Envisioning a DIY CyberCarpet for Omnidirectional Motion

[Synexious]

The CyberCarpet is the only omnidirectional stationary motion system that is anywhere near practical for consumer VR. It will most likely be decades still until omnidirectional treadmills are affordable, and the VirtuSphere is too big and expensive, as well as technically suboptimal (I think walking on top might be better, but that would require even more space!). There is also an idiotic tile-based system, the CirculaFloor (perhaps the tile idea could be made to work, but it seems unlikely - maybe with a square gridline? The current system is far too slow. The OmniWalk hexagonal system is interesting...). That leaves us with one last hope, else we're restricted to bidirectional motion on a conventional treadmill, or running in place with an accelerometer. With this thread, I would like us VR enthusiasts to assess the feasibility of constructing our own CyberCarpets. CyberCarpets are featured in the Michael Crichton novel/film "Disclosure," and have been realized by a research team at Max Planck. They consist of several thousand (I believe the prototype contained 4000) balls held in small wells. The problem with the prototype is that high speeds cause the balls to pop out. To prevent this, I've envisioned two potential solutions, a) place a second layer over the balls, with holes slightly smaller than the diameter of the balls. Hopefully there would be enough surface protruding to walk on. And b) use magnets. I don't know if magnets would affect walking, or if they would even work to keep the balls in place well. Solution a) seems to be the better option to me right now, though 2:07 in the first video gives me pause, as well as the fact that it seems so obvious yet wasn't implemented. So perhaps magnets, after all... how do they work, again?

That's it. It's really quite simple compared to the other options, and I believe it could be done for less than $5000. Perhaps significantly less, I don't know. What do you think? Is that a realistic price point? What materials would we use, for the balls and the plates? Plastic? Metal? Wood? A belt actuator, basically a bidirectional, square treadmill, was also added to the prototype. It turns 360 degrees, tracking the user. I'm not entirely sure what effect this would have, beyond forcing continuous motion. It would obviously drive the price up significantly. I'm very interested to hear the community's thoughts. Has anyone ever had a similar idea? This could be a breakthrough of HMZ-T1 magnitude, or greater. Unless we develop our own ball carpets, we'll probably have to wait twenty years at least to get our own omnidirectional stationary motion systems, and until we get those, cyberpresence will remain limited.

[youtube]http://www.youtube.com/watch?v=bdGH3eOuGIM[/youtube]
[youtube]http://www.youtube.com/watch?v=qhKsVoKsl8s[/youtube]
[youtube]http://www.youtube.com/watch?v=VFkyV7d5t8o[/youtube]
Combining a ball carpet with a system similar to this one (replace Vuzix HMD with Sony) would enable total VR (WirelessHD or similar technology could be used to eliminate the laptop, and enable near photorealistic visuals from a high-end custom PC). Everything audiohaptivisual would be covered, full range of natural motion. Potential mods to the HMZ-T1 could increase FOV, or a custom HMD with as much as 180° FOV could be made, perhaps even a Surround or 3D Surround HMD, though that would require a wired connection to a laptop.
[youtube]http://www.youtube.com/watch?v=lV3j2Yxv7jY[/youtube]
Near-photorealistic environments which you could be physically present in with this system:
[youtube-hd]http://www.youtube.com/watch?v=OBw0R1AU ... re=related[/youtube-hd]

[bobv5]

Will think about it, see if I come up with any bright ideas. Might actually be possible. ONly took the guys on th other thread three days to come up with a viable gun tracker, so who knows?

[Aeroflux]

Very interesting stuff. Maintenance of this 4000+ array would be crippling though. The problem with having a layer is that it needs to segmented to allow structural support...and then you have to deal with alignment issues.

I would say the jacket needs to cover 3/4 of the bearing and the bearings need to be packed in closer proximity to each other. Make the array segmented into modular groups so bearings can be inspected/maintained/replaced with ease. Larger arrays can be made over time from the same parts.

With a bi-directional treadmill I imagine the user will find it difficult to suddenly change direction, but not having used a setup like this I'd like to try it to see what restrictions of movement exist. You mentioned magnets...perhaps in the future we will create a magnetic treadmill that can drive all angles of movement and apply better resistance. Put a highly magnetic substance (i.e. iron fillings) in flexible container and use electromagnet drives to vary the resistance. You could use optical tracking, like a modified optical mouse to translate movement in the virtual world.

Just thinking off the top of my head...

[Synexious]

The bidirectional treadmill is square and turns 360° underneath the balls, driving them. I'm not sure exactly what benefit this provides. Why do you think maintenance would be a problem? Would something like this wear down fast?

[Aeroflux]

The bidirectional treadmill is square and turns 360° underneath the balls, driving them. I'm not sure exactly what benefit this provides. Why do you think maintenance would be a problem? Would something like this wear down fast?

I was referring to the maintenance for the bearing array. Even if you restrict footwear of the user, repeated use will eventually contaminate the bearings and cause drag on the more frequently used bearings. You'll end up with an uneven resistance over time. That's why I was thinking of dividing them into modular components. It would make maintaining the bearings easier.

[cybereality]

How about this: don't use the floor at all. THERE IS NO FLOOR. Its all virtual!!!!!

Get a harness type of thing and suspend yourself from the floor. I found some Yoga-swings that look like they will work and only $450:
http://www.yogaswings.com/2010/01/28/rock-it-stand/" onclick="window.open(this.href);return false;

[youtube]http://www.youtube.com/watch?v=cNActial8u4[/youtube]

Get a mountain climbing harness for like $60, and some straps to mount it to the swing:
http://climbhigh.com/climbing/momentum- ... djust.html" onclick="window.open(this.href);return false;




Build a sandbox that just barely touches your feet. Just enough for haptic feedback and to allow you to turn, but you are always actually hanging from the swing. Lets say that costs another $100.

In total just a little over $600. Not bad. You think this will work?



Also, how was this guy not mentioned:

[youtube]http://www.youtube.com/watch?v=UYE6dqaereU[/youtube]

[brantlew]

How about combining the strengths of both the yoga harness and the roller shoes. The yoga harness keeps you in place and upright but you don't really get to put weight on your legs and would make walking awkward. The roller shoes by themselves look difficult to balance. I can't imagine running in them. But if the harness were used to balance you and you could just run in place with the wheels then that might be a decent system. It would be even better if you could motorize the wheels so that you could use resistance to simulate different terrain. Crouching might be difficult but jumping would certainly be supported. Actually it could even handle safe acrobatics so you get do a VR martial arts game.

[Synexious]

Brant, that's a good idea! Yes, I don't see how the yoga harness would be any better than just turning in place by itself. In fact, it would be worse. But running in place with roller shoes might work. That might actually work really well, and would be the cheapest solution. Cyber, I'd actually seen the roller shoes but had forgotten about them. A strange idea I had in the past few days is to put a square treadmill on a pole. The device senses when the user is turning and turns the treadmill accordingly. That probably wouldn't work, though. I wonder how hard motorizing the shoes would be. I would want them to be quiet, how much money do you think motorizing them would cost? Where could we get comfortable, shoelike roller shoes? A ball carpet would be better, but I think the harness and roller shoes, motorized or not, would actually be the absolute cheapest system.

[bobv5]

WOuld probably need to build the shoes, rather than modify a bought pair. Wouldn't be too difficult. The hard part would be the software, motion tracking, speed control, wheel speed sensors, idk what else. But that stuff is going to be a problem with any powered solution.

[pierreye]

Just an idea. Maybe we can get kinect and mount it to view from the top down view. Then it can detect our position, facing direction and movement to translate it into WASD. I think it is quite possible as someone had manage to integrate Kinect in GlovePIE.

http://kinect.dashhacks.com/glovepie" onclick="window.open(this.href);return false;
http://kinect.dashhacks.com/kinect-hack ... y-bindings" onclick="window.open(this.href);return false;
(using FAAST to translate into keyboard macro)

Or we can mount a wiimote as top down view again and detect movement by putting some IR LED on our shoulder or leg and translate that into movement. We can translate the LED movement and calculate whether we are walking or running.

[brantlew]

Wiimote solution I feel like would be fairly limited in terms of allowable motions. Assuming you use 4 LEDs (the max) you could position one on each knee and maybe on each shoulder. Orientation and gate speed would probably be fairly simple, but determining motions like backward motion, side-stepping, crouching, and jumping could be a fairly difficult problem.

I think the Kinect is the easier route because the software already handles limb tracking out of the box, so there are a huge number of possibilities in terms of mapping physical movement to character movement. I think they would work better mounted in the normal horizontal position - maybe two of them at 90 degree angles in case they have trouble interpreting limb motion from behind.

[Okta]

Im having visions of Cybers harness and frame and a tray of marbles to walk on... not sure how we can turn then with no traction....

This is the main thing that had stopped my VR tinkering, trying to find a movement method.

[brantlew]

I hate to play devil's advocate here, but the more I think about the harness/roller concept, the less workable it seems. I feel like there would always be a compromise between freedom of movement and balance. The more weight and support that was given over to the harness, the more balanced you would be and the faster that you could move your legs - but at the expense of any other type of motion like easy turning. Less support would give you more mobility but less balance. In either scenario though I don't think it would feel very much like actually walking or running because you are just going through the leg motions and not actually distributing your weight properly. The whole thing might be really awkward and very limiting just so you could essentially "pedal" in place.

I think the better compromise might just be to use Kinect cameras and limb tracking and walk/run in place to activate motion. It still has an artificial feel to it, but at least you are using your muscles properly and your other motions are not limited in any way. Plus, what could be cheaper?

[foisi]

I think the better compromise might just be to use Kinect cameras and limb tracking and walk/run in place to activate motion. It still has an artificial feel to it, but at least you are using your muscles properly and your other motions are not limited in any way. Plus, what could be cheaper?

From what I know a kinect camera cannot work with another one at the same time. This is due to the fact that it has an IR projector with a dot matrix which is unique for each kinect device and it would interfere with the one of the other camera.
I don't know if only one Kinect can track your body in 360°.
Another limitation is that it can see you with accuracy in depth from between 4 and 11 feet away. Maybe you can have multiple Kinects that turn on and off depending from where you are ? but you have to know that once a kinect is plugged into an usb port, the projector is automatically turned on (and I don't know if it is possible to disable from software)

[brantlew]

From what I know a kinect camera cannot work with another one at the same time.

I'm not aware of that limitation. I've never worked with a Kinect but I assumed it was possible because this guy did it.

[youtube]http://www.youtube.com/watch?v=5-w7UXCAUJE[/youtube]

I've been thinking about the walking in place idea a bit more and realized that it opens up the possibility of a mobile platform for terrain simulation. Admittedly, this would be really dangerous and only appropriate in certain circumstances but if the visuals matched the terrain accurately, your body could prepare for the slope changes and you could stay balanced on one of these. This would really be great in a game like SkyRim where you are walking around a huge mountainous world. Maybe attach a 6'x6' piece of plywood to the top so you have a lot of area to place your feet.

[youtube]http://www.youtube.com/watch?v=8bLwvoPD ... re=related[/youtube]

Here's a $5000 one but I'm sure some industrious DIY'er could figure out how to make one for <$1000

http://www.ckas.com.au/2dof_low_cost_systems_50.html

[Okta]

I hate to play devil's advocate here, but the more I think about the harness/roller concept, the less workable it seems. I feel like there would always be a compromise between freedom of movement and balance. The more weight and support that was given over to the harness, the more balanced you would be and the faster that you could move your legs - but at the expense of any other type of motion like easy turning. Less support would give you more mobility but less balance. In either scenario though I don't think it would feel very much like actually walking or running because you are just going through the leg motions and not actually distributing your weight properly. The whole thing might be really awkward and very limiting just so you could essentially "pedal" in place.

No matter what we come up with there is huge compromise. I dont like the jogging on the spot idea a lot. Aside from it feeling wrong it would be difficult to implement strafing (side stepping) which would be very important for immersion and is a built feature in almost every game that should be taken advantage of with simple key binding just like forward/backward/jump/duck.

How about the harness with an oily/slippery floor surface? You could put down a decent amount of pressure and its simple and cheap as, just a little messy and dangerous I think it could give a lot of tactile feel to running in VR, it would feel all erm...a word i hate....surreal?
The harness should be suspended by a sprung tension. This would help with reasonable foot pressure and make jumping possible
Another problem with the harness is it looks like a wedgie torture device. Are those things designed to 'sit' in like ab sailing? Vertical harnessing might be a bit more tricky/expensive, we need some 'circ du soleil' techs in here

That leaves turning. At this stage i can only think of spinning yourself via an encircling hand rail as part of the frame...lame....

[brantlew]

Aside from it feeling wrong it would be difficult to implement strafing (side stepping) which would be very important for immersion and is a built feature in almost every game that should be taken advantage of with simple key binding just like forward/backward/jump/duck.

How about this. When I actually stand up and physically strafe at a walking pace, I'm basically crossing one leg in front of the other to accomplish that. Why not use that very intuitive motion to implement strafing? So if you cross your left leg in front of the right and walk in place, you will strafe right. And vice-versa.

Plus if you want to implement a side dodge (as opposed to a continuous sideways strafe walk) you could just do that directly. Just allow a single quick step to the side.

[Aeroflux]

This would really be great in a game like SkyRim where you are walking around a huge mountainous world.

Only if the game doesnt treat mountains as a cascade of invisible walls like Oblivion did.

Sorry, off topic, but Ive followed that series since day one. In Daggerfall you could actually climb. Since then the mountains are just for pretties.

[brantlew]

The demos of SkyRim I have seen definitely include a lot of mountainous terrain - not just backgrounds.

[Aeroflux]

What I meant by saying "cascade of invisible walls" was that mountainous terrain wasn't directly traversable in Oblivion and Morrowind. You had to follow paths, which is a bit hypocritical since the game was built in a sandbox environment (except Arena of course). It would not allow you to traverse a steep incline. I wouldn't want to simulate such a steep incline without virtual belay anchors of course.

[cybereality]

Ok, how about something more realistic and practical. I know this is going to sound ghetto, but hear me out. Years ago I had a Playstation DDR pad and hooked it up to the PC. I mapped the buttons to the WASD so that if I hopped forward it would walk forward, to the left for strafe, etc. Then I used a P5 glove for camera/gun control and my trusty ESLA shutter glasses on a CRT. Low-fi, I know, but it was actually pretty fun. Now I think this same concept would work, but just needing some more modern tech.



The basic control setup would be like you were standing in a circle about 4 feet diameter. In the middle on the circle is stand still. In the front of the circle (relative to your forward orientation) is move forward. If you go a little further it could run. Back is similar. Left or right (relative to you) would strafe. You could of course move forward and strafe at the same time by going diagonal. To stop again just hop into the middle. I think this is ideal, because it will naturally confine you to a small area, lets say 6' x 6', which anyone can make room in their house for. For more realism, you could walk in place. This could be done with a Kinect, maybe a Wiimote mounted on the ceiling (with IR emitters on the top of the HMD). Even better, with a positional tracker (6/9-DOF) on the HMD this can be implemented without any external hardware with line-of sight issues. I'm not sure, but the Vuzix 6TC tracker may even allow this (have to check the SDK). I know this is not as cool as we would dream, but its something that's available and affordable right know (lets say $150 for the tracker and some blood/sweat).

[bobv5]

Hmmm.

All a mater of personal preference, but I think I would rather stick with a little joystick on the gun.

When you walk, you just think "walk over that way" not "move right leg here, then left leg there". It seems like having think about putting feet in right place would lower immersion, not increase it. Especially when you are moving a fair amount anyway, turning, ducking, jumping and leaning.

A cool solution for homebrew software would be to not walk at all. I mean, why walk when you could have a jetpack? I imagine brains could be fooled pretty well when sat lent way back in high fov gear. Ever feel like your moving/floating in bed after a long gaming session or a trip to the theme park? (umm, that happens to other people, not just me, right?) I think I saw something about it on a forum somewhere, perhaps this one, but I cant remember the word I need to search for.

http://flogistondesign.com/chair.htm" onclick="window.open(this.href);return false;

[Okta]

I had forgotten you mentioned this a few times before and looking at it now it seems like it would work pretty well.
Just the basic matt mapped to the wasd would be pretty cool. At least you get the notion of movement controlled by the feet, you just need to imagine you are wearing roller skates This would still be better IMO than using the wii nunchuck joystick for movement.

What would be extra cool is if the matt were mapped to go in the direction you press relative to your facing. Just mapping the matt normally would lock you into forward facing all the time without turning, the key binding must spin to follow you if you know what i mean.
The game would require movement keys not bound to mouse look and you might need some led's and a camera to track body/hip facing.

[cybereality]

@Okta: Well I was using the dance pad as an example so you guys could see where the idea came from. I think it could be better implemented with positional trackers. Maybe something that was worn on a belt or something. You could also use optical, but I think it would get more complicated. That way the "forward" is always relative to where you are facing. I'm hoping I can do this all with the Vuzix 6TC tracker. They claim its 6-DOF, but that could mean different things. I also wonder about the possibility of hacking the 6TC tracker to work without the 1200VR. You can buy the tracker alone for $130, and the Vuzix SDK it pretty simple from what I recall. But there are also those IMUs from the other thread, so thats not the only option.

[brantlew]

I agree with Cyber. Other than the full omni-treadmill solution, I think all the contraptions are going to be large, complex, awkward, and in the end not very satisfying. I think sticking to basics is the better approach. Using your natural body motions, keeping it cheap, and within a confined space is going to be a lot more practical in the end. But I agree the foot sensor pad could be improved a lot.

Some combination of Kinect(s) and/or Wiimote sensors is the way to go. If you just wanted the foot sensor pad concept but have it rotate and align with your body, then probably just an overhead Wiimote would do the trick. The IR solution would also be a lot more forgiving than a physical footpad sensor. You wouldn't have to think very hard about absolute foot placement - just relative direction of motion. The IR would let you drift around the floor and keep your "center" calibrated.

Personally though, I think a more complex interface would also be more satisfying as long as you keep the controls intuitive. Limb tracking via the Kinect opens up a lot of possibilities - assuming you have game support for it. Variable speed walking, variable height crouching, jumping, laying down, arm crawling, dodging, and even strafing could all be mapped to either exact or very similar real life motions. Sure those motions might not be as simple to perform as they are in games now. I would probably throw up if I tried to duck and jump as often as I do in a FPS. But if you want realism - there you have it.

[Okta]

Cyber: Also dont forget that Carl Kenner blocked Glovepie from working with a Vuzix device attached
Just had a thought, run glovepie on a proxy machine and forward commands to the main pc...

[WiredEarp]

Its possible you could lie back on a weight bench type thing, on an incline, with your feet in devices kind of like Novint Falcons, and use that to move. That way you wouldn't have all the weight on the harness (or have to wear a full harness). This would also allow force feedback, so you could feel things. You wouldn't be able to cross your legs easily (depending on setup) but it would probably provide the best sense of realism. A similar arrangement for your arms, like the Novint XIO, would allow you to get force feedback for the arms.
I stole most of these ideas from lawnmower man

Currently, I dont see it as being at all easy to add realistic walking. I'm just going to stick with the stick and gestures etc to move around for now.

I had my first immersive moment last night with my Z800. Am working on setting it up with COD5, and walked into a pylon base and looked up - for a moment, it all looked, and more importantly, FELT, real. Had a couple of other similar moments moving around the forest. I think having the headtracking combined with the gun is definitely helping my immersion, plus I no longer need to turn heaps. I can just look left/right and line up the gun... and turn when I really need to, but now due to the double mice acceleration, I can do a 360 in about 270 degrees.

I've noticed as well that as PalmerTech said in his thread, my body has no idea of its real position. Had several moments where i've taken off the headset and realised i was facing a totally different direction than I expected. I only have the lights off, a full cover hmd would probably be even better.

OT wierd idea - could you design HMD optics, silver dots, mirrors, prisms, etc, so that they reflected the edges of the display screen and wrapped them over the bezel? That way, your black edges would be the same colour as the edge pixels, and quite possibly help immersion significantly over just having a black border. It would be like a budget ambilight system.

[Synexious]

Why did Carl do that? There can be no legitimate explanation for what he did. That's unacceptable.

EDIT: On Google, I found a thread from here in which Cyber said Carl and Vuzix had a dispute. What was it over?

[WiredEarp]

Can't imagine it would be too hard to reenable that/noop out that anti Vuzix routine, unless he's put more than usual effort into protecting it...

[Okta]

Why did Carl do that? There can be no legitimate explanation for what he did. That's unacceptable.

EDIT: On Google, I found a thread from here in which Cyber said Carl and Vuzix had a dispute. What was it over?

Carl was also on the verge of obliterating Glovepie because its power usage was destroying the environment....

[bobv5]

WiredEarp-- Its not that difficult to just make a mini ambilight system. Did some experiments, it does help immersion, but I'm not very good at programming so never got it working quite right. But that is my own fault for starting from scratch, instead of copying someone elses design.

http://blogger.xs4all.nl/loosen/articles/408184.aspx" onclick="window.open(this.href);return false;

EDIT-- I didn't make boblight. Just the same name!

Aphradonis-- He did that because he was banned from the vuzix forums. What I read is that he was banned because he refused to help some guy who wanted to use vr920 to train figter pilots. It's difficult to tell what really happened because someone at vuzix went round deleting the threads.

http://www.mtbs3d.com/phpBB/viewtopic.php?f=21&t=2188" onclick="window.open(this.href);return false;

I'm not sure if glovepie will work with vuzix gear attatched and just refuse to interact with it, or not start at all with vuzix gear. Big difference if you are using vuzix for display and something else for tracking.

I also read that this restriction will be dropped in the next version of glovepie when he gets round to releasing it.

[PalmerTech]

Omnidirectional motion would be pretty awesome! I am going to be a contrarian, though.

I don't think you will ever get the same sense of immersion from any of these solutions as actually having a huge space to walk around, unfortunately. But hey, hopefully we can get close!

@WiredERP: You should make a thread about bezel enhancement/ambilight systems, I have some stuff I want to post there, and do not want to derail this thread.

I think the CyberCarpet could be made fairly easily, as well. Ideally a large one! I also think that using larger ball bearings would make for looser tolerances, easier maintenance, and much easier construction. Probably somewhere around the size of small marbles? You would want to wear pretty stiff bottomed footwear (Like boots) so you do not feel the rougher texture, but I think it could work very, very well.

[cybereality]

@Okta: Ah yes, I know. That is why I am writing my own driver: http://www.youtube.com/watch?v=mlfZyRzfVMM" onclick="window.open(this.href);return false;

@Aphradonis: Evidently Carl went off on some political rant and that wasn't allowed on the forums (isn't allowed here either FYI). So he got banned and, as retaliation, he dropped support for the VR920. Kinda sucks, but whatever. I should have my mouse-emulator done soon.

[Synexious]

The ball carpet would be close to reality, and a large enough omnidirectional treadmill with proper inertia compensation is indistinguishable from reality for walking on flat surfaces. ODTs could also be made inclinable.

[Okta]

Check this new thread of the BF3 sim with this new onmi carpet http://www.mtbs3d.com/phpBB/viewtopic.p ... 268#p65268" onclick="window.open(this.href);return false;

Looks like a possible DIY build.

[PalmerTech]

Wow, I am going to have to watch that episode! Looks like VR is about to get a big push, marketing wise.

I will have to see that treadmill in action more, it looks a lot easier to build. Seems like the software to control it would be pretty tough, though; I mean, what if you decide to turn 45 degrees near the edge of the platform? It would have to reel you in pretty fast before you ran out of space!

[brantlew]

They don't show much of the treadmill in action. I would like to see a lot more of it. It looks like it has some limitations but it does appear to the best possibility for a DIY system discussed on this thread so far. Basically 16 independently controlled belts under those horizontal rotating rods. I don't think the software would be near as difficult as the mechanical build however. That's a lot of precision placed moving parts! I wonder if 8 would work almost as well? I love the fact that the center piece is not moving so you have a firm foundation to jump and kneel on. I'm curious to see someone jog or run in a straight line on it to see if there are balance issues. Like Palmer I am also skeptical about the freedom of movement once you get far enough away from the center. I don't think you would be able to run and then turn quickly. For controlled walking it looks like it would work pretty well though.

What's your best guesstimate of the DIY cost? I would guess about 2K.

PS. If you really wanted to get crazy you could mount it on a 2DOF platform to simulate hills.

[Synexious]

I think a ball carpet would work better and be cheaper to build. It would also require less space.

[PalmerTech]

I don't like that the center does not move, it is one of the things that makes the software so hard. It needs to somehow be tracking your gait well enough to predict when you are about to step onto it, and try to compensate down to very low speeds as you step onto the plate, and then slowly ramp it back up as you step off the other side.

As far as price, it could be pretty cheap if you somehow got it all right on the very first try. The hard part is the software, and incorporating that into existing games somehow.

[brantlew]

The way I envision it, you would not transfer from one roller across the plate to the opposite roller unless your motion stopped. So for example if you walked backward, you would move onto the rear roller and walk in place. Then if you immediately switched direction the roller would just reverse and you would walk forwards in place (still) on the rear roller. Only when you stop moving does the roller gently push you back to the center. I'm sure that's a bit of a simplification. You can probably jump all around and exceed its limitations, but the idea would be to avoid transferring over the center during motion and then return to the center during rest.

The center might actually be helpful because you have a bit of a movement buffer where small non-directional movements (shifting side-to-side, leaning, turning) will not trigger any roller motion. You have to make a really directed and easily detectable movement to get on the rollers and engage them.

Tracking the gait speed seems like the most difficult thing, but I think engaging the proper rollers is pretty straight forward. Plus the system is extendable. I bet 24 or 32 rollers would really smooth things out and allow you all kinds of side stepping action.

The things that might be awkward would be running forward and turning 90 degrees. Also there might be a strange transition as the roller guided you back onto the plate when standing still. It couldn't just push you all the way onto the flat without causing some balance problems.