Interesting notes. With regard to presence creating a feeling of isolation, that's not necessarily a bad thing. I agree that it's worth looking at ways that it can be managed (try very low, continuous white noise, especially using indistinct voices) for cases when you don't want it, but in cases where VR is simply being used as an interface, the feeling of being around other people may be distracting. For instance, if I'm visiting a virtual mall then I want to feel like there are other people around me, but if I'm using a virtual store interface I want to have that solo experience. It may be that this is also a matter of preference. Some people panic when they feel isolated, some people relax when they feel isolated. It may be that allowing the user to tune compensatory behavior is the best route for now.

Testing all of these assumptions out is why I'm building a framework for building VR applications in modern web browsers: https://www.github.com/capnmidnight/VR

It's been fascinating to try things that I thought would be awesome and see them not work out at all, while also trying things that I thought were not going to be a big deal but turned out to be hugely important.

For example, I thought speech commands were going to be a really big deal, but they turned out to be almost completely useless. It's really difficult to get a speech engine to recognize very short phrases, and longer phrases are cumbersome. It just doesn't make for anywhere approaching a good experience.

And then there are the weird things, like combining actions from one hand over the Leap Motion with mouse motions in the other hand. I think it's because the mouse allows me to make very broad movements and the Leap allows me to make small ones, and the two combined cover areas that the other doesn't do well.

Anyway, really fascinating stuff.

From my brief VR experience I noticed many commonalities (and some differences) with your experience. I did a rollercoaster (no, not THAT one). Having a seated position matching the real-world seated position was hugely helpful for combating simulator sickness (as compared to some of the awful FPS attempts). The frame rate was also hugely important. Graphics features that can be turned on/off were vastly context dependent. I managed to turn off shadow receiving for everything except the coaster cars without any drama. But turning those last shadows off drained it of life - the flickering superstructure shadows are what a rollercoaster is all about. I agree that distant vistas are curiously dull. I get that binocular vision doesn't help much in this case, but sight-seers everywhere love distant mountains and oceans. I found that particle effects were good to provide a spatial sense, e.g. floating dandelion seeds. I'd love to incorporate avatars more. Do you think fake avatars would work at all if the player lacks a kinect/hydra/leap? For example infer the position of arms, show them holding a controller, when a button is pressed show the fingers pressing the buttons? It may be super-weird. Such an experimental field right now!

@jefferytitan: I don't think the 75fps framerate target that Carmack talks about is strictly necessary, but it definitely needs to be at least 60fps. But what is surprising is that scenes don't need to be extremely detailed to achieve presence. I would say it's better to achieve high framerate than to achieve high polygon count, high res textures, etc. We used to play 3D games that were nothing more than billboarded sprites and were fine with it. Something like that would still work in VR, as long as the framerate and interactions were smooth.

On avatars, I've been using a model of a fat, cartoon bear that I threw together with a simple running animation. The arms are not rigged to my Leap Motion yet, but you mostly can't see the arms. Having a body and feet to see, though, has been nice. I like it, at least. YMMV.

However, with the Leap, I've noticed that it absolutely *is* necessary to have a fully rigged hand model. Simple spheres for each of the knuckles works fine, but a single sphere for the entire hand is not good enough. However, I'm having some network performance issues with the Smartphone HMD sync in Psychologist.js, so I might have to degrade the visual in that case (i.e. transmit only the hand's location and not all of the individual finger joints).

But without direct, kinematic sync with body motion, it's best to show nothing. I've found any sort of user-oriented movement that was not triggered by the user's actual movement to be the most likely to cause motion sickness. This is where I refer to presence being a double-edged sword. If you haven't achieved presence, the user (or rather, me) still feels like they are looking at a screen and non-didactic movement isn't disorienting. When you get the user feeling like they are in the scene, they need to be in complete control of what is attributed to them.

So in short, if you do use some sort of fake hand model, I'd try to keep it abstract. Show the model of the tool held in the hand but do not show the hand itself, and keep movement to a minimum.

@cozzie: I have the Samsung Galaxy Note 3 for doing the Smartphone-style HMDs, and a cardboard box that I built myself for holding it, with a strap I cribbed from a pair if head-wearable magnifying lenses (but not the lenses, I bought lenses specifically on Amazon.com). It's not Google Cardboard, because Google Cardboard was announced the day I ordered the lenses.

I also just acquired an Oculus Rift DK2. I spent a little time last night and most of this morning getting it to work with Psychologist.js (okay, I officially hate that name now, it's too long). It's up and running now, but requires specially development builds of Chromium or Firefox.