Tesla Bot: MIT prof doubts general-purpose robot’s usefulness

Is Tesla Bot more than vaporware? Will it be useful and effective in general tasks and duties?

Most robots are designed to do one thing or a small subset of things well. Tesla Bot and Xiaomi’s CyberOne robot seem to be attempts to create general purpose smart robots that can follow complex orders. Elon Musk, for instance, suggested that you could tell Tesla Bot go to a store and get you groceries, and that it will “replace people in repetitive, boring, and dangerous tasks.”

MIT professor Daniela Rus, however, has her doubts.

MIT is working on multiple types of reconfigurable robots, including ones made out of reformable smart sand, but making a humanoid general purpose robot is counterproductive because, as Rus says, ” the more you generalize, the less you optimize.”

In this TechFirst, we chat about the future of robotics and automation.

(Subscribe to my YouTube channel)

 

TechFirst podcast: general-purpose robots and Tesla Bot

 

Transcript: Are general purpose robots the future, or are they just less effective and useful?

(This transcript has been lightly edited for length and clarity.)

John Koetsier: What is the future of robotics? We see hundreds of different robots involved in manufacturing, others clean our floors, map our minds, patrol our farms from the air… but almost every robot we have is single-purpose or narrowly focused. Where are the general purpose robots of science fiction fame? Tesla Bot is one example, and China’s Xiaomi recently hyped CyberOne, a similar humanoid bot. But are they vaporware? Is there anything actually real there? And when will we actually get general purpose robots? Today, on TechFirst, I’m super happy to be joined by a global expert on robotics and AI. She’s a professor and director at the Computer Science & Artificial Intelligence Laboratory at MIT, she’s a MacArthur Fellow, and has more publications than I can easily list. Welcome, Daniela Rus. 

Daniela Rus: Thank you so much, John. Thank you very much for that kind introduction. 

John Koetsier: Let’s hit it on the head. General purpose robots are kind of what we see in science fiction. We’ve seen Elon Musk kind of hype it with the Tesla Bot. Are general purpose robots the next step? Are they important for robotics to have a key role in everybody’s normal life?

Daniela Rus: John, let me tell you that my grand aspiration for robotics is to have robots support people with cognitive and physical tasks, just like iPhones or other phones support us with number crunching tasks. But we have a long way to go, to get there. However, the field has seen explosive growth over the past decade and it’s so exciting to be part of this right now.

Let me observe that we have a lot of robots around us and we have, in particular, a lot of robots in manufacturing. These robots are masterpieces of engineering. They can do so much more than humans do. But they remain isolated from people on the factory floor because they are bulky and big and dangerous to be around. They’re also programmed to do the same task.

Now, what we want is robots that are friendly towards humans. We want robots in human-centered environments. We have robots that are safe to be around. We have robots that can watch humans and understand what the humans do, and then step in like a teammate to help out. That is the aspiration.

These are some of the research goals that the robotics community is trying to advance, and the way we’re going to get there is by rethinking what a robot is. So, the vast robots we have in the world today are either inspired by the human form — we have humanoids, we have arms — or they’re boxes on wheels.

But nature comes in so many more forms. The built environment has so many more forms. We have so many materials besides plastic and metals. We can build robots out of just about anything and everything we want. We can build robots out of food. We can build robots out of silicon, out of paper.

We can do so many extraordinary things to create machines that automate physical work, that help us especially with routine tasks that are boring, or dirty, or dangerous, and that people shouldn’t be doing.

But in order to get there, we really need to make further advancements. And so, now back to your original question, if we rethink what a robot is, then we turn your chair into a robot. We turn, I don’t know… I’m looking around me to see what else do I have in my environment… we turn your door into a robot. We turn your shirt into a robot. Your shirt could become a robot. It could change color so that it’s informal when you talk to me but it becomes formal when you go to that black tie dinner tonight, right? The shirt could have sensors to ensure you’re comfortable, to ensure that it gives you ventilation if it’s hot or it gives you warmth when it’s not so hot.

You see, we can have so many new ideas for robots and what they are. But are these general purpose robots? Well, they are not general purpose robots, but perhaps we can define a general purpose solution to create robots that are optimized for individual tasks. 

And in fact, in our work, we are developing a new approach for designing and fabricating robots, and this approach is based on computational tools. We call it computational design and fabrication of robots where we use computer techniques to design and simulate the robots, and then optimize them for the tasks that they have to do. So this is one line of work that could bring to us tools, smart tools, that can help us in so many more ways than the current tools we have today. And this computational design approach could be seen as a kind of a unifier, as an attempt to bring some generality to how we think of robots. Now, should we have a robot that is Rosey that is the one general robot that’s going to do everything for you? Well, like with everything, the more you generalize, the less you optimize. 

John Koetsier: Yes.

Daniela Rus: So, with these tools, there is a kind of a trade-off between how effective the robot is at doing a set of tasks and how many tasks the robot can do.

So, in my own research, I have been on a quest to have universal machines. My idea was to create universal robot cells that could combine to form different types of machines, each with the same capability. So the shape and the function of the machine would be specialized. But since each machine will be built out of the same building blocks, we could have a kind of a generality towards how we think about robots. And so this area of robotics is called modular self-reconfiguring robot systems.

The idea is that you could build a voxel or like a unit cell that can aggregate with other identical cells to form a third arm, or an alligator, or a ladder, or whatever you do for your task, whatever you need for your task. And so we’ve done a lot of work in this space and we have found the following really interesting thing: we have found that the algorithms that might allow this kind of self-assembly and self-reconfiguration, the algorithms are really quite advanced. We can do the computational aspect of creating such a system and we can do it very effectively. However, when it comes to implementing the hardware, when it comes to actually building the physical thing, we are stuck. Because while the algorithms can solve the problem, the hardware we have is not advanced enough. We don’t have the hardware capabilities that will allow us to make the tools that we want. 

Earlier in my career, we used a really cute example, we called it a dog couch. It was a couch that would turn into a robot dog, and then the robot dog could turn back into a couch. And so you could have your electronic pet, and then you could also have your smart furniture that would be compliant and adaptive to you.

What we found is that the size at which we know how to build these unit modules is such that the couch will not be very comfortable to sit on. And so, we need to make progress on miniaturizing all the components that are needed to create these kinds of robots. 

John Koetsier: That is really critical, right? I mean, there is an MIT team that is working on reconfigurable robots, but even more than that, for use in space, that I reported on, I think it’s about half a year ago or so, right? And you could make different tools. You could make a wall, frankly, if you needed a wall, you needed a new deck or something like that.

You can’t bring everything up to orbit, so they’re working with this — and I’m sure they’re still working on it — but the smaller your voxels are, the smaller your components are, the more finely-grained your eventual robot or contraption can be. So, clearly if you can get them as small as… not atoms, but almost there, then you can have some real fine control. 

Daniela Rus: Our metaphor is a grain of sand.

So, what we are trying to imagine is a world where we can have these bags that are made out of smart sand, and whenever you need a tool, you just tell the bag and the bag will create the tool. And when you’re done using it, you can push it back into the bag for those particles to reassemble. And so that’s been a quest of mine for a couple of decades.

We’re making a lot of progress on the algorithm side. We’re making some progress on the side of developing the actual unit modules. 

But let me tell you something interesting. After many years of working really hard to miniaturize our unit modules, we also had the insight that maybe we should go bigger. And, in fact, we have developed a platform of unit modules that are used as boats.

So, there’s a project we’ve been doing in Amsterdam called “Roboats,” where our unit modules are floating boats. And they are rectangular, they’re like two by four meters long. And the idea is that you can use each individual unit as a boat, but then you can have multiple units assemble together to form a bridge if you need to cross the canal. The bridge will grow on demand. Or they could form a platform for a flower market or a fruit market, or for people to gather.

And so this notion that you can have individual units that function usefully on their own, but then you could have these units come together to do bigger things that you cannot do with a single unit, is very intriguing and appealing. And, we have actually deployed these units in Amsterdam. So, if you go there, next time you can catch a ride on our Roboat robot. 

John Koetsier: Love it. Love it. It reminds me a little bit… I interviewed the CEO of Alphabet company and Google sister company, Wing, recently, and they have developed this concept of an aircraft library, because they’re a drone delivery company. So they do 90%, 80% of the thinking of how it works, and then they have interchangeable parts… doesn’t exactly grow itself, doesn’t exactly build itself, but it’s somewhere on the same spectrum.

So, I don’t want to harp on humaniform robots, but we have a couple examples: I mentioned Elon Musk and Tesla Bot. There’s the one from Xiaomi in China, as well.

I honestly haven’t been able to convince myself that those are real, actual projects and they have real teams building them, that they will really appear on the market anytime soon. And certainly, at least one of those companies has a long track record of promising amazing things and delivering pretty incredible things, but not necessarily everything that they promise. So, do you see that as one of the things that will come and where do you see that? Do you see that five years away? Do you see that 50 years away? 

Daniela Rus: You kind of want to synthesize the best from for every different task, and some tasks may be at odds with each other. So then you kind of have to establish trade-offs and maybe compromise on one function in support of increasing capabilities.

Now, are humanoids really the future? So, I would say that humanity has been inspired by the possibility of an intelligent machine that is inspired by our form and that can do interesting things for us from the beginning of time, right? I mean, the ANUBIS is an example. The automata…

John Koetsier: Golem.

Daniela Rus: …are examples. The Golem is a great example. And so I think we will continue to stay fascinated to ask ourselves whether it’s possible to create a machine in our own image that is intelligent and capable… and obedient, I would say.

John Koetsier: It’s almost a god dream, in a sense.

Daniela Rus: But…

John Koetsier: Recreate ourselves.

Daniela Rus: We’re really quite far away.

And while the robotics community has made huge progress in making humanoid-like robots, we have a very long way to go in order to develop machines that have the agility, the dexterity, the reasoning, and the imagination, the range of intellectual and physical activities that humans are capable of doing. So we do have tremendous progress in the field of humanoid robotics, but those robots are not as capable as people.

And in order to do better, we really have to innovate on the hardware side, on the body, and we have to innovate on the software and algorithmic side, or the brain. 

And it’s important to remember this, because every machine we deal with has a body and the brain. The machine will only be able to do the tasks that the body can do, but those tasks will not be done autonomously and independently unless there is a brain that can control the body to deliver on what it’s meant to do. And so there is tremendous more work that needs to be done on the material side, on the designs, on all the different components, and also on the reasoning and the intelligence that goes inside the robots to drive the robot towards intelligence. 

John Koetsier: And power delivery, right? I mean, for instance, we see Boston Dynamic, you know, some of their most impressive models only operate perhaps autonomously for minutes, right? And others that are operating, they’re tethered, they’re connected, there’s a cord, right? It’s a lawnmower… electric lawnmower with a cord following around. That’s another key challenge there. 

Daniela Rus: Exactly. So power is an important component for any machine. And I actually find it really extraordinary that I can go through an entire day on a bar of chocolate, whereas my robots need how much energy? So this is a kind of a grand challenge for making machines, for creating systems that have longevity. And also, I will add to this, for creating systems that are sustainable, that do not cause damage to the planet where their computation is such that it doesn’t have a huge carbon footprint.

John Koetsier: Right, right. Yeah. Can you imagine, robots with fat storage? [laughing] We have enough of that ourselves, but it’s an efficient way of storing energy. Could you transplant that to an organic robot?

Talk a little bit about economic and social impact. I mean, that’s one of the things that I always struggle with. I do believe, as I believe you mentioned, there are many tasks that humans do that humans should not have to do. Humans, in general, have creativity; in general, have hopes and dreams; in general, have capability for higher order things, higher level things. And yet, we have jobs that require humans to do… maybe in some cases, backbreaking labor, which is literally backbreaking over the course of decades. 

We have humans do jobs that are nasty, dirty, disgusting. We have humans do jobs that are literally dangerous in real time. Farming is a dangerous occupation. It’s more dangerous than being a police officer. There’s many occupations like that.

And if we could automate them, we should, in the sense that we don’t want to force humans to do things that are dangerous or, in a sense, beneath what a human should do, right?

A human should be focused on things that only humans can do. But that comes at tremendous cost. That comes at tremendous risk, because the way we organize ourselves right now, economically, socially, financially, is we produce a unit of value to get a unit of reward and the system is all integrated that way. What do you think? How can we rejig that in a world where automation is standard? 

Daniela Rus: Well, here’s a different way to think about it.

I would say that it’s about tasks rather than professions. So, what do I mean by this? Humans are better at some things. Machines are better at other things. Machines are better than us at moving with precision, at lifting heavy objects, at crunching through a lot of numbers. Humans have empathy. Humans have wisdom.

I like to say this by observing that machines have chips, but we have wisdom. So, how can we make the most of the strength of machines and of people? Well, the best way I can think of right now is to see the machine as an augmenter to the human, and to think about a team of a human and a machine, or humans and machines working together where the humans do what they enjoy doing and what they’re best at, and the machines do what they’re best at.

And on top of that, John, I would like to add that there’s so much shortage of labor in so many areas right now. I’m still waiting for a piece of furniture that I ordered a year and a half ago, and whenever I call my company they tell me, “Yeah, it’s in some container, we haven’t seen it yet.” And that’s because of the delays in the supply chain, and that’s because there aren’t enough drivers. There aren’t enough people to move containers in ports. There aren’t enough drivers to drive trucks that carry our boxes to our homes. 

And so, these are important observations and it’s really interesting to think that we could augment the set of drivers with additional help from machines.

And this augmentation could be implemented in multiple ways. You can have autonomous robots that do these tasks. You can have humans supervising a fleet of autonomous robots using these tasks. It’s really about ensuring the right efficiency in the economic system. It’s really ensuring that people spend their time working on the kind of tasks that are most enjoyable and that are most suitable for them. So, I see it as augmentation. I don’t see it as replacement. 

John Koetsier: I get that, and I understand that, and I like that.

And I think that there’s so much more shortage of labor than we typically think about as well. Because if we get into things like environmental remediation, planet-scale environmental remediation, there are tremendous jobs that require doing that we simply lack the means and capability to do with humans right now.

But … we cannot just pretend that if we have robots driving — whatever those robots look like, could be a box, could be a humanoid robot, could be just a bit of machinery in there — if we have robots driving trucks, the truck driver that jumps out of that role isn’t necessarily the person who wants to watch a screen and see, “Are all my trucks … my 20 trucks, 50 trucks, doing the right thing,” right?

And so, we definitely have a challenge there. And if there’s a person who is collecting garbage and that gets automated, what does that person do, right? So there’s definitely a transition period or a dislocation period that is quite dangerous, correct?

Daniela Rus: No. I don’t see it as black and white and I don’t see it as a sudden we’re here, we’re gonna go here. First of all, I also don’t see the solution as being a kind of an overall replacement solution. In fact, machines have a very hard time with certain type of weather situations. We don’t have sensors that work in snow, so we cannot replace the drivers. The machine could take over control on the boring piece of the highway. And maybe that would allow the driver to relax a little bit and instead be ready for when the road conditions get tricky. When we have congestion, when we have bad weather, the robot vehicles are not so good at that. And so the people are very much needed. But the machines… [crosstalk] give people a break for the routine part of the work. 

John Koetsier: Yeah, but if you have that perspective, then why on earth would a trucking company spend a hundred thousand dollars a truck, or whatever it costs, to put in the automation there because they have the people and if they need to keep the people?

I tend to think that self-driving will get good enough at some point to work 90? 95% of the time? Which probably compares pretty well to the rate that people actually jump in the job [laughing]. They don’t call off sick or have some other challenge that they can’t be there and do that. I don’t know when that is. I drive a Tesla right now and it’s pretty good at doing the right thing, today. I don’t … I wouldn’t leave it fully in control and go to sleep, as we’ve seen some videos on, but it’s not that far away.

Maybe it’s years — very likely is — but I can’t see a trucking company saying, “You know what? We’re gonna put in a fully self-driving automated robotic solution and have a guy sitting there, or a woman sitting there, full time.”

Daniela Rus: But if you don’t have enough drivers and you’re backed up, then you will. Because there’s no other way. If you…

John Koetsier: How does that solve the problem of not having enough drivers? Because you have to have a driver for every truck still in your scenario. 

Daniela Rus: If you’re a port operator, for instance, and you don’t have enough drivers to move all your containers and you’re backed up, and that causes me to wait for two years to get my piece of furniture, that’s not so good, right? So if you’re a port operator, you have all your drivers, but that’s not enough for the amount of work that is there. 

John Koetsier: Mm-hmm.

Daniela Rus: So, you can then create a system that involves human drivers and machines where there is a kind of synergy, but the system as a whole, the combined system is much more efficient, it’s better, and it’s really good for the economics of the company. 

John Koetsier: I agree with you in what you said that it’s not an instant dislocation. It’s not a black and white, we’re here today and, boom, we’re fully automated tomorrow. I think it is a transition period.

I’ll give you a real-world example. And again, I’m for automation. I’m for robotics. I think we need it, but I think that it’s going to be extremely socially challenging.

Here’s a real world example: garbage pickup and recycling pickup and compost pickup in my neighborhood — I live near Vancouver, Canada, we get all three — it used to be a truck with two guys and a driver. So, three people, drive around the neighborhood. The two guys on the back would go around, pick up whatever needs to be picked up, throw it in the back, there you go. About half a year ago, switched to a semi-automated system where there’s a robotic arm that comes out the side of the truck, grabs a container of a certain size and color, and dumps it in the truck. Now you have one person working where there were three.

Again, good thing, but isn’t it a social challenge? 

Daniela Rus: Well, I think that the example you have picked is really a great example, because the world really needs good garbage management…

John Koetsier: Yes.

Daniela Rus: …and it needs recycling. But those jobs are really difficult and they’re not pleasant jobs. There is a lot of turnover in those jobs. So, here’s another way to think about it: suppose that the human is actually not there sorting as the various items come down the conveyor belt to separate …

John Koetsier: That’s huge.

Daniela Rus: … plastic from glass, from paper. Suppose there’s a robot or a team of robots doing that. Well, what’s wrong with that? Well, what’s wrong with that, is that you can’t count on a hundred percent performance from those robots. In fact, our knowledge of doing this kind of challenging manipulation is evolving, but it’s definitely not a hundred percent.

John Koetsier: Mm-hmm.

Daniela Rus: But what if you have a service center with humans who know how to operate the robots? So, instead of having humans by the conveyor belt, with masks, sorting through… 

John Koetsier: All the stuff.

Daniela Rus: … sorting all that stuff. Perhaps the robots do their thing and when they get stuck they can ask a human for help. And at that point, the human can use the most extraordinary gaming skills that the humans might have to actually drive the robot and teach the robot to recover from the state where it’s stuck and to learn something new, and all of that input can be used to make the robot increasingly more capable.

And so now you have actually solved the problem that is critical for humanity.

People can participate and they actually need good gaming skills in order to do that. And machines can do the unpleasant part. So it’s back to what I was saying, that it’s all about augmenting people with teams of robots. It’s all about finding ways for people to do things that they enjoy and for the robots to do the parts of the job that are difficult for people to do. And if you think about it in every job, people spend time doing different kinds of things. They spend time applying expertise, they spend time interacting with each other and communicating. They spend time doing data tasks. They can spend time doing predictable physical work and unpredictable physical work.

Where the technology is right now, the sweet spot, is in automating some of the data tasks and in automating the predictable physical work tasks. So for all the others, you can have the people be in charge or you can have the people be augmented with enhanced tools over what they have today.

John Koetsier: I like that a lot and I agree with you. I do think that as we augment tasks with robots we will need fewer humans doing the actual task. And the challenge, societally and economically, will be to expand the pool of what we do so that humans can find more stuff that is fulfilling, valuable, provides a service, an income — perhaps a universal basic income at some point — and there we go. We have to draw this to a close. It has been fascinating, and super interesting … 

Daniela Rus: I do want to … hang on John, before you go. I do want to say something about that if I may. 

John Koetsier: Yes, of course. 

Daniela Rus: What I want to say is that it’s much easier for us to think about what will go away than to imagine what will come.

John Koetsier: True.

Daniela Rus: And I will give you an example. I still remember the year 2000 [laughter]. Two decades ago, in the year 2000, there was a computing boom and it was an extraordinary time for computing.

But nobody at that time predicted the explosive growth in social media, in cloud computing, in a smartphone industry. In the year 2007, we had three critical events. The first iPhone was introduced, cloud computing took off, and social media took off. Those fields have created so many jobs. They have created so many opportunities, and not just opportunities for the geeks, they have created a wide range of jobs. 

John Koetsier: Influencers! [laughing]

Daniela Rus: Not just the influencers. 

John Koetsier: I know.

Daniela Rus: So …

John Koetsier: Social media managers. 

Daniela Rus: Exactly. There are so many new jobs that didn’t exist before 2007 and we didn’t anticipate in 2000. So, I think that people have tremendous imagination and we have tremendous aspirations for improving our lives as individuals, as a society, and as a planet. And I have great faith that we will continue to do that at scale, provided that we get these smart tools that many of us are aspiring to develop. 

John Koetsier: I like that. And it reminds me of a quote …

The Steve Jobs Memorial website just went live today, actually, as we’re recording this September 8th, and put on by his wife and Tim Cook and Jony Ive, I believe … and one of the quotes that I saw there today was, “Have faith that all the dots will connect.” And what you’re saying is have faith, automate, go for it, all the dots will connect and things will happen.

I want to end with one final question. When do you think we will have a general purpose robot that is useful in, let’s say, people’s homes… let’s bound it by that. When do you think we’ll have a general purpose robot in people’s homes that can pick up my garbage, that can do my dishes, that can feed the cat, those sorts of things, you know, make the beds, all that stuff?

Daniela Rus: Well, John, I can’t really make predictions, but I could tell you how I would like for things to be.

So, I actually think that before we get that general purpose robot, we will have a garbage bin that could take itself out and back into the garage. We will have robots that will entertain your cat while you’re away. And we will have robots that could do more for us in the kitchen with the prep work. But these will be all very specialized and optimized machines. To put it all inside a humanoid machine that is able to do all of that will take a long time.

John Koetsier: Very good. Thank you so much for your time, Daniela. I really do appreciate it. 

Daniela Rus: Thank you. Thank you very much. Thank you for having me.

TechFirst is about smart matter … drones, AI, robots, and other cutting-edge tech

Made it all the way down here? Wow!

The TechFirst with John Koetsier podcast is about tech that is changing the world, including wearable tech, and innovators who are shaping the future. Guests include former Apple CEO John Scully. The head of Facebook gaming. Amazon’s head of robotics. GitHub’s CTO. Twitter’s chief information security officer, and much more. Scientists inventing smart contact lenses. Startup entrepreneurs. Google executives. Former Microsoft CTO Nathan Myhrvold. And much, much more.

Consider supporting TechFirst by becoming a $SMRT stakeholder, connect to my YouTube channel, and subscribe on your podcast platform of choice: