Multi-user VR experiences with A-Frame (2017) | Hacker News Reader

Multi-user VR experiences with A-Frame (2017)

hacks.mozilla.org

121 pointssp33232 comments

I think beyond even VR, just having this standard markup language for building 3D scenes is amazing. My biggest gripe over modern game engines is the lack of a programmatic means for building scenes. You always end up with massive proprietary binary “scene” files that are a pain to put in VCS.

This is a necessary evil unfortunately. When loading 3D content from disk, every cycle counts. Therefore storing and parsing this content in plain text is just too inefficient. A number of 3D scenes are quite large in size as well, and benefit highly from compression.

This is admittedly a naive question— but wouldn't something between the two work well? (A-Frame is arguably akin to that already)

A 3d markup language that can be compiled? At least for prototyping, it might be useful, no?

Does that exist? Or is it a problem that's already been tossed around and buried?

This exists, it’s very common to have a human readable format for development and storage in version control that gets converted to a binary format for release. What doesn’t really exist is a common inter-engine format.

VRML ?

https://en.wikipedia.org/wiki/VRML

Lots of common formats exist for scene data just very few of them are actually used by game engines in a manner that would make them somewhat interoperable.

Probably most the most commonly shared is something like the Tiled TMX format for tilemaps.

I guess Blender Python scripts come kind of close to this, but that's not really a markup language. (And 3D Modeling not a game scene file)

How about glTF? It's a JSON based scene description which uses inline data or references to external binary files, with an option for a single binary "glb" data file.

https://github.com/KhronosGroup/glTF

https://github.com/KhronosGroup/glTF/tree/master/specificati...

glTF is already gaining some traction as a model format. But it’s scene support is more what you’d see for modelling software than what you’d have for a game engine where the scene description is more about the game entities and the data glTF provides is usually abstracted out.

glTF is a great idea for providing a sensible interchange format for models though. Even this data usually gets converted into a custom binary format for the runtime engine to consume depending on use. Although as you note glTF has its own binary format and I understand it’s going to be improved with support for compression like Basis so perhaps that will reduce the need for custom formats for all but the most optimised cases.

You're probably right, at least for the near future. There's another A-Frame article about how to make an immersive game with low-poly models. https://hacks.mozilla.org/2018/03/immersive-aframe-low-poly/ If you're really pressed for size though, you can combine binary coding on disk with A-Frame, as Platane did in Vernissage. https://hacks.mozilla.org/2018/01/lessons-learned-from-the-a...

Why not just compile the thing? Textual source code for authoring, optimised binary for execution.

We do this for executable code, is there any reason why we shouldn't do this for assets as well?

I’ve worked on 3D pipelines that did just that. The text format was even, gasp, XML! As long as the large arrays were large strings of numbers and not individual nodes per component, it was actually pretty efficient and very fast to convert.

This is essentially what three.js does with WebGL content. Using JSON;

https://github.com/mrdoob/three.js/wiki/JSON-Geometry-format...

Everything old is new again...

https://en.wikipedia.org/wiki/VRML

Hey, one of the A-Frame creators here. Looking past the markup, A-Frame is closer to a Unity-esque web framework (based on ECS) than a standard declarative file format. An ECS-based three.js framework that has some DOM flavor. Developing with it sort of feels similar to Vue.js.

Unity scenes are stored as yaml in the project, and are only baked to binary for built executables.

Time is a flat circle. I remember being obsessed with VRML in 1997 and specifically there was a site called CyberTown which you could chat in a 3d world.

Thanks for posting! networked-aframe is great, can't wait for it to continue to improve.

We're the creators and maintainers of A-Frame and just went through the last YC batch as Supermedium (https://news.ycombinator.com/item?id=16318151). Let us know if you have any questions!

> Compatibility > The sharedspace component requires a browser with WebRTC support which means any recent version of the most popular mobile and desktop browsers.

Just a cautionary note... let's see, how to put this fairly...

When Mozilla announced the general availability of WebVR, they meant on Windows. Around that time, WebVR on Linux had achieved a milestone... of sometimes crashing immediately after starting, instead of always crashing before. It's been described as 'caring so little about Mac and Linux, that even the awareness of non-support has been limited'.

And even on Windows, Mozilla described WebVR as ready for use something like a year ago. And yet there's a thread here today, about trying A-Frame on Firefox, and being made sick, and this being a known problem.

So least at some points in the past, Mozilla's blog posts about WebVR's usability, have been... aspirational. Motivational. And not full and fair disclosure.

This is awesome! Somebody just a few months ago show us how they built a WebRTC multiplayer VR A-frame game.

The got the 3D world to permanently persist too, demo here: https://twitter.com/marknadal/status/962013903004684288

Ready Player One's "Oasis" seems to already be real!

Thanks! We created A-Frame and are now also building Supermedium (YC W18), a fully VR browser. If you're interested in RPO, we build an OASIS in a week (with A-Frame) featured on Supermedium: https://supermedium.com/blog/oasis

People hunt for an egg among hundreds of generated worlds to receive a prize and access to the code ("take control of the OASIS"). Though someone found it yesterday.

Anyone know how a-frame compares to OpenCroquet/Cobalt/Qwak/etc. ?

The latter seems to be way more developed, with P2P networking, "3D hyperlinks", integration with XMPP/VNC/etc., integrated IDE for live-editing, etc. Those were all in Smalltalk rather than JS, but there are also JS ports/successors like Virtual World Framework, 3dicc, etc.

A-Frame's just a general ECS-based JS/VR framework on top of three.js with access to any Web API. It's extensible and modular so P2P networking (or Websockets or Firebase) have been done as community components. The community is active, and it mirrors web development (versus like trying to do a Unity in the browser).

3D "hyperlinks" have been done, they can take whatever shape or interaction you want (e.g., portals https://supermedium.com/blog/oasis), just have some event trigger a window.location change.

Just last night I tried using A-Frame with a Vive but was disappointed by the performance. The framerate is fine but there seems to be a tiny lag, between the time when the headset position changes until the worldspace camera is updated. Made me feel a little nauseous.

Yeah, it's the latency in the Firefox browser on Vive. Not with A-Frame, but there's some delay communicating with the SDK and transferring that data specially on SteamVR.

Perhaps try narrowing the optical field of view? You might test by holding the HMD away from your face. Or by unlocking the Vive side disks and adjusting the display outward. Or by using a thicker pad. A comfort-mode-like restriction of peripheral vision.

If it's your own page, you might try putting the Vive camera video in background. Doing passthrough AR, I regularly used Vive (and now WMR) down around 30 fps. But I displayed the browser directly in the HMD. So no WebVR display latency. And I don't do lens correction (old laptop with integrated graphics), which has a bit of a "comfort mode" effect. And no predictive tracking, so the rendered graphics have lag, but no judder. And displayed directly, it was easy.[1] So YMMV.

Depending on your level of desperation, there's my largely-abandoned WebVR-like stack for Vive on linux. But I didn't get around to supporting A-Frame, just Three.js. But by keeping it simple, some things the complex native WebVR stacks have struggled with these last 2 years on Windows, were much easier.

[1] https://cdn.rawgit.com/mncharity/node-webvr-alt-stack/master...

For small values of "Reality".

There are many far better multi-user VR experiences. Sansar. SineSpace. High Fidelity. Even Second Life. It's easy to do the simple demo, and really hard to do a full large virtual world.

Definitely multi-user tooling is still early, and the blog post was meant as a simple demo, but rapidly improving. At least "bunch of people in a world" is extremely feasible (syncing heads/hands). The community is still growing too. But I think there'll be more unique multiuser experiences coming from the Web versus Second Life recreations owned by companies.

The actual world creation part isn't too bad. We built hundreds of generated interlinked worlds and dropped in multiuser in a week (https://supermedium.com/blog/oasis). But still improving.

Are you an active user in those applications? What's your favorite part about them?

VR social experiences are by far the best ones.

Once eye tracking is implemented, social experiences will take on a new level.

Eye tracking is already there.

Tobii showed vives retrofitted with eye trackers, and snapdragon 845 vr devkits with integrated trackers.

I tried both and it works really well.

I'm not so sure that it will completely solve the social interaction problem though. I've tried the demo where you can see your avatar in the mirror and observe the eyes move, but there was still something off about it. Maybe it was just the art style, but I suspect that eyeball movements are not the only part of equation - there's a lot of very mobile musculature around the eyes that tends to change the shape of the eye as people make expressions with their faces - and that part was not there in the demo.

Does anyone have any thoughts on JanusVR (which also offers multi-user VR web experiences)?

VRchat is basically the same thing but better. Janus has its own kind of markup language which is pretty cool but nobody has done anything huge with it yet. If it was developed to fill in where VRchat doesn't, it may be better. the decentralized nature is pretty nice.

Reply on news.ycombinator.com