

31:53.640 --> 31:58.640
 I would sell less than I would if I did nothing at all

31:58.720 --> 32:00.400
 and that needs to change.

32:00.400 --> 32:02.880
 So it's not a question about earning trust.

32:02.880 --> 32:05.000
 I think that's necessary but not sufficient.

32:05.000 --> 32:08.440
 We need to figure out how to have a comfortable set

32:08.440 --> 32:13.440
 of what is the grade A meat standard applied to privacy

32:14.200 --> 32:18.400
 that customers can trust and understand

32:18.400 --> 32:20.480
 and then use in the buying decisions.

32:23.040 --> 32:25.520
 That will reward companies for good behavior

32:25.520 --> 32:29.880
 and that will ultimately be how this moves forward.

32:29.880 --> 32:32.680
 And maybe be part of the conversation

32:32.680 --> 32:34.800
 between regular people about what it means,

32:34.800 --> 32:36.280
 what privacy means.

32:36.280 --> 32:38.400
 If you have some standards, you can say,

32:38.400 --> 32:41.080
 you can start talking about who's following them,

32:41.080 --> 32:42.680
 who's not, have more.

32:42.680 --> 32:45.440
 Because most people are actually quite clueless

32:45.440 --> 32:47.320
 about all aspects of artificial intelligence

32:47.320 --> 32:48.400
 or data collection and so on.

32:48.400 --> 32:49.920
 It would be nice to change that

32:49.920 --> 32:52.760
 for people to understand the good that AI can do

32:52.760 --> 32:56.520
 and it's not some system that's trying to steal

32:56.520 --> 32:58.760
 all the most sensitive data.

32:58.760 --> 33:02.640
 Do you think, do you dream of a Roomba

33:02.640 --> 33:05.240
 with human level intelligence one day?

33:05.240 --> 33:10.240
 So you've mentioned a very successful localization

33:10.520 --> 33:11.880
 and mapping of the environment,

33:11.880 --> 33:14.360
 being able to do some basic communication

33:14.360 --> 33:16.560
 to say go clean the kitchen.

33:16.560 --> 33:21.360
 Do you see in your maybe more bored moments,

33:22.880 --> 33:24.840
 once you get the beer,

33:24.840 --> 33:27.000
 just sit back with that beer

33:27.000 --> 33:30.800
 and have a chat on a Friday night with the Roomba

33:30.800 --> 33:32.840
 about how your day went.

33:34.120 --> 33:37.560
 So through your latter question, absolutely.

33:38.640 --> 33:40.720
 To your former question as to whether Roomba

33:40.720 --> 33:43.680
 can have human level intelligence, not in my lifetime.

33:45.200 --> 33:48.440
 You can have you, you can have a great conversation,

33:49.680 --> 33:51.960
 a meaningful conversation with a robot

33:53.920 --> 33:55.400
 without it having anything

33:55.400 --> 33:57.760
 that resembles human level intelligence.

33:57.760 --> 34:02.760
 And I think that as long as you realize

34:02.800 --> 34:05.280
 that conversation is not about the robot

34:06.360 --> 34:08.600
 and making the robot feel good.

34:08.600 --> 34:13.600
 That conversation is about you learning interesting things

34:14.880 --> 34:18.400
 that make you feel like the conversation

34:18.400 --> 34:21.240
 that you had with the robot is

34:23.800 --> 34:27.280
 a pretty awesome way of learning something.

34:27.280 --> 34:30.800
 And it could be about what kind of day your pet had.

34:30.800 --> 34:35.040
 It could be about, how can I make my home

34:35.040 --> 34:36.160
 more energy efficient?

34:36.160 --> 34:39.600
 It could be about, if I'm thinking about

34:39.600 --> 34:41.720
 climbing Mount Everest, what should I know?

34:44.320 --> 34:46.480
 And that's a very doable thing.

34:48.600 --> 34:51.480
 But if I think that that conversation

34:51.480 --> 34:53.640
 I'm gonna have with the robot is,

34:53.640 --> 34:56.760
 I'm gonna be rewarded by making the robot happy.

34:56.760 --> 34:58.720
 But I could have just put a button on the robot

34:58.720 --> 35:00.280
 that you could push and the robot would smile

35:00.280 --> 35:02.120
 and that sort of thing.

35:02.120 --> 35:04.160
 So I think you need to think about the question

35:04.160 --> 35:06.680
 in the right way.

35:06.680 --> 35:11.520
 And robots can be awesomely effective

35:11.520 --> 35:14.400
 at helping people feel less isolated,

35:14.400 --> 35:17.520
 learn more about the home that they live in

35:17.520 --> 35:21.920
 and fill some of those lonely gaps

35:21.920 --> 35:23.760
 that we wish we were engaged

35:23.760 --> 35:25.640
 learning cool stuff about our world.

35:25.640 --> 35:30.640
 Last question, if you could hang out for a day

35:30.640 --> 35:34.640
 with a robot from science fiction, movies, books

35:34.640 --> 35:39.640
 and safely pick, safely pick its brain for that day,

35:39.640 --> 35:41.640
 who would you pick?

35:41.640 --> 35:42.640
 Data.

35:42.640 --> 35:43.640
 Data.

35:43.640 --> 35:44.640
 From Star Trek.

35:44.640 --> 35:48.640
 I think that data is really smart.

35:48.640 --> 35:52.640
 Data's been through a lot trying to go and save the galaxy

35:52.640 --> 35:57.640
 and I'm really interested actually in emotion and robotics.

35:58.640 --> 36:00.640
 And I think you'd have a lot to say about that

36:00.640 --> 36:05.640
 because I believe actually that emotion plays

36:05.640 --> 36:10.640
 an incredibly useful role in doing reasonable things

36:11.640 --> 36:14.640
 in situations where we have imperfect understanding

36:14.640 --> 36:15.640
 of what's going on.

36:15.640 --> 36:18.640
 In social situations when there's imperfect information.

36:18.640 --> 36:23.640
 In social situations also in competitive

36:23.640 --> 36:25.640
 or dangerous situations,

36:26.640 --> 36:30.640
 that we have emotion for a reason.

36:30.640 --> 36:35.640
 And so that ultimately, my theory is that as robots

36:35.640 --> 36:36.640
 get smarter and smarter,

36:36.640 --> 36:38.640
 they're actually going to get more emotional

36:38.640 --> 36:46.640
 because you can't actually survive on pure logic.

36:46.640 --> 36:51.640
 Because only a very tiny fraction of the situations

36:51.640 --> 36:55.640
 we find ourselves in can be resolved reasonably with logic.

36:55.640 --> 36:57.640
 And so I think data would have a lot to say about that

36:57.640 --> 36:59.640
 and so I could find out whether he agrees.

36:59.640 --> 37:02.640
 What, if you could ask data one question

37:02.640 --> 37:05.640
 and you would get a deep, honest answer to,

37:05.640 --> 37:06.640
 what would you ask?

37:06.640 --> 37:08.640
 What's Captain Picard really like?

37:08.640 --> 37:12.640
 Okay, I think that's the perfect way to end the call

37:12.640 --> 37:14.640
 and thank you so much for talking today.

37:14.640 --> 37:16.640
 I really appreciate it.

37:16.640 --> 37:45.640
 My pleasure.