Colin Angle: iRobot
AI 与机器学习音乐与艺术技术与编程商业与创业心理与人性
📋 章节目录
暂无章节信息
🔑 关键词
robotrobotshomeroombadataroboticsgoingirobotconversationvisiondoncostprivacyautonomouscompanyrolefutureenvironmentincrediblehard
💬 精彩语录
暂无语录
🎙️ 完整对话（636 条）
Lex Fridman (00:00.000)
The following is a conversation with Colin Angle.
以下是与科林·安格的对话。
Lex Fridman (00:02.800)
He's the CEO and co founder of iRobot,
他是 iRobot 的首席执行官兼联合创始人，
Lex Fridman (00:05.840)
a robotics company that for 29 years
一家拥有 29 年历史的机器人公司
Lex Fridman (00:08.180)
has been creating robots that operate successfully
一直在创造能够成功运行的机器人
Lex Fridman (00:11.180)
in the real world.
在现实世界中。
Colin Angle (00:12.560)
Not as a demo or on a scale of dozens,
不是作为演示或数十个规模，
Lex Fridman (00:15.600)
but on a scale of thousands and millions.
但规模达到了数以千计的规模。
Colin Angle (00:18.860)
As of this year, iRobot has sold more than
截至今年，iRobot 的销量已超过
Lex Fridman (00:21.900)
25 million robots to consumers,
2500万个机器人给消费者，
Colin Angle (00:25.700)
including the Roomba vacuum cleaning robot,
包括 Roomba 真空清洁机器人，
Lex Fridman (00:28.160)
the Bravo floor mopping robot,
Bravo 拖地机器人，
Lex Fridman (00:29.960)
and soon the Terra lawn mowing robot.
很快还有 Terra 割草机器人。
Lex Fridman (00:33.960)
29 million robots successfully operating autonomously
2900万台机器人成功自主运行
Colin Angle (00:37.640)
in real people's homes,
在真实的人们家里，
Lex Fridman (00:39.640)
to me is an incredible accomplishment
对我来说是一项令人难以置信的成就
Colin Angle (00:42.060)
of science, engineering, logistics,
科学、工程、物流、
Lex Fridman (00:45.060)
and all kinds of general entrepreneurial innovation.
以及各类一般创业创新。
Colin Angle (00:48.760)
Most robotics companies fail.
大多数机器人公司都失败了。
Lex Fridman (00:51.320)
iRobot has survived and succeeded for 29 years.
iRobot 已经生存并取得了 29 年的成功。
Colin Angle (00:56.320)
I spent all day at iRobot,
我在 iRobot 待了一整天，
Lex Fridman (00:58.860)
including a long tour and conversation with Colin
Colin Angle (01:01.440)
about the history of iRobot,
Lex Fridman (01:03.480)
and then sat down for this podcast conversation
Colin Angle (01:06.740)
that would have been much longer
Lex Fridman (01:08.560)
if I didn't spend all day learning about
Lex Fridman (01:10.780)
and playing with the various robots
Lex Fridman (01:12.340)
and the company's history.
Colin Angle (01:13.980)
I'll release the video of the tour separately.
Lex Fridman (01:17.520)
Colin, iRobot, its founding team, its current team,
Lex Fridman (01:21.900)
and its mission has been and continues to be
Lex Fridman (01:24.660)
an inspiration to me and thousands of engineers
Colin Angle (01:27.620)
who are working hard to create AI systems
Lex Fridman (01:30.260)
that help real people.
Colin Angle (01:33.020)
This is the Artificial Intelligence Podcast.
Lex Fridman (01:35.660)
If you enjoy it, subscribe on YouTube,
Colin Angle (01:38.020)
give it five stars on iTunes,
Lex Fridman (01:39.820)
support it on Patreon,
Colin Angle (01:41.300)
or simply connect with me on Twitter
Lex Fridman (01:43.300)
at Lex Friedman, spelled F R I D M A N.
Lex Fridman (01:47.140)
And now, here's my conversation with Colin Angle.
Lex Fridman (01:51.120)
In his 1942 short story, Runaround,
Colin Angle (01:55.200)
from his iRobot collection, Asimov proposed
Lex Fridman (02:00.000)
the three laws of robotics in order,
Colin Angle (02:02.880)
don't harm humans, obey orders, protect yourself.
Lex Fridman (02:06.840)
So two questions.
Lex Fridman (02:07.720)
First, does the Roomba follow these three laws?
Lex Fridman (02:11.680)
And also, more seriously,
Lex Fridman (02:14.800)
what role do you hope to see robots take
Lex Fridman (02:17.160)
in modern society and in the future world?
Lex Fridman (02:21.320)
So the three laws are very thought provoking
Lex Fridman (02:25.760)
and require such a profound understanding
Colin Angle (02:31.400)
of the world a robot lives in,
Lex Fridman (02:36.280)
the ramifications of its action and its own sense of self
Colin Angle (02:40.040)
that it's not a relevant bar,
Lex Fridman (02:45.040)
at least it won't be a relevant bar for decades to come.
Lex Fridman (02:50.120)
And so if Roomba follows the three laws,
Lex Fridman (02:54.560)
and I believe it does,
Colin Angle (02:58.000)
it is designed to help humans, not hurt them,
Lex Fridman (03:00.920)
it's designed to be inherently safe,
Lex Fridman (03:03.120)
and we designed it to last a long time.
Lex Fridman (03:07.200)
It's not through any AI or intent on the robot's part.
Colin Angle (03:11.600)
It's because following the three laws
Lex Fridman (03:14.960)
is aligned with being a good robot product.
Lex Fridman (03:19.620)
So I guess it does,
Lex Fridman (03:23.120)
but not by explicit design.
Lex Fridman (03:27.240)
So then the bigger picture,
Lex Fridman (03:28.800)
what role do you hope to see robotics, robots take
Lex Fridman (03:33.040)
in what's currently mostly a world of humans?
Lex Fridman (03:37.360)
We need robots to help us continue
Colin Angle (03:42.360)
to improve our standard of living.
Lex Fridman (03:46.120)
We need robots because the average age
Colin Angle (03:51.320)
of humanity is increasing very quickly,
Lex Fridman (03:55.040)
and simply the number of people young enough
Lex Fridman (03:59.760)
and spry enough to care
Lex Fridman (04:01.600)
for the elder growing demographic is inadequate.
Lex Fridman (04:08.800)
And so what is the role of robots?
Lex Fridman (04:11.400)
Today, the role is to make our lives a little easier,
Colin Angle (04:14.880)
a little cleaner, maybe a little healthier.
Lex Fridman (04:18.480)
But in time, robots are going to be the difference
Colin Angle (04:22.120)
between real gut wrenching declines
Lex Fridman (04:25.680)
in our ability to live independently
Lex Fridman (04:28.080)
and maintain our standard of living,
Lex Fridman (04:30.240)
and a future that is the bright one
Colin Angle (04:34.880)
where we have more control over our lives,
Lex Fridman (04:37.480)
can spend more of our time focused
Colin Angle (04:40.560)
on activities we choose.
Lex Fridman (04:44.560)
And I'm so honored and excited
Colin Angle (04:47.920)
to be playing a role in that journey.
Lex Fridman (04:50.400)
So you've given me a tour.
Colin Angle (04:51.720)
It showed me some of the long histories,
Lex Fridman (04:53.960)
now 29 years that iRobot has been at it,
Colin Angle (04:57.120)
creating some incredible robots.
Lex Fridman (04:59.200)
You showed me Pacbot.
Colin Angle (05:01.160)
You showed me a bunch of other stuff that led up to Roomba,
Lex Fridman (05:04.480)
that led to Braava and Terra.
Lex Fridman (05:08.640)
So let's skip that incredible history
Lex Fridman (05:14.080)
in the interest of time,
Colin Angle (05:15.040)
cause we already talked about it.
Lex Fridman (05:16.120)
I'll show this incredible footage.
Colin Angle (05:18.040)
You mentioned elderly and robotics in society.
Lex Fridman (05:22.640)
I think the home is a fascinating place for robots to be.
Lex Fridman (05:26.280)
So where do you see robots in the home?
Lex Fridman (05:29.800)
Currently, I would say, once again,
Colin Angle (05:31.640)
probably most homes in the world don't have a robot.
Lex Fridman (05:34.520)
So how do you see that changing?
Lex Fridman (05:36.160)
What do you think is the big initial value add
Lex Fridman (05:39.840)
that robots can do?
Lex Fridman (05:41.920)
So iRobot has sort of, over the years,
Lex Fridman (05:44.960)
narrowed in on the home, the consumer's home,
Colin Angle (05:49.360)
as the place where we want to innovate
Lex Fridman (05:53.120)
and deliver tools that will help a home
Colin Angle (05:59.360)
be a more automatically maintained place,
Lex Fridman (06:04.240)
a healthier place, a safer place,
Lex Fridman (06:06.800)
and perhaps even a more efficient place to be.
Lex Fridman (06:11.480)
And today, we vacuum, we mop,
Colin Angle (06:15.040)
soon we'll be mowing your lawn.
Lex Fridman (06:16.960)
But where things are going is,
Colin Angle (06:22.720)
when do we get to the point where the home,
Lex Fridman (06:27.080)
not just the robots that live in your home,
Lex Fridman (06:29.120)
but the home itself becomes part of a system
Lex Fridman (06:32.160)
that maintains itself and plays an active role
Colin Angle (06:35.960)
in caring for and helping the people live in that home.
Lex Fridman (06:40.760)
And I see everything that we're doing
Colin Angle (06:43.200)
as steps along the path toward that future.
Lex Fridman (06:46.160)
So what are the steps?
Lex Fridman (06:47.720)
So if we can summarize some of the history of Roomba,
Lex Fridman (06:53.280)
you've mentioned, and maybe you can elaborate on it,
Lex Fridman (06:55.520)
but you mentioned that the early days
Lex Fridman (06:57.240)
were really taking a robot from something that works
Colin Angle (07:02.320)
either in the lab or something that works in the field
Lex Fridman (07:04.880)
that helps soldiers do the difficult work they do
Colin Angle (07:10.200)
to actually be in the hands of consumers
Lex Fridman (07:12.640)
and tens of thousands, hundreds of thousands of robots
Colin Angle (07:15.640)
that don't break down over how much people love them
Lex Fridman (07:18.480)
over months of very extensive use.
Lex Fridman (07:21.440)
So that was the big first step.
Lex Fridman (07:22.840)
And then the second big step was the ability
Colin Angle (07:26.000)
to sense the environment, to build a map, to localize,
Lex Fridman (07:29.920)
to be able to build a picture of the home
Colin Angle (07:32.560)
that the human can then attach labels to
Lex Fridman (07:34.600)
in terms of giving some semantic knowledge
Colin Angle (07:38.400)
to the robot about its environment.
Lex Fridman (07:40.880)
Okay, so that's like a huge, two big, huge steps.
Colin Angle (07:46.320)
Maybe you can comment on them,
Lex Fridman (07:47.520)
but also what is the next step
Lex Fridman (07:51.040)
of making a robot part of the home?
Lex Fridman (07:54.720)
Sure, so the goal is to make a home
Colin Angle (07:57.840)
that takes care of itself,
Lex Fridman (08:01.280)
takes care of the people in the home,
Lex Fridman (08:03.680)
and gives the user an experience of just living their life
Lex Fridman (08:07.840)
and the home is somehow doing the right thing,
Colin Angle (08:10.840)
turning on and off lights when you leave,
Lex Fridman (08:14.120)
cleaning up the environment.
Lex Fridman (08:17.240)
And we went from robots that were great in the lab,
Lex Fridman (08:24.920)
but were both too expensive
Lex Fridman (08:27.200)
and not sufficiently capable to ever do an acceptable job
Lex Fridman (08:32.880)
of anything other than being a toy or a curio in your home
Colin Angle (08:37.280)
to something that was both affordable
Lex Fridman (08:42.080)
and sufficiently effective to drive,
Colin Angle (08:45.720)
be above threshold and drive purchase intent.
Lex Fridman (08:50.600)
Now we've disrupted the entire vacuuming industry.
Colin Angle (08:55.440)
The number one selling vacuums, for example, in the US
Lex Fridman (08:59.760)
are Roombas, so not robot vacuums, but vacuums,
Lex Fridman (09:02.840)
and that's really crazy and weird.
Lex Fridman (09:05.480)
We need to pause that. I mean, that's incredible.
Colin Angle (09:08.000)
That's incredible that a robot
Lex Fridman (09:10.440)
is the number one selling thing that does something.
Colin Angle (09:15.480)
Yep. Something as essential as vacuuming.
Lex Fridman (09:17.880)
Yep. Congratulations.
Colin Angle (09:20.000)
Thank you. It's still kind of fun to say,
Lex Fridman (09:22.360)
but just because this was a crazy idea
Colin Angle (09:26.520)
that just started, you know, in a room here,
Lex Fridman (09:30.880)
we're like, do you think we can do this?
Colin Angle (09:33.600)
So, hey, let's give it a try.
Lex Fridman (09:35.960)
But now the robots are starting to understand their environment.
Lex Fridman (09:42.760)
And if you think about the next step,
Lex Fridman (09:46.240)
there's two dimensions.
Colin Angle (09:48.640)
I've been working so hard since the beginning of iRobot
Lex Fridman (09:52.960)
to make robots are autonomous,
Colin Angle (09:55.000)
that, you know, they're smart enough
Lex Fridman (09:57.560)
and understand their task enough,
Colin Angle (09:59.400)
they can just go do it without human involvement.
Lex Fridman (10:04.080)
Now what I'm really excited and working on
Lex Fridman (10:07.360)
is how do I make them less autonomous?
Lex Fridman (10:10.560)
Meaning that the robot is supposed to be your partner,
Colin Angle (10:15.680)
not this automaton that just goes and does what a robot does.
Lex Fridman (10:20.120)
And so that if you tell it,
Colin Angle (10:23.440)
hey, I just dropped some flour by the fridge in the kitchen,
Lex Fridman (10:27.080)
can you deal with it?
Colin Angle (10:28.920)
Wouldn't it be awesome if the right thing just happened
Lex Fridman (10:32.720)
based on that utterance?
Lex Fridman (10:35.200)
And to some extent, that's less autonomous
Lex Fridman (10:37.840)
because it's actually listening to you,
Colin Angle (10:40.080)
understanding the context and intent of the sentence,
Lex Fridman (10:44.360)
mapping it against its understanding
Colin Angle (10:47.360)
of the home it lives in and knowing what to do.
Lex Fridman (10:52.640)
And so that's an area of research.
Colin Angle (10:56.320)
It's an area where we're starting to roll out features.
Lex Fridman (10:59.360)
You can now tell your robot to clean up the kitchen
Lex Fridman (11:02.840)
and it knows what the kitchen is and can do that.
Lex Fridman (11:05.840)
And that's sort of 1.0 of where we're going.
Colin Angle (11:10.400)
The other cool thing is that we're starting
Lex Fridman (11:13.000)
to know where stuff is.
Lex Fridman (11:14.600)
And why is that important?
Lex Fridman (11:15.960)
Well, robots are supposed to have arms, right?
Colin Angle (11:21.480)
Data had an arm, Rosie had an arm, Robbie the robot had an arm.
Lex Fridman (11:25.200)
I mean, robots are, you know, they are physical things
Colin Angle (11:27.680)
that move around in an environment
Lex Fridman (11:29.480)
and they're supposed to like do work.
Lex Fridman (11:31.200)
And if you think about it,
Lex Fridman (11:34.080)
if a robot doesn't know where anything is,
Lex Fridman (11:37.200)
why should it have an arm?
Lex Fridman (11:38.720)
But with this new dawn of home understanding
Colin Angle (11:44.320)
that we're starting to go enjoy,
Lex Fridman (11:47.680)
I know where the kitchen is.
Colin Angle (11:49.320)
I might in the future know where the refrigerator is.
Lex Fridman (11:52.000)
I might, if I had an arm, be able to find the handle,
Colin Angle (11:55.240)
open it and even get myself a beer.
Lex Fridman (11:58.440)
Obviously, that's one of the true dreams of robotics
Colin Angle (12:01.880)
is to have robots bringing us a beer
Lex Fridman (12:03.480)
while we watch television.
Colin Angle (12:05.240)
But, you know, I think that that new category of tasks
Lex Fridman (12:10.960)
where physical manipulation, robot arms,
Colin Angle (12:14.200)
is just a potpourri of new opportunity and excitement.
Lex Fridman (12:20.160)
And you see humans as a crucial part of that.
Lex Fridman (12:23.760)
So you kind of mentioned that.
Lex Fridman (12:26.240)
And I personally find that a really compelling idea.
Colin Angle (12:28.880)
I think full autonomy can only take us so far,
Lex Fridman (12:33.920)
especially in the home.
Lex Fridman (12:35.320)
So you see humans as helping the robot understand
Lex Fridman (12:38.880)
or give deeper meaning to the spatial information.
Colin Angle (12:43.560)
Right. It's a partnership.
Lex Fridman (12:46.760)
The robot is supposed to operate according to descriptors
Colin Angle (12:52.760)
that you would use to describe your own home.
Lex Fridman (12:57.040)
The robot is supposed to, in lieu of better direction,
Colin Angle (13:02.200)
kind of go about its routine,
Lex Fridman (13:03.920)
which ought to be basically right,
Lex Fridman (13:07.640)
and lead to a home maintained in a way
Lex Fridman (13:12.800)
that it's learned you like,
Lex Fridman (13:14.960)
but also be perpetually ready to take direction
Lex Fridman (13:21.560)
that would activate a different set of behaviors
Colin Angle (13:26.360)
or actions to meet a current need
Lex Fridman (13:28.960)
to the extent it could actually perform that task.
Lex Fridman (13:32.360)
So I got to ask you, I think this is a fundamental
Lex Fridman (13:35.400)
and a fascinating question,
Colin Angle (13:37.000)
because iRobot has been a successful company
Lex Fridman (13:39.800)
and a rare successful robotics company.
Lex Fridman (13:42.360)
So Anki, Jibo, Mayfield Robotics with their robot curry,
Lex Fridman (13:46.760)
SciFi Works, Rethink Robotics, these are robotics companies
Colin Angle (13:50.560)
that were founded and run by brilliant people.
Lex Fridman (13:54.040)
But all, very unfortunately, at least for us roboticists,
Colin Angle (13:59.680)
all went out of business recently.
Lex Fridman (14:02.120)
So why do you think they didn't last longer?
Lex Fridman (14:05.120)
Why do you think it is so hard to keep a robotics company alive?
Lex Fridman (14:10.640)
You know, I say this only partially in jest
Colin Angle (14:14.120)
that back in the day before Roomba,
Lex Fridman (14:17.800)
you know, I was a high tech entrepreneur building robots.
Lex Fridman (14:23.920)
But it wasn't until I became a vacuum cleaner salesman
Lex Fridman (14:26.400)
that we had any success.
Colin Angle (14:29.400)
So, I mean, the point is technology alone
Lex Fridman (14:34.200)
doesn't equal a successful business.
Colin Angle (14:37.680)
We need to go and find the compelling need
Lex Fridman (14:43.560)
where the robot that we're creating
Colin Angle (14:47.600)
can deliver clearly more value to the end user
Lex Fridman (14:53.600)
than it costs.
Lex Fridman (14:55.400)
And this is not a marginal thing
Lex Fridman (14:59.000)
where you're looking at the scale and you're like,
Colin Angle (15:00.400)
yeah, it's close.
Lex Fridman (15:01.800)
Maybe we can hold our breath and make it work.
Colin Angle (15:04.360)
It's clearly more value than the cost of the robot
Lex Fridman (15:11.560)
to bring, you know, in the store.
Lex Fridman (15:13.840)
And I think that the challenge has been finding
Lex Fridman (15:17.440)
those businesses where that's true
Colin Angle (15:24.760)
in a sustainable fashion.
Lex Fridman (15:28.240)
You know, when you get into entertainment style things,
Colin Angle (15:34.600)
you could be the cat's meow one year,
Lex Fridman (15:38.240)
but 85% of toys, regardless of their merit,
Colin Angle (15:43.360)
fail to make it to their second season.
Lex Fridman (15:45.600)
It's just super hard to do so.
Lex Fridman (15:48.720)
And so that's just a tough business.
Lex Fridman (15:53.840)
And there has been a lot of experimentation
Colin Angle (15:57.760)
around what is the right type of social companion,
Lex Fridman (16:02.680)
what is the right robot in the home
Colin Angle (16:05.840)
that is doing something other than tasks people do every week
Lex Fridman (16:14.600)
that they'd rather not do.
Lex Fridman (16:17.520)
And I'm not sure we've got it all figured out right.
Lex Fridman (16:20.920)
And so that you get brilliant roboticists
Colin Angle (16:23.000)
with super interesting robots
Lex Fridman (16:25.720)
that ultimately don't quite have that magical user experience
Lex Fridman (16:32.880)
and thus that value benefit equation remains ambiguous.
Lex Fridman (16:40.600)
So you as somebody who dreams of robots changing the world,
Lex Fridman (16:45.760)
what's your estimate?
Lex Fridman (16:48.800)
How big is the space of applications
Colin Angle (16:53.280)
that fit the criteria that you just described
Lex Fridman (16:55.760)
where you can really demonstrate an obvious significant value
Lex Fridman (17:00.560)
over the alternative non robotic solution?
Lex Fridman (17:05.720)
Well, I think that we're just about none of the way
Colin Angle (17:10.040)
to achieving the potential of robotics at home.
Lex Fridman (17:13.400)
But we have to do it in a really eyes wide open,
Colin Angle (17:20.440)
honest fashion.
Lex Fridman (17:21.920)
And so another way to put that is the potential is infinite
Colin Angle (17:25.440)
because we did take a few steps,
Lex Fridman (17:27.080)
but you're saying those steps are just very initial steps.
Lex Fridman (17:29.680)
So the Roomba is a hugely successful product,
Lex Fridman (17:32.600)
but you're saying that's just the very, very beginning.
Colin Angle (17:34.280)
That's just the very, very beginning.
Lex Fridman (17:36.560)
It's the foot in the door.
Lex Fridman (17:38.040)
And I think I was lucky that in the early days of robotics,
Lex Fridman (17:45.120)
people would ask me, when are you going to clean my floor?
Colin Angle (17:48.440)
It was something that I grew up saying,
Lex Fridman (17:53.600)
I got all these really good ideas,
Lex Fridman (17:54.800)
but everyone seems to want their floor clean.
Lex Fridman (17:58.120)
And so maybe we should do that.
Colin Angle (18:02.400)
Yeah, your good ideas.
Lex Fridman (18:03.400)
Earn the right to do the next thing after that.
Lex Fridman (18:05.840)
So the good ideas have to match with the desire of the people
Lex Fridman (18:10.120)
and then the actual cost has to like the business,
Colin Angle (18:13.240)
the financial aspect has to all match together.
Lex Fridman (18:16.640)
Yeah, during our partnership back a number of years ago
Colin Angle (18:21.240)
with Johnson Wax, they would explain to me
Lex Fridman (18:24.080)
that they would go into homes and just watch how people lived
Lex Fridman (18:32.520)
and try to figure out what were they doing
Lex Fridman (18:35.560)
that they really didn't really like to do,
Lex Fridman (18:39.960)
but they had to do it frequently enough
Lex Fridman (18:42.440)
that it was top of mind and understood as a burden.
Colin Angle (18:51.720)
Hey, let's make a product or come up with a solution
Lex Fridman (18:55.840)
to make that pain point less challenging.
Lex Fridman (19:02.040)
And sometimes we do certain burdens so often as a society
Lex Fridman (19:07.400)
that we actually don't even realize,
Colin Angle (19:09.400)
like it's actually hard to see that that burden
Lex Fridman (19:11.480)
is something that could be removed.
Lex Fridman (19:13.200)
So it does require just going into the home and staring at,
Lex Fridman (19:17.120)
wait, how do I actually live life?
Lex Fridman (19:19.560)
What are the pain points?
Lex Fridman (19:21.080)
Yeah, and getting those insights is a lot harder
Colin Angle (19:26.400)
than it would seem it should be in retrospect.
Lex Fridman (19:29.360)
So how hard on that point?
Colin Angle (19:33.120)
I mean, one of the big challenges of robotics
Lex Fridman (19:37.440)
is driving the cost down to something
Colin Angle (19:42.200)
that consumers, people would afford.
Lex Fridman (19:45.680)
So people would be less likely to buy a Roomba
Colin Angle (19:48.840)
if it cost $500,000, which is probably
Lex Fridman (19:53.320)
sort of what a Roomba would cost several decades ago.
Lex Fridman (19:58.040)
So how do you drive, which I mentioned is very difficult,
Lex Fridman (20:02.200)
how do you drive the cost of a Roomba or a robot down
Lex Fridman (20:05.320)
such that people would want to buy it?
Lex Fridman (20:07.920)
When I started building robots, the cost of the robot
Colin Angle (20:11.600)
had a lot to do with the amount of time it took to build it.
Lex Fridman (20:15.480)
And so that we build our robots out of aluminum,
Colin Angle (20:18.400)
I would go spend my time in the machine shop
Lex Fridman (20:21.400)
on the milling machine, cutting out the parts and so forth.
Lex Fridman (20:28.000)
And then when we got into the toy industry,
Lex Fridman (20:29.680)
I realized that if we were building at scale,
Colin Angle (20:34.280)
I could determine the cost of the Roomba
Lex Fridman (20:35.920)
instead of adding up all the hours to mill out the parts,
Lex Fridman (20:38.880)
but by weighing it.
Lex Fridman (20:42.080)
And that's liberating.
Colin Angle (20:44.200)
You can say, wow, the world has just
Lex Fridman (20:48.600)
changed as I think about construction
Colin Angle (20:51.560)
in a different way.
Lex Fridman (20:53.160)
The 3D CAD tools that are available to us today,
Colin Angle (20:56.920)
the operating at scale where I can do tooling and injection
Lex Fridman (21:02.920)
mold, an arbitrarily complicated part,
Lex Fridman (21:07.120)
and the cost is going to be basically
Lex Fridman (21:09.840)
the weight of the plastic in that part,
Colin Angle (21:13.960)
is incredibly exciting and liberating
Lex Fridman (21:16.360)
and opens up all sorts of opportunities.
Lex Fridman (21:18.560)
And for the sensing part of it, where we are today is instead
Lex Fridman (21:26.760)
of trying to build skin, which is really hard.
Colin Angle (21:30.640)
For a long time, I spent creating strategies and ideas
Lex Fridman (21:38.240)
around how could we duplicate the skin on the human body
Colin Angle (21:42.720)
because it's such an amazing sensor.
Lex Fridman (21:47.880)
Instead of going down that path, why don't we focus on vision?
Lex Fridman (21:53.920)
And how many of the problems that
Lex Fridman (21:57.280)
face a robot trying to do real work
Lex Fridman (22:02.880)
could be solved with a cheap camera and a big ass computer?
Lex Fridman (22:09.680)
Moore's law continues to work.
Colin Angle (22:12.360)
The cell phone industry, the mobile industry
Lex Fridman (22:16.440)
is giving us better and better tools that can run
Colin Angle (22:19.360)
on these embedded computers.
Lex Fridman (22:21.000)
And I think we passed an important moment maybe
Colin Angle (22:27.880)
two years ago where you could put machine vision
Lex Fridman (22:33.640)
capable processors on robots at consumer price points.
Lex Fridman (22:39.520)
And I was waiting for it to happen.
Lex Fridman (22:42.960)
We avoided putting lasers on our robots to do navigation
Lex Fridman (22:49.480)
and instead spent years researching
Lex Fridman (22:51.800)
how to do vision based navigation
Colin Angle (22:54.600)
because you could just see where these technology
Lex Fridman (23:00.400)
trends were going.
Lex Fridman (23:01.520)
And between injection molded plastic and a camera
Lex Fridman (23:06.680)
with a computer capable of running machine learning
Lex Fridman (23:10.800)
and visual object recognition, I could
Lex Fridman (23:12.920)
build an incredibly affordable, incredibly capable robot.
Lex Fridman (23:18.640)
And that's going to be the future.
Lex Fridman (23:21.160)
So on that point with a small tangent,
Lex Fridman (23:23.400)
but I think an important one, another industry in which I
Lex Fridman (23:27.000)
would say the only other industry in which there
Colin Angle (23:30.760)
is automation actually touching people's lives today
Lex Fridman (23:34.760)
is autonomous vehicles.
Lex Fridman (23:37.560)
What the vision you just described
Lex Fridman (23:40.160)
of using computer vision and using cheap camera sensors,
Colin Angle (23:44.400)
there's a debate on that of LIDAR versus computer vision.
Lex Fridman (23:48.160)
And the Elon Musk famously said that LIDAR
Colin Angle (23:54.360)
is a crutch that really in the long term,
Lex Fridman (23:58.360)
camera only is the right solution, which echoes some
Colin Angle (24:02.000)
of the ideas you're expressing.
Lex Fridman (24:03.480)
Of course, the domain in terms of its safety criticality
Colin Angle (24:06.760)
is different.
Lex Fridman (24:07.600)
But what do you think about that approach
Lex Fridman (24:10.640)
in the autonomous vehicle space?
Lex Fridman (24:13.360)
And in general, do you see a connection
Colin Angle (24:15.160)
between the incredible real world challenges
Lex Fridman (24:18.480)
you have to solve in the home with Roomba?
Lex Fridman (24:20.680)
And I saw a demonstration of some of them, corner cases
Lex Fridman (24:24.000)
literally, and autonomous vehicles.
Lex Fridman (24:27.800)
So there's absolutely a tremendous overlap
Lex Fridman (24:31.600)
between both the problems a robot vacuum
Lex Fridman (24:36.560)
and an autonomous vehicle are trying to solve
Lex Fridman (24:38.560)
and the tools and the types of sensors
Colin Angle (24:41.840)
that are being applied in the pursuit of the solutions.
Lex Fridman (24:47.960)
In my world, my environment is actually
Colin Angle (24:53.880)
much harder than the environment an automobile travels.
Lex Fridman (24:57.320)
We don't have roads.
Colin Angle (24:58.800)
We have t shirts.
Lex Fridman (25:01.280)
We have steps.
Colin Angle (25:02.680)
We have a near infinite number of patterns and colors
Lex Fridman (25:07.120)
and surface textures on the floor.
Colin Angle (25:10.160)
Especially from a visual perspective.
Lex Fridman (25:11.960)
So the way the world looks is an infinitely variable.
Colin Angle (25:18.840)
On the other hand, safety is way easier on the inside.
Lex Fridman (25:22.520)
My robots, they're not very heavy.
Colin Angle (25:26.480)
They're not very fast.
Lex Fridman (25:28.280)
If they bump into your foot, you think it's funny.
Lex Fridman (25:32.680)
And autonomous vehicles kind of have the inverse problem.
Lex Fridman (25:39.400)
And so that for me saying vision is the future,
Colin Angle (25:45.920)
I can say that without reservation.
Lex Fridman (25:49.440)
For autonomous vehicles, I think I
Colin Angle (25:51.720)
believe what Elon's saying about the future
Lex Fridman (25:56.840)
is ultimately going to be vision.
Colin Angle (25:59.040)
Maybe if we put a cheap lighter on there as a backup sensor,
Lex Fridman (26:02.000)
it might not be the worst idea in the world.
Lex Fridman (26:03.800)
So the stakes are much higher.
Lex Fridman (26:05.000)
The stakes are much higher.
Colin Angle (26:05.840)
You have to be much more careful thinking through how far away
Lex Fridman (26:09.080)
that future is.
Colin Angle (26:10.720)
Right.
Lex Fridman (26:11.520)
But I think that the primary environmental understanding
Colin Angle (26:17.880)
sensor is going to be a visual system.
Lex Fridman (26:21.800)
Visual system.
Lex Fridman (26:23.000)
So on that point, well, let me ask,
Lex Fridman (26:25.560)
do you hope there's an iRobot robot in every home
Lex Fridman (26:29.440)
in the world one day?
Lex Fridman (26:31.880)
I expect there to be at least one iRobot robot in every home.
Colin Angle (26:38.120)
We've sold 25 million robots.
Lex Fridman (26:41.160)
So we're in about 10% of US homes, which is a great start.
Lex Fridman (26:47.080)
But I think that when we think about the numbers of things
Lex Fridman (26:52.240)
that robots can do, today I can vacuum your floor,
Colin Angle (26:57.520)
mop your floor, cut your lawn, or soon
Lex Fridman (27:00.040)
we'll be able to cut your lawn.
Lex Fridman (27:02.880)
But there are more things that we could do in the home.
Lex Fridman (27:06.640)
And I hope that we continue using the techniques I described
Colin Angle (27:12.240)
around exploiting computer vision and low cost
Lex Fridman (27:16.240)
manufacturing that we'll be able to create these solutions
Colin Angle (27:20.960)
at affordable price points.
Lex Fridman (27:22.640)
So let me ask on that point of a robot in every home,
Colin Angle (27:25.600)
that's my dream as well.
Lex Fridman (27:26.880)
I'd love to see that.
Colin Angle (27:29.960)
I think the possibilities there are indeed
Lex Fridman (27:31.880)
infinite positive possibilities.
Lex Fridman (27:34.520)
But in our current culture, no thanks to science fiction
Lex Fridman (27:39.760)
and so on, there's a serious kind of hesitation, anxiety,
Colin Angle (27:45.320)
concern about robots, and also a concern about privacy.
Lex Fridman (27:51.480)
And it's a fascinating question to me
Lex Fridman (27:55.040)
why that concern is amongst a certain group of people
Lex Fridman (27:59.560)
is as intense as it is.
Lex Fridman (28:02.840)
So you have to think about it because it's a serious concern.
Lex Fridman (28:05.480)
But I wonder how you address it best.
Lex Fridman (28:08.000)
So from a perspective of vision sensors,
Lex Fridman (28:09.840)
so robots that move about the home and sense the world,
Lex Fridman (28:14.320)
how do you alleviate people's privacy concerns?
Lex Fridman (28:19.720)
How do you make sure that they can
Lex Fridman (28:21.080)
trust iRobot and the robots that they share their home with?
Lex Fridman (28:26.680)
I think that's a great question.
Lex Fridman (28:28.200)
And we've really leaned way forward on this
Lex Fridman (28:33.760)
because given our vision as to the role the company intends
Colin Angle (28:39.440)
to play in the home, really for us,
Lex Fridman (28:43.720)
make or break is can our approach
Colin Angle (28:48.160)
be trusted to protecting the data
Lex Fridman (28:50.920)
and the privacy of the people who have our robots?
Lex Fridman (28:53.520)
And so we've gone out publicly with a privacy
Lex Fridman (28:57.600)
manifesto stating we'll never sell your data.
Colin Angle (29:00.360)
We've adopted GDPR not just where GDPR is required,
Lex Fridman (29:05.480)
but globally.
Colin Angle (29:09.120)
We have ensured that images don't leave the robot.
Lex Fridman (29:18.000)
So processing data from the visual sensors
Colin Angle (29:22.080)
happens locally on the robot.
Lex Fridman (29:23.680)
And only semantic knowledge of the home with the consumer's
Colin Angle (29:30.720)
consent is sent up.
Lex Fridman (29:32.720)
We show you what we know and are trying
Colin Angle (29:35.400)
to go use data as an enabler for the performance of the robots
Lex Fridman (29:45.000)
with the informed consent and understanding of the people who
Colin Angle (29:50.720)
own those robots.
Lex Fridman (29:55.720)
We take it very seriously.
Lex Fridman (29:56.840)
And ultimately, we think that by showing a customer that
Lex Fridman (30:03.840)
if you let us build a semantic map of your home
Lex Fridman (30:07.320)
and know where the rooms are, well, then
Lex Fridman (30:09.240)
you can say clean the kitchen.
Colin Angle (30:11.720)
If you don't want the robot to do that, don't make the map.
Lex Fridman (30:14.480)
It'll do its best job cleaning your home.
Lex Fridman (30:16.960)
But it won't be able to do that.
Lex Fridman (30:18.720)
And if you ever want us to forget that we know that it's
Colin Angle (30:21.120)
your kitchen, you can have confidence
Lex Fridman (30:24.400)
that we will do that for you.
Lex Fridman (30:26.800)
So we're trying to go and be a data 2.0 perspective company
Lex Fridman (30:35.800)
where we treat the data that the robots have
Colin Angle (30:39.200)
of the consumer's home as if it were the consumer's data
Lex Fridman (30:43.200)
and that they have rights to it.
Lex Fridman (30:47.360)
So we think by being the good guys on this front,
Lex Fridman (30:50.880)
we can build the trust and thus be entrusted
Colin Angle (30:55.080)
to enable robots to do more things that are thoughtful.
Lex Fridman (31:00.160)
You think people's worries will diminish over time?
Colin Angle (31:04.760)
As a society, broadly speaking, do you
Lex Fridman (31:07.040)
think you can win over trust not just for the company,
Lex Fridman (31:10.600)
but just the comfort that people have with AI in their home
Lex Fridman (31:14.920)
enriching their lives in some way?
Colin Angle (31:17.040)
I think we're in an interesting place today
Lex Fridman (31:19.560)
where it's less about winning them over
Lex Fridman (31:22.360)
and more about finding a way to talk about privacy in a way
Lex Fridman (31:26.720)
that more people can understand.
Colin Angle (31:28.880)
I would tell you that today, when there's a privacy breach,
Lex Fridman (31:33.320)
people get very upset and then go to the store
Lex Fridman (31:37.000)
and buy the cheapest thing, paying no attention
Lex Fridman (31:39.320)
to whether or not the products that they're buying
Colin Angle (31:42.080)
honor privacy standards or not.
Lex Fridman (31:44.600)
In fact, if I put on the package of my Roomba,
Colin Angle (31:50.040)
the privacy commitments that we have,
Lex Fridman (31:53.560)
I would sell less than I would if I did nothing at all.
Lex Fridman (31:58.680)
And that needs to change.
Lex Fridman (32:00.360)
So it's not a question about earning trust.
Colin Angle (32:02.880)
I think that's necessary but not sufficient.
Lex Fridman (32:04.920)
We need to figure out how to have
Colin Angle (32:06.520)
a comfortable set of what is the grade A meat
Lex Fridman (32:10.800)
standard applied to privacy that customers can trust
Lex Fridman (32:17.640)
and understand and then use in their buying decisions.
Lex Fridman (32:23.040)
That will reward companies for good behavior
Lex Fridman (32:25.520)
and that will ultimately be how this moves forward.
Lex Fridman (32:29.880)
And maybe be part of the conversation
Colin Angle (32:32.640)
between regular people about what it means,
Lex Fridman (32:34.800)
what privacy means.
Colin Angle (32:36.280)
If you have some standards, you can say,
Lex Fridman (32:38.440)
you can start talking about who's following them,
Colin Angle (32:41.080)
who does not have more.
Lex Fridman (32:42.640)
Because most people are actually quite clueless
Colin Angle (32:45.440)
about all aspects of artificial intelligence,
Lex Fridman (32:47.320)
the data collection, and so on.
Colin Angle (32:48.440)
It would be nice to change that for people
Lex Fridman (32:50.560)
to understand the good that AI can do.
Lex Fridman (32:52.800)
And it's not some system that's trying to steal
Lex Fridman (32:56.560)
all the most sensitive data.
Lex Fridman (32:58.760)
Do you think, do you dream of a Roomba
Lex Fridman (33:02.640)
with human level intelligence one day?
Lex Fridman (33:05.280)
So you've mentioned a very successful localization
Lex Fridman (33:10.520)
and mapping of the environment, being
Colin Angle (33:12.160)
able to do some basic communication to say,
Lex Fridman (33:14.680)
go clean the kitchen.
Lex Fridman (33:16.600)
Do you see in your maybe more bored moments,
Lex Fridman (33:22.880)
once you get the beer, to sit back with that beer
Lex Fridman (33:27.080)
and have a chat on a Friday night with a Roomba
Lex Fridman (33:30.840)
about how your day went?
Lex Fridman (33:34.160)
So to your latter question, absolutely.
Lex Fridman (33:38.640)
To your former question as to whether a Roomba
Colin Angle (33:40.760)
can have human level intelligence, not in my lifetime.
Lex Fridman (33:45.280)
You can have you.
Colin Angle (33:46.440)
I think you can have a great conversation,
Lex Fridman (33:49.720)
a meaningful conversation with a Roomba
Colin Angle (33:54.000)
without it having anything that resembles
Lex Fridman (33:56.320)
human level intelligence.
Lex Fridman (33:59.200)
And I think that as long as you realize that conversation
Lex Fridman (34:04.000)
is not about the robot and making the robot feel good.
Colin Angle (34:08.600)
That conversation is about you learning interesting things
Lex Fridman (34:14.880)
that make you feel like the conversation that you
Colin Angle (34:18.720)
had with the robot is a pretty awesome way
Lex Fridman (34:24.840)
of learning something.
Lex Fridman (34:27.400)
And it could be about what kind of day your pet had.
Lex Fridman (34:30.920)
It could be about how can I make my home more energy efficient.
Colin Angle (34:36.240)
It could be about if I'm thinking about climbing
Lex Fridman (34:40.000)
Mount Everest, what should I know?
Lex Fridman (34:44.440)
And that's a very doable thing.
Lex Fridman (34:48.640)
But if I think that that conversation
Colin Angle (34:51.480)
I'm going to have with the robot is
Lex Fridman (34:53.760)
I'm going to be rewarded by making the robot happy,
Colin Angle (34:56.840)
well, I could just put a button on the robot
Lex Fridman (34:58.720)
that you could push and the robot would smile.
Lex Fridman (35:00.600)
And that sort of thing.
Lex Fridman (35:02.200)
So I think you need to think about the question
Colin Angle (35:04.200)
in the right way.
Lex Fridman (35:06.720)
And robots can be awesomely effective at helping people
Colin Angle (35:12.680)
feel less isolated, learn more about the home
Lex Fridman (35:16.320)
that they live in, and fill some of those lonely gaps
Colin Angle (35:21.960)
that we wish we were engaged learning
Lex Fridman (35:24.320)
cool stuff about our world.
Colin Angle (35:26.640)
Last question.
Lex Fridman (35:28.800)
If you could hang out for a day with a robot
Colin Angle (35:32.280)
from science fiction, movies, books,
Lex Fridman (35:35.760)
and safely pick its brain for that day, who would you pick?
Colin Angle (35:42.080)
Data.
Lex Fridman (35:43.320)
Data.
Colin Angle (35:43.880)
From Star Trek.
Lex Fridman (35:45.280)
I think that A, data is really smart.
Colin Angle (35:49.600)
Data has been through a lot trying
Lex Fridman (35:51.440)
to go and save the galaxy.
Lex Fridman (35:53.520)
And I'm really interested actually in emotion
Lex Fridman (35:59.600)
and robotics.
Lex Fridman (36:01.200)
And I think you'd have a lot to say about that.
Lex Fridman (36:03.120)
Because I believe actually that emotion
Colin Angle (36:08.120)
plays an incredibly useful role in doing reasonable things
Lex Fridman (36:14.280)
in situations where we have imperfect understanding of
Colin Angle (36:16.920)
what's going on.
Lex Fridman (36:18.000)
In social situations when there's imperfect information.
Colin Angle (36:20.760)
In social situations, also in competitive or dangerous
Lex Fridman (36:26.320)
situations that we have emotion for a reason.
Lex Fridman (36:32.880)
And so that ultimately, my theory
Lex Fridman (36:36.320)
is that as robots get smarter and smarter,
Colin Angle (36:38.520)
they're actually going to get more emotional.
Lex Fridman (36:41.400)
Because you can't actually survive on pure logic.
Colin Angle (36:49.120)
Because only a very tiny fraction of the situations
Lex Fridman (36:53.800)
we find ourselves in can be resolved reasonably with logic.
Lex Fridman (36:57.320)
And so I think Data would have a lot to say about that.
Lex Fridman (36:59.600)
And so I could find out whether he agrees.
Colin Angle (37:02.360)
If you could ask Data one question,
Lex Fridman (37:04.840)
you would get a deep, honest answer to what would you ask.
Lex Fridman (37:08.560)
What's Captain Picard really like?
Lex Fridman (37:12.520)
OK, I think that's the perfect way to end it.
Colin Angle (37:14.320)
Colin, thank you so much for talking today.
Lex Fridman (37:16.040)
I really appreciate it.
Colin Angle (37:16.960)
My pleasure.
🔗 相关节目