You’re reading Engineering Paper, and here’s the latest design and simulation software news.
In response to a recent edition of this column, The agentic era for mechanical design, a reader named Hugo Manning reached out to share his experience with agentic AI. He works for Spare Parts 3D (SP3D) as a lead product manager for Theia, an AI-assisted tool that turns 2D drawings into 3D models.
The premise isn’t unique. Earlier this year, for example, Dassault Systèmes announced that Leo, one of its three virtual companions, could turn PDF drawings into 3D models in xDesign. Another example is Ortho2CAD, a vision language model that generates CAD models from orthographic drawings. Yet another is gNucleus AI, which, among other AI capabilities, advertises an image-to-CAD feature.
The list goes on, but Manning believes Theia stands apart from its competitors.
“The solution we offer is a fully validated file… an engineer is going to look at it and make sure that all the dimensions and features are correct,” Manning told me. “That’s our value proposition, as opposed to what most of the market offers, which is to plug it into an AI and see what the AI gives you.”
Theia, then, is not a standalone AI tool. Rather, it’s a service that happens to use AI. “AI handles geometric extraction and pattern recognition at speed, engineers handle validation and edge cases,” reads the website.
As long as the 3D model is accurate, whether it’s created by an AI or human doesn’t much matter from a user perspective. You simply upload a drawing and, sometime later, receive a 3D STEP file.
The time and cost depends on the complexity of the drawing, which is automatically determined by the number of dimensions, views, types of geometric features, and so on. The price is shown to the user up front and ranges from €9 to €200 ($10.50 to $233) per drawing, paid in credits purchased with a monthly subscription.
Unlike conventional AI tools, Theia isn’t instantaneous, since everything is human-validated.
“About 75% of the parts we do we return within a day,” Manning told me. “Our perspective is we’re going to take longer, but we’re going to provide it at a much higher quality and maybe a higher cost.”
You can find more information on Theia’s website.
Hololight’s spatial computing magic
Spatial computing is a technology that seems perpetually on the brink of breakthrough. The various flavors of this technology, such as virtual and augmented reality (VR/AR), have made steady progress over the years. There are many examples of engineers using spatial computing, and many companies investing in the tech. Yet spatial computing has never grown as much as its promise would suggest.
One developer working to solve that problem is Hololight, which provides an enterprise extended reality (XR) streaming platform. I recently covered an update to that platform which aimed to make XR design reviews faster and more stable. I’ve since had the chance to speak with Florian Haspinger, Hololight’s founder and managing director, to learn more about it.
“We engineered a lot of stuff on a computer, and it was always painful to switch to the 3D model on the 2D screen,” Haspinger recalled of his motivation to start Hololight back in 2015. He and three co-founders decided that augmented reality was the way forward for humanity, and that the way forward for Hololight was pixel streaming.
“We outsource the compute to a server,” Haspinger explained. “We render everything on the server side with big GPUs, and then we pixel stream it to the end device.” Without having to handle heavy computation, XR headsets can be lighter, cooler, and more battery efficient.
Hololight works with headset manufacturers to integrate its pixel streaming technology with their devices, aiming for compatibility with as many devices as possible—including tablets and web browsers.
“We built a Snap client, a Quest client, HTC is part of it, Pico is part of it, Apple Vision Pro, HoloLens One, HoloLens Two… Even on iPad and Android devices, it works pretty well. A client device could also be just a browser in Windows or on a Mac, so you can just stream into the browser of your device as well,” Haspinger said.
The biggest potential drawback of pixel streaming—for spatial computing or any other reason—is latency. If there’s even a small lag between action and response, the application can feel choppy or even nauseating.
“We’re not magicians. We can’t reduce latency,” Haspinger said.
By that he means network latency, which depends on multiple factors, such as the distance between server and client, that no application can change. As for what Hololight can control, Haspinger says “the latency we add from our system is very minimal.”
They may not be magicians, but Hololight has a trick to deal with high network latency. The platform predicts the user’s trajectory and calculates the frames it should show after the network delay, allowing the headset to project its best guess at a properly-timed image.
“Even with 200 milliseconds of network latency, you still feel a very immersive experience,” Haspinger said. “So you don’t get motion sickness. You don’t have the feeling it’s laggy. We can give you an experience that feels real-time, even with the latency on the network side. That’s one part of the Hololight magic.” (Maybe they are magicians?)
I’ve long been interested in spatial computing, so I’m glad to see progress—no matter how slow—with the technology. If you’re using Hololight specifically or spatial computing in general, I’d love to hear about your experiences at malba@arrowfly.com.
One last link
Engineering.com contributor Peter Bilello explains why you should Stop measuring product development progress in dollars.
Got news, tips, comments, or complaints? Reach me at malba@arrowfly.com.
The post Drawing-to-CAD AI with a human twist appeared first on Engineering.com.