Podcast: Vibe Manufacturing: MIT's AI Text-to-Manufacturing System

In this episode, we explore MIT’s new system that turns spoken ideas into physical assemblies by linking speech-to-text, text-to-CAD, CAD-to-assembly, and robotic fabrication - pointing to a future where making hardware could be as easy as prompting an AI with your voice.

This podcast is sponsored by Mouser Electronics. 

Episode Notes

(4:34) MIT researchers “speak objects into existence” using AI and robotics

This episode was brought to you by Mouser, our favorite place to get electronics parts for any project, whether it be a hobby at home or a prototype for work. Click HERE to learn more about the role of AI in product design.

Become a founding reader of our newsletter: http://read.thenextbyte.com/

Transcript

From the replicator in Star Trek to the microwave that made McDonald's and Spy Kids, the people have yearned for an engineering appliance that could make them whatever they wanted, despite them lacking the skill level to make it themselves. And I am honored to be the one that delivers the news. MIT might have just cracked this. So, if that's got you excited, let's take a bite out of this episode and get into it.

What's up friends, this is The Next Byte Podcast where one gentleman and one scholar explore the secret sauce behind cool tech and make it easy to understand.

Farbod: What's going on, friends? Welcome back. Like you heard, we're talking about product design, which is probably Daniel and I's favorite topic. Like, obviously, we like robotics. We like materials. But the combination of it as it distills down into product design, that's our favorite bit. But before we get started with today's article, let's talk about today's sponsor, Mouser Electronics. Now, if you've been rocking with us all about 2025 or really any time in the last five years, which is a crazy milestone we're coming on, you know we love Mouser Electronics because they take really interesting information and they distill it down in a way that's really easy to understand in all these technical content resources that they have. So sometimes they write journal articles, sometimes they create videos. The one that I want to talk about is a journal article that we're going to link to the show notes where they're talking about the role of AI in product design. And Mouser has this tendency of generally looking into this crystal ball, and the crystal ball is really just composed of the academia and industry partners that they have, and predicting things pretty spot on earlier than anyone else. So, they have this interview with a McKinsey advisor who has 15 years of experience doing product design. This person has over 30 patents, has used their genius and their innovation to save companies over $500 million over the years that he's been working. And he's giving all these interesting statistics based on the analysis, the market trends that they're seeing with AI with respect to product design, saying things like it could unlock $60 billion of potential in the R&D space, and it could reduce product development cycle by 70% or more. And all of that is super, super relevant to what we're talking about today.

Daniel: I will say, I typically have a cynical outlook toward consultants and advisors and stuff like that. Appreciate that this is someone who has experience actually doing and they say like the bad phrases like those who can't do teach in my mind. It's like those who can't do consult sometimes. And this is just me like super strong bias, but…

Fabod: I'm not cynical but.

Daniel: Yeah, I am cynical. But this is a really, really good interview. And I appreciated the pragmatic approach, looking at how AI can unlock different potential in product research and design, but specifically how AI can unlock different steps in product research and design. And then obviously, looking at the potential implications of reducing product development cycle times by 70 plus percent. That means engineers can iterate faster. That means you better products faster. That just means the rate of improvement in the world around us. And again, my other bias here is engineers are at the center of the universe and engineers run everything.

Fabod: A lot of hot takes today.

Daniel: Yeah, a of hot takes there to unpack. But if you make it faster for engineers to solve problems and do them in a time-effective, efficient manner, we will see a lot more improvement in the way we interact with the world around us.

Fabod: Less than three minutes into the episode and you're already dropping these hot bombs.

Daniel: I know. Hopefully, I haven't lost a couple of listeners already.

Fabod: If they weren't tuned in before, they're tuned in now.

Daniel: Yeah, exactly.

Fabod: But what I was going to say, another thing that stood out to me is in a world where when it comes to AI and engineering, let's say 90% of the focus is on software, Mouser actually took the time and focused on something different. Not only is that refreshing, but it gives an interesting bit of insight into a world where it seems like everybody has television. So, seeing a focus on product research and design, not just in the software realm, but in the physical realm as well, super interesting. And again, highs really nicely into today's article. And speaking of which, that's going to be us going to MIT. That reminds me also, we have content from MIT we haven't released yet. Hopefully, we'll do it first thing in 2026, but be on the lookout for that. But bringing it back to today's episode, MIT, they have this speech to reality. I don't know what else I would call this. Speech to product reality AI and robotic combination system. And the premise here was, you know, we've gotten used to going to our LLM agent of choice, let's say ChatGPT, and prompting it to do various things like plan me a trip, get me a list of restaurants to go to tonight with my girlfriend, whatever. And we've been able to automate a lot of these processes.  And it seems like we have various components of a bigger puzzle which is automating product development from scratch. So, imagine if instead of you having an idea and then having to go to your CAD software and drawing it up and then taking that drawing and then putting it to your CAM software and taking it from the CAM software and giving it to a manufacturer and having the manufacturer put it together and then assemble it and then give it back to you. What if we could just cut everything else out and you just give the idea and what you get delivered is the final assembly from a robotic system. That was the idea.

Daniel: Yeah, it's awesome. As someone who's been excited about accelerating product development cycles, if you couldn't tell from the early in our episode, looking at the stack, if we start with the physical assembly of what you already want, we already have robots that can do physical assembly. We already have CAD programs that can turn 3D geometry into a set of instructions for robots to assemble. We already have programs that can help us take text and turn them into geometry that that CAD program can understand. And we even have engines that can take speech and turn them into text. So, it's like we've got this whole chain of things that haven't yet been integrated, but we do have speech to text. We do have text to CAD. We do have CAD to CAM and we do have CAM to actual assembly. Those all exist in fragmented systems. Here's someone who's doing the work to integrate all of that and do some research on how we can take speech and just turn it straight into reality.

Fabod: And it's this glue that's fusing all these different pieces together to create something meaningful. And I guess we gave the precursor. Let's dive into some of the actual components of the system to begin with. You went from the robotic system up. I'm going to go reverse. So, at the very beginning, you have user speech coming in. So, Daniel and I saying, hey, I want a table with a single leg created. And then that voice has to then go through a natural language processing system. Odds are you've used those systems on your daily life. So, if you're using Alexa, if you're using Siri, we did an interview with Assembly AI. They're powering the majority of these platforms from Google to Gemini.

Daniel: If you’re reading the notes, show notes or transcript from this podcast episode.

Fabod: There you go. That's what it is.

Daniel: That’s the natural language processing.

Fabod: It takes the raw audio data transcribes it into text. And then exactly like you were saying, now we can take this text information and create some sort of a geometry, a representation for our CAD software to try and understand. So, it takes single-legged table, creates the mesh, the general form of it into the 3D space, gives it to the CAD program. The CAD program then dimensionalized it, make sense of it. That then goes into this discrete component system. And this is kind of like a piece of their secret sauce where imagine Lego blocks, but the Lego blocks have magnets on all sides so that the robot can pick and place these almost as pixels to build whatever geometry it wants. It's leveraging whatever the CAD system is outputting to come up with what it would look like with this Lego system. And then it goes into another system that says, cool, now we have all these pieces, but we have to consider physics as well. Like we're building this thing in the real world. Just like you would build Ikea furniture, you can't start in the middle, go to the top, and then do the bottom at the end. Or maybe you can if you're a wiz, I don't know. But the idea is that it could fall apart. So, this assembly sequence program tells the robot in what steps everything should be assembled for it to have the structural rigidity it needs. And then finally, it goes to the robotic system that brings it all together. It follows the steps, and it gives you what you were looking for.

Daniel: One of my favorite things about all of this is like, when you talk about discretizing stuff into the components, that's a part of the design process, but it's also a precursor for the assembly process. And so, where design and fabrication are being solved together, this engine only generates objects that are actually buildable using their components. So, the geometry is like real time adjusted for the real-world constraints. This is in contrast to the way that we're doing a lot of design and assembly today with humans is like as a human as a product designer I might make a certain design and then I send it to my manufacturer, my manufacturer says I can't do that. And then they send feedback back and say fix it this way. So, then I fix it make some tweaks then I send it back to them and they say okay I can do that, but you need to change this this and this. There's a lot of like iterative cycles right now between manufacturers and designers. And they say the best designers are people who have done manufacturing before. So, designers who are designing with the context of manufacturing. And they say some of the best manufacturers are people who have done design before, so they can manufacture with the context of design. In this case, they've made their AI process work such that design and fabrication are solved together in one step. So, you're actually cutting out a lot of these potential cycles where you have to iterate and go back and forth and not back and forth. But also, as a designer, imagine having the ability to basically one shot your design and you only create objects that are actually manufacturable and then imagine as a manufacturer only ever getting designs that are actually manufacturable, it makes it a lot more efficient. And that's something that's exciting to me for when they start to add more sophisticated methods of manufacturing into their arsenal. So, like right now, like you said, they're using basically the coolest analogy is like little tiny blocks with magnets on each side so that they can build something pixel by pixel by pixel. When they get into the point of being able to integrate other real-world manufacturing systems and real-world assembly systems into this arsenal, if they can keep this process where design and fabrication are solved together at the same time, I think this is where it'll really do a lot of the heavy lifting.

Fabod: Totally agree with you. I mean, as is, this isn't something that I would want to have in my home as a consumer or in my workshop as a professional. But imagine this thing where you cut out these magnetic Lego pieces and you introduced, for example, additive manufacturing units where components can get uploaded into those units, parts are made, the robot can grab it, and of course it still has the assembly requirements so it knows how to piece them all together. That is far more interesting to me. And I think the hardest part in these frameworks is always those first baby steps to get something off the ground and then let the world iterate on it. And that is how I'm approaching this and perceiving the accomplishment of this team is that they've shown that these various  siloed systems can be brought together and we can add that glue and here's how and now the rest of the world, the manufacturing lovers like you run wild with it and iterate and let's see what it looks like in the next year or two with various manufacturing equipment included.

Daniel: Yeah. I like the idea right now of this fast, accessible, low-waste fabrication where they can reuse these models. I'm going to be super excited when they get to a point of using multiple robots to do assembly, using more sophisticated inputs from me as the user, and then obviously more sophisticated outputs in the manufacturing realm. We'll get to a point where you basically get to do physical reality completely on demand the way that like right now with LLMs and generative models, like you can be like, oh, like make me a picture of a moose standing on a mountain surrounded by flowers. And it will do that. Just imagine being able to be like, I don't know, Jarvis make me a water bottle and it makes me a water bottle. And it's like, oh, that's, that's cool.

Fabod: Yeah. But I was trying to think of how else I could connect this to the general audience. Again, unfortunately, I don't think this is going to be something you're gonna see in your home anytime soon. But what it could unlock is, quote unquote, vibe coding has enabled a lot of people who might not have had a programming background to easily step or dip their toes into this world of programming and create. And even if it's low fidelity, it gets them something that they can hand off to someone else so that their vision becomes real and it's easier to explain what they wanted to make. And the cost of doing software has now come down quite a bit as well. I think a system like this will help us inch forward toward that same vision, but now in the physical world, just like at-home 3D printers did in the last 10 years, 15 years or so.

Daniel: And then the big sci-fi future version of this, they mention it in the article, is like the replicator in Star Trek or the robots in Big Hero 6. The main researcher says this is a way where you simply have to think of the reality and be able to communicate it out loud. And then he's working towards a future where you can create a physical object basically just by imagining it, which is really cool.

Fabod: Did you ever watch Spy Kids?

Daniel: Yes.

Fabod: The second one where they have that microwave where they press the button.

Daniel: Yeah, we've talked about this in the podcast.

Fabod: It's just so stuck in my mind that every time we talk about something like this, that's where my brain goes. Yeah, that's the future I want to be in.

Daniel: Well, I mean, we've seen cooking robots before. And like the discretizing components part of this is actually quite easy because there's a lot of like learning out there in cookbooks to understand like the correct ratios of different ingredients. It would be cool to see them hook up a fridge to a stove and say like with all these robotic arms and be like, yo, I want a cheesecake and just watch it go to work making a cheesecake.

Fabod: That would be incredible.

Daniel: Cool.

Fabod: One day, soon, hopefully. But I think that's a good place to end the episode on. Just a quick wrap up. Basically, we've all dreamed about these replicators that we saw in Star Trek, or if you're me, the microwave that makes McDonald's from Spike Kids. And the folks over at MIT think they've cracked the system. They've taken all these achievements in engineering and computer science over the last two decades or so, where we have robotic assembly. We have text to CAD, and then we now have speech to text and LLMs, put it together so that as a user you can say, want something like a single leg table, and out comes a single leg table put together by a robot. What this means for today is maybe not so significant, probably nothing you're gonna get in your home, but what this means for tomorrow or the next year, the next five years, is very likely a streamlined method of manufacturing that will do what AI did to software, but now for hardware for the average Joe. So super exciting.

Daniel: Love it.

Fabod: Thanks folks.

As always, you can find these and other interesting & impactful engineering articles on Wevolver.com.

To learn more about this show, please visit our shows page. By following the page, you will get automatic updates by email when a new show is published. Be sure to give us a follow and review on Apple podcasts, Spotify, and most of your favorite podcast platforms!

--

The Next Byte: We're two engineers on a mission to simplify complex science & technology, making it easy to understand. In each episode of our show, we dive into world-changing tech (such as AI, robotics, 3D printing, IoT, & much more), all while keeping it entertaining & engaging along the way.