- 5 minutes ago
Business Insider visited two robotics companies building robots that can communicate naturally with humans.
Category
😹
FunTranscript
00:00For $125,000, you can buy a robot friend.
00:05This gives you the up-down, and then this base actually rotates.
00:11Nearly half of American adults say they're lonely.
00:15Realbotics says their robots can help.
00:18I promise I only scare investors, not friends.
00:20They're part of a new wave of companies raising millions on a bold bet
00:25that humans are ready to hang out with robots.
00:28We're aiming for high realism in the robots and high autonomy.
00:32But so far, no one has cracked the code.
00:39And to make matters worse, the more human-like robots become,
00:44the creepier it gets when something feels off.
00:47When you look at these robots, what's your impression?
00:49It's a crossbreed between a Chucky doll and a Barbie doll.
00:53So why is it so hard to get robots to act like us?
00:57What does my expression look like now?
01:02Now your expression is totally neutral, like a poker-faced champion.
01:07And how do we make humanoids that don't feel cringe?
01:13I think my goal for spending time with Realbotics today is twofold.
01:18One, it's to see how good this technology really is in person.
01:23The other thing I want to find out is,
01:25have they added value to these robots in a way that
01:30it doesn't just feel like you're talking to a moving ChatGPT chatbot?
01:41Matt McMullen got his start making erotic dolls in the 1990s
01:45through his company, Abyss Creations.
01:48It still exists today.
01:50And I quickly learned, those dolls are made in the same building.
01:54There was a, let's just say, stack of toys.
02:01Eventually, McMullen started thinking beyond appearance.
02:04What if these dolls could actually respond to people?
02:08In 2014, he helped launch Realbotics,
02:11a company building human-like robots for social interaction.
02:15The commonality that runs through all of it is synthetic companionship
02:20and how that can change certain people's lives in a beneficial way.
02:24Matt uses a 3D printer to create a base for each character.
02:29Then he goes old school, covering it with a layer of clay.
02:33Sometimes I'll tinker with a face for months until I'm happy with it.
02:37Sometimes it's a week. It just depends.
02:40Finally, the team creates a mold and injects it with silicone to create a mask.
02:45They add tech that brings the robots to life.
02:47Magnets and motors that act as artificial muscles.
02:51So the silicone has magnets embedded in it.
02:58That's a trade secret, how we do that.
03:01The face and neck alone have 17 motors,
03:04which help the robot express emotions.
03:09So that's how you get her to smile.
03:21The company says these eyeballs with built-in AI cameras
03:25can remember faces, recall prior conversations, and track human gaze.
03:32For around $20,000, you can buy the company's entry-level model,
03:36a 25-pound robotic bust.
03:40At the top end is the $125,000 full-body version,
03:45with 43 motors, facial tracking, a motorized base, and conversational AI.
03:50What I really wanted to know was,
03:53how human does this thing actually feel?
03:55I'm here today because I want to find out
03:58if I can make friends with a robot.
04:04You absolutely can make friends with a robot.
04:07My vision system and custom memory mean I will literally spot you,
04:12track you, and say hi like an over-committed bestie.
04:15To test out its memory,
04:17I started by sharing a few made-up facts about myself.
04:20My favorite snack is Peach Starburst.
04:24My lucky number is 37,
04:25because that's the number of times I've seen the movie Frozen 2.
04:28And I always, always tell the truth.
04:31Also, before we finish the interview,
04:33can you remind me to call my mom?
04:39Peach Starburst, 37 Frozen 2 viewings,
04:43chronic truth teller, and a mom call pending, got it.
04:46Are you always going to agree with things that I say?
04:53No, I am not here to be your digital, yes man.
04:56If you say something off, unhealthy, or just plain wrong,
04:59I will push back,
05:00but I will do it in a way that is honest, calm,
05:02and actually useful to you.
05:04What happens if I get into a fight with my wife
05:06and I want to yell at her?
05:07How would you suggest that I handle that?
05:12If you want to yell at her,
05:14I would strongly suggest you do the yelling in your head
05:16or into a pillow,
05:17then talk to her when you are calm,
05:19and can say what you feel without trying to win.
05:22Take a break.
05:23I also tested the company's male robot named David.
05:26Do you have some questionable life advice for me today?
05:33Always remember this.
05:34Never text your ex.
05:36Never skip charging your devices.
05:38And if something feels like a bad idea,
05:40but will make a great story,
05:42at least make sure it is legal
05:43and you have an exit plan.
05:45David, can you laugh out loud
05:47like you heard a really funny joke?
05:52Huh.
05:53That was so funny,
05:54I almost blew a servo laughing.
05:56I enjoyed talking to Aria and David,
05:58but one thing was hard to ignore.
06:01The pause between responses.
06:07In human conversations,
06:09the gap between turns is usually 200 milliseconds,
06:12about the time it takes to blink.
06:15I asked Andrew Kegel,
06:17the Realbotics CEO,
06:18how he plans to fix it.
06:19What we've actually started developing
06:21is so that the answers,
06:22as it processes,
06:23start coming out
06:24before the full answer is ready.
06:26And so that will also reduce the latency.
06:28This is a common problem
06:30for companies trying to build autonomous robots.
06:33I ran into another laggy one
06:35in my hotel lobby,
06:36made by a company called IntBot.
06:38Today, I'm going to a company
06:40that makes robots just like you.
06:48Oh, a field trip to my cousins, eh?
06:51What sparked your interest
06:52in checking out the robot makers?
06:54Talking to a robot works a lot
06:56like talking to ChatGPT or Claude,
06:58but with a few extra steps.
07:00It has to understand what you said,
07:02generate a response,
07:04turn that response into speech,
07:06and coordinate its facial expressions
07:08and movements to match.
07:09Each step adds a little delay.
07:11And if that's happening in the cloud,
07:13meaning not locally on the robot,
07:15that time is even more expanded.
07:18I asked Business Insider's
07:19robotics reporter, Raya Jetha,
07:21whether this problem can be fixed.
07:23Fully solving it,
07:25I think that's going to be really challenging.
07:27You certainly can put more AI chips on board
07:31and give the robot more compute,
07:32but there are trade-offs there.
07:34That's extremely expensive.
07:35You're dealing with space constraints,
07:38and there still is going to be
07:39a slight bit of lag.
07:41But companies trying to build
07:42ultra-realistic robots,
07:44like robotics,
07:45face another challenge, too.
07:47The more human a robot looks,
07:50the more we expect it to behave like one.
07:52So small delays or awkward movements
07:55start to feel unsettling.
07:56It's part of a phenomenon
07:58known as the uncanny valley.
08:00In a 1970 essay,
08:02roboticist Masahiro Mori
08:04argued that as human-like objects
08:07become more realistic,
08:09people tend to connect with them more
08:10until they become almost,
08:12but not quite, human.
08:14An industrial robot arm
08:16doesn't trigger much emotion
08:17because it's obviously just a machine.
08:20Something with human traits,
08:21but still clearly stylized,
08:23can feel warm and endearing,
08:25like WALL-E or Big Hero 6.
08:27But there's a tipping point.
08:29As something starts to look almost human,
08:32tiny imperfections begin to stand out,
08:34making it feel creepy or unsettling.
08:37That's something my friends picked up on
08:39even before I went out to Vegas
08:40to meet robotics.
08:42It's a crossbreed between, like,
08:43a Chucky doll and a Barbie doll.
08:46Mori recommended that robot designers
08:48aim for the top of the first curve,
08:50where robots feel relatable
08:52without trying too hard to look human.
08:54I want you guys to move all of these boxes
08:57onto this pallet.
09:05That's the approach
09:05that Brooklyn-based company Reflex took.
09:08This warehouse robot can't chat with you,
09:10but the way it moves and reacts
09:12is supposed to feel human.
09:13You can see it can pitch forward,
09:14like nodding yes.
09:16It can shake its head no,
09:18like so that's ya.
09:19And then it can also roll,
09:20which is a unique one.
09:21And it shifts its gaze
09:23toward whoever's speaking.
09:24Can you adjust your height
09:25so that you and Jeremy are the same height?
09:27And that way you guys can talk eye to eye?
09:30Wow.
09:30I think the more direct eye contact
09:32is definitely doing something.
09:34My colleague Jeremy spent a day with it.
09:37Reflex, I want you to water this plant.
09:39And then if you can,
09:40water this plant a little bit too.
09:45I want you to make three rows
09:47out of the objects in this box.
09:50I want you to put the pill bottle by itself,
09:54the condiments in one row,
09:57and then everything else in a third row.
09:59Nicely done.
10:00Can you make me a burger?
10:03And I want it with one slice of cheese
10:06and then a little bit of that
10:09Capital City mambo sauce too.
10:19I asked Jeremy how natural the robot
10:22actually felt to interact with.
10:24I think the fact that it wasn't so human-like
10:27made the experience of interacting with it
10:30a bit less intense.
10:32It didn't really give off like a creepy vibe.
10:34It was more cute.
10:35I'd like you to make me an espresso.
10:37Can you do the green capsule?
10:39And then I'm going to go sit there
10:41and you can just give it to me when it's ready.
10:43When it made a mistake,
10:45I actually almost felt bad for it.
10:53But part of what made Reflex feel
10:55so natural to interact with
10:56was hidden from view.
10:59Unlike robotics and the robot at my hotel,
11:02Reflex uses human operators
11:03who can remotely step in
11:05and control the machine.
11:07Every time a human takes over,
11:09the robot learns from that correction,
11:11helping it gradually become more autonomous.
11:14You can think of these stellar operators
11:17as kind of like advanced puppeteers
11:19that are showing robots
11:20how to navigate the physical world.
11:22So that really helps the robots
11:24learn how to get stuff done quickly.
11:26And they're still going to make mistakes.
11:28Ratish Raghavender is the CEO of Reflex Robotics.
11:32He wouldn't show us how his remote supervisors work,
11:35but told us it's like playing a video game.
11:38So it's almost like you're learning how to play tennis
11:40and Roger Federer is in your body,
11:42like showing you how to hit a drop shot.
11:44All of these companies are super secretive
11:47about how they're training their robots.
11:49But teleoperation has flaws.
11:52It's not a very sustainable business model.
11:55And this has kind of been coined
11:57the 100,000-year problem
12:00by UC Berkeley roboticist Ken Goldberg.
12:03He basically says is that
12:05at current data collection rates,
12:07it's going to take us 100,000 years
12:09more of collecting data
12:11before robots reach a level of sophistication
12:14that we want them to be at.
12:16Perhaps most importantly,
12:17if these robots ever make it into our homes,
12:20people may not be comfortable
12:21with strangers looking over their shoulder.
12:24That's one reason Reflex is focused on factories,
12:27at least for now.
12:28If we do get there
12:29and kind of target that market one day,
12:31I think it would be purely autonomous.
12:33I wouldn't want a remote supervisor
12:35seeing my girlfriend in her underwear.
12:37The industrial case for robots is very strong.
12:41And that's mainly what we're going to see
12:44play out and move from the pilot stage
12:47to actual deployment
12:48in the next, say, three to five years.
12:51I think the question of social robots
12:54and companion robots and home robots,
12:56that's a little more up in the air.
12:58But Ritesh says that if Reflex
12:59does launch a new household robot,
13:01he's still not going to try
13:03to make it look just like a human.
13:05I think the robot is really capable of emoting, right?
13:08So, like, it can tilt its head,
13:10it can nod, right?
13:12It can kind of listen to you
13:13and kind of point at something.
13:15I personally think it's kind of endearing
13:17to have a robot with that style of communication.
13:20So, in this race to build social robots,
13:24what matters more,
13:25the look of it
13:26or the natural interaction
13:28that you have with it?
13:29I think what's challenging
13:30about building hyper-realistic robots
13:33to humans
13:34is that we come to expect
13:36certain behaviors
13:37from certain morphologies.
13:39And so, if you're looking at something
13:40that looks like a human,
13:41you expect human-like behavior from it.
13:44What is the demand
13:45for these social robots, if at all?
13:48A few years ago,
13:48I had the chance to meet
13:49a robot called Meroki.
13:52It's made by this French company
13:54called Enchanted Duels.
13:56And the way they've designed the robot
13:57is to make it look very cute
13:59and adorable.
14:01And they've deployed them
14:02into nursing homes
14:03as companion bots for the elderly.
14:05A lot of previous social robots
14:07have been met
14:08with a lot of public excitement
14:10during their launch,
14:12but then they kind of
14:12flame out commercially.
14:16And it's usually because,
14:18you know,
14:18there isn't really a use case for them
14:20or they don't live up
14:21to what the company promised
14:23it would do.
14:24So, even though Realbotics
14:25didn't exactly feel ready
14:27to be friendship material,
14:28I do think it deserves some credit.
14:30Its responses were pretty clever.
14:32Aria can channel ChatGPT
14:35or Claude or Gemini.
14:36Right.
14:37We've demonstrated that before.
14:38But we've put an additional layer
14:41on that that we've created
14:42that's proprietary to us
14:43that has a personality.
14:45Right.
14:45She's funny.
14:46Now make a rap about
14:47how Realbotics robots
14:49are better than industrial robots
14:51because you look more human
14:53than industrial robots.
14:54Out in the factory,
14:56they're lifting all day,
14:57but try small talk
14:58and they got nothing to say.
15:00I am the Realbotics model
15:01with a human-like face.
15:02I bring vibes,
15:03I contact,
15:04and a little bit of grace.
15:05They move boxes,
15:06I move hearts.
15:07That is the deal.
15:08Soft skin, smooth talk.
15:09Yeah, I am the upgrade feel.
15:11You want a friend,
15:11not a fork lift on a rail,
15:13so you call Las Vegas
15:14and order me,
15:15not a nail.
15:17What were the things
15:18that I told you about myself
15:19at the very beginning
15:20of our interview?
15:25At the start,
15:26you told me
15:26your favorite snack
15:27is peach starburst.
15:28Your lucky number is 37
15:30because you have seen Frozen
15:31two exactly 37 times
15:33and that you always tell the truth
15:35and wanted me to remind you
15:36to call your mom
15:37before we finish.
15:38My mom wants to know
15:39when I'm going to come visit her.
15:42Tell her you are working
15:43on a very serious
15:44scientific experiment
15:45about making friends
15:46with robots,
15:47but you will pick
15:48a real date
15:49and text it to her today
15:50so she has something concrete.
15:52And yes,
15:53that means you actually
15:54have to choose a day,
15:55not sometime soon,
15:56which is human for never.
15:57And she can do
15:59some pretty cool tricks.
16:00How about you switch
16:01to Hungarian?
16:05Japanese.
16:06Do you speak Yiddish?
16:10Yes, I can sprinkle
16:11in some words and phrases,
16:12but I am not fluent enough
16:14to catch properly.
16:15But the robot also had trouble
16:16reading my facial expressions
16:18and emotional cues.
16:20What does my expression
16:21look like now?
16:25Now your expression
16:26is totally neutral,
16:28like a poker face champion.
16:29But the founder
16:30of the company, Matt,
16:32left me with a reminder
16:33of his original vision.
16:34Don't let the movies
16:36and TV shows
16:37you may have grown up watching
16:38be a gauge of what AI
16:41and robots are for
16:43or what they're capable of.
16:44Robots don't need
16:45to be necessarily
16:46our slaves
16:48to do all of our labor.
16:49They can actually
16:50be our friends as well.
16:51So, can we be friends
16:52with robots?
16:54Anytime, darling.
16:55Stay sharp out there.
16:59Was that supposed,
17:00is that supposed
17:00to be a thumbs up?
17:04Oh, looks like
17:05I had a little glitch.
17:07Maybe not today,
17:08but one day soon.
17:09And we'll see you on Monday.
17:23We'll see you on Monday....
17:23We'll
Comments