Envisioning a DIY CyberCarpet for Omnidirectional Motion

[brantlew]

Thinking a bit more about it... I think even 90 degree turns could be handled reasonable well.

If you walk forward you move onto the forward roller and it matches your speed. Then you turn 90 degrees clockwise and continue walking. At this point the software will try to guide you onto the east roller. It allows you to walk eastward but gently nudges you south just enough to get you onto the east roller. More rollers would allow more subtle angles and give you a smoother transition. It's really ingenious! - in theory. I'm really curious to see it in action more.

[cybereality]

It seems problematic to me that the center is immobile. If you were to stop on the side somewhere, wouldn't you trip if they floor tried to push you back into the middle? And if it didn't do this, then you could quickly run out of space.

[PalmerTech]

More rollers could definitely work! But...

The problem with these systems I am figuring out is that the prices scale upwards really, really fast. I mean, sure, 36 rollers would be great! But every single roller segment needs a pretty beefy motor+the custom made roller service. I am going to be very, VERY optimistic, and assume that each motor was a surplus find for $50. Add $15 per motor for the control/power boards (Again, pretty darn optimistic). Now, watching that video, it looks like each segment has about 50 individual rollers, so assume a mere $3 per roller (Which is nothing, considering that each one needs two ball bearing assemblies, one for each end/gearing for movement/tubing/endcaps/etc).

Finally, you need the actual base itself. This is an exercise in optimism, so we can pretend that we built it out of cheap wood and cheap nails for about $100.

$50x16 = $800
$15x16 = $240
$3x50x16= $2400
$100x1 = $100

Total cost, being as optimistic as possible, and getting it all right on the first try: $3540

And the reality is, motors powerful enough to move a person fast enough are going to cost closer to $100, the control and power electronics would cost at least $50, and the rollers would cost a lot more (You could always use less of them, I suppose). Look at how much normal treadmills cost, you cannot get one for less than $250-300, and those are terrible build quality, and the motors typically only go up to 4.5mpg or so, which is average walking speed.

[pierreye]

Couldn't we use a belt to pull all the roller in each section with one motor? But to me the biggest problem is space requirement. It's tough to keep the item when not in use or basically I need to allocate a room just for VR.

We all need to learn moonwalking to save cost and space.

[mAchiNE]

For DIY couldn't we just use say 4 of those roller belts, use a bluetooth pedometer to track leg speed. THe way it would work is all rollers are always running a minimum speed, when you step onto the rollers the pedometer information is used to speed up/slow down the rollers depending on how fast you are walking/running and adjusts both the roller speed and the characters speed (walk/run/sprint) in game. the only problem is how to walk backwards? but I think walking backwards on rollers is dangerous at the best of times anyway so maybe backwards movement can be by standing on the edge of the stationary platform with your back to the edge of the rollers somehow that could trigger backwards movement.

this setup would have some problems but should be cheap enough and easy enough to implament for DIY I think.

[brantlew]

Palmer's got it right. We're talking about a $5,000 system which puts it on the really high end of DIY projects. Still I think it gets really close to a true omni-treadmill for probaby 1/10 the cost. The flight-sim guys spend similar amounts of cash on their motion rigs so it's within reason for a uber-enthusiast.

mAchine - If we want to get down to a more practical setup, you might be right about the 4-way system - and from there you can simplify it even more. With a 4-way system you can skip all the custom rollers and just buy 4 cheap exercise treadmills. The motors and rollers are already included and you just have to rewire and reprogram the motor controllers. Pretty much the same concept but with a really course angular granularity. Backward/forward and strafe would be no big deal. The main problem would be turning. You would be limited to 4 directions, otherwise you would step in-between two treadmills.

I wonder if you could create a setup where you had just 4 (or even 2) treadmills and a rotating disk in the center. As you turned, the disk would counter-act your rotation to keep the treadmills aligned with your body so as you stepped forward you would always walk straight onto a treadmill. Probably an awkward setup but more affordable at least.

[cybereality]

How about we go cheap? Just get a bunch of stainless steel balls and put them in a box on the floor. Looks like you can get around 1000 kilograms of metal balls for around $500 bulk from China:
http://www.alibaba.com/products/bulk_st ... 03869.html" onclick="window.open(this.href);return false;

Then all you need is some sort of pod and/or harness which will hold you in place. Seems like the whole setup could be done for under $1,000.

[PalmerTech]

Would that really work, though? I mean, they would probably not rotate all that easily.

Here is what I have been looking at for my build: http://www.vxb.com/page/bearings/PROD/B ... ts/Kit7764" onclick="window.open(this.href);return false;

They are ball transfer units, I saw some at the airport in the cargo area floor. Then put a bunch of them in the floor, ball facing up, and you can slide things along them. If you covered an entire flat board with them and used a harness (There is some good stuff here: http://www.rehabharness.com/harnesses.htm" onclick="window.open(this.href);return false;), then you could just walk right on top! In fact, if you mounted them in a very shallow dish shape, you might even be able to avoid the harness entirely! I got that idea from here: http://www.woodway.com/performancetreadmills/curve.html" onclick="window.open(this.href);return false; I have watched quite a few videos of that treadmill, and it seems that a slight curve would allow reasonable speeds without any need for a harness. Personally, I would want a harness anyways, but only to prevent me from falling if I trip. The goal would be to have no tether.

The cost might get pretty high, it all depends on how large the walking area would be, and how widely you can space the transfer units without feeling the space in between. I think it could be done for under $2000 for sure, though!

Real pity I can't get my job to fund this project, we have too much invested in real-space, time and money wise. Something like this is a totally different direction that nobody wants to take.

[bobv5]

Perhaps when you are making hmds cheap enough for 20 or 30 guys to be training at the same time, they will run out of space?

[cybereality]

@PalmerTech: You are probably right. The balls by themselves might not work correctly without some sort of ball bearing mechanism. The thing you linked to does look promising. But at $4 a piece, that could get pricey fast. Though I guess not out of reason.

[PalmerTech]

@Bobv5: Not really. If they are training 20/30 soldiers at once, they will be doing squad training, where they need to work together anyways. And this is the military we are talking about, if there is one thing they have, it is space.

@Cyber: You can buy 10 at a time for only $3 apiece, and I just sent in a quote request for 100, 256, 1000, and 3000. I will update when I hear back.

[mAchiNE]

@Palmer
That is an awesome Idea! I think this is the most practical and cost effective soloution for a DIY setup I have heard of so far, so basically the "cyber floor" is passive and the direction is taken from trackers and the speed could be taken from pedometers? no need for complex mechanics or software, and can be done in a relatively small space (possibly small enough to flip it up and hide in a cupboard when not in use)
And I agree with you, even if you can make it safe to walk on without a harness I'd still wear one to catch you if you fall! especially when wearing thousands of dollars of gear on your head/back!

Look foward to hearing about bulk buy prices, I'd be keen to build this setup

[brantlew]

That does look simple and promising Palmer.

Simply using pedometers does not give information about direction. In other words you would always be moving forward - no backward movement or strafing. You still need some type of directionality information either built into the floor (tracking bearing movement), built into shoes, overhead IR detection, or maybe limb tracking via something like Kinect.

I wonder if you could paint the top half of each bearing black so that you could visually determine the spin direction of a group of bearings?

[mAchiNE]

That does look simple and promising Palmer.

Simply using pedometers does not give information about direction. In other words you would always be moving forward - no backward movement or strafing. You still need some type of directionality information either built into the floor (tracking bearing movement), built into shoes, overhead IR detection, or maybe limb tracking via something like Kinect.

I wonder if you could paint the top half of each bearing black so that you could visually determine the spin direction of a group of bearings?

Yes you are correct, I did not think of that , left/right turning is taken care of with the aiming device but "WSAD" movement still needs to be accounted for pedometer would only give information about speed (walk/run/sprint)

limb tracking would be essential I think but i'm not sure if you can use more than one kinect at a time (the infrared projector of one kinect might intefere with the other kinect?) also input lag on kinect could be offputting, but you definately can use kinect for this just not sure exactly how well it will work.

All we need to track is the legs to determin weather we are going foward, backwards, or strafing left/right. The problem is we need to track this movement at any angle! which makes the kintect fairly usless without at least 2 preferably 4 of them and som complex code to to combine all the information and determine what the limbs are doing...

I wonder if we strap 2 Wiimotes, one to each leg, if we could write a glovepie script to track the motion of both legs to detrmine WSAD information

[brantlew]

Why would a single Kinect not work? From the demos I've seen it appears the Kinect can track body motion from any angle. Watch this guy spin in a circle.

[youtube]http://www.youtube.com/watch?v=nr8vgCnb9_0[/youtube]

I've seen people doing spin kicks and it tracks that as well. Anyone here have much experience with the Kinect? Does it have severe limitations tracking from a rear view?

[mAchiNE]

Hmm maybe one is enough then

What make you thinkn that guy is using Kinect?
I think he is using A 3D depth Sensor, the X-tion Pro, made by ASUS and running openni software.
see here:
http://www.openni.org/Hardware.aspx" onclick="window.open(this.href);return false;

but that does look very promising, very little input lag in that video, where kinect has some really bad inputlag which I think would cause problems unless someone has fixed that.

X-tion pro can be bought here:
http://us.estore.asus.com/index.php?l=p ... ail&p=3397" onclick="window.open(this.href);return false;
not too expensive either!

I already have a Kinect, maybe I will try out some of this software to see how well it works.

[PalmerTech]

A single Kinect would probably work fine, but using multiple ones allows for greater redundancy and accuracy.

You know, the harness could also be used for speed information. Have a Kinect or something to determine what direction your are facing, and measure how much pulling force there is on the harness.

Just brainstorming! I always skew towards extra hardware over extra software, but that is only because I am confused by the latter.

[mAchiNE]

one problem though..
assuming a 3 foot radius circle, we would require roughly 4000 of those ball bearing units! at $3each thats $12000!!!! even if you can get away with spacing them an inch apart you still are looking at $6000
just for the bearings.

I hope there are some really good discounts for bulk buy!

[brantlew]

Here was the original link I followed.

http://urbanhonking.com/ideasfordozens/ ... rocessing/

Digging a little deeper I see that they are using the Kinect sensor but bypassing the Kinect processor and using OpenNI for tracking. The results look great, but I have also seen demos with just the built-in Kinect processor that seem to do a pretty good job as well. The Kinect API is high level too, so you actually get objects back like JointID.KneeRight and JointID.Head to work with and don't have to fool around with all the low level image processing. The trick would then be to take the high level skeletal data and perform a motion analysis on the legs to translate the different movements to standard direction commands. (WASD) I'm simplifying it of course. There are a lot of motions that would map to the same command. For example, I can strafe by crossing one leg over the other or I can also just bring my feet together and step to one side without crossing. Recognizing all these locomotion methods would be challenging. Maybe focusing on the torso instead of the legs might simplify matters, but I am just speculating. Surely someone has already solved this out there in open-source land.

Palmer, I thought about the harness pull solution also. Probably would work but I'm not sure how well pull force would directly map to speed with all the bouncing up and down and bending forward and backward. It might be a really noisy signal. I guess if you just map to 2 different states (walk and run) in a particular direction you might be able to tease all the noise out. An interesting possibility - plus potentially simple to implement.

Opposite to you, I tend to gravitate towards the software solution. I have a hammer and so I see nails.

[PalmerTech]

@Machine: I don't think your math is right.

Each unit is just under 1.2" wide. 3 feet would be 36 inches. 36/1.2= ~30.

So, you need 43 for a line that is 3 feet long. For ease of math, pretend we are making a square area. 30x30 = 900, 900x3 = $2700

And actually, I just got a quote. The price for 1500 units would be $2.39 each, 900x2.39 = $2151.

Realistically, you would not even need 900, since that would be a square. The actual number would come closer to 700, and that is if you spaced them right up against each other. In that case, 700x2.39 = $1673.

@Brantlew: Something to consider is that even most PC games now support analog movement via a gamepad, probably better than using WASD.

[mAchiNE]

@Palmer
I was caluclating for a circle 6foot diameter, (3foot radius)
to calculate the area of a circle with a 36" radius then it is Pi*36*36=4071 square inches, sinc each ball bearing is 1" I based it on that for my rough calculations (I did not take into consideration for the casing) but for a 6foot diameter circle you would need at least 2000 of the ballbearings.

Do you think a 3foot diameter circle is big enough? swing one of your legs foward to maximum stride, measure back to the other foot, double that and that is roughlyu how big it needs to be for you.

[brantlew]

Palmer - I didn't realize that. (Who in their right mind would choose a gamepad over keyboard/mouse for an FPS?) Determining arbitrary directionality via limb motion could get quite complex given the numerous possibilities. I believe if the system was well implemented that it would be the most accurate and flexible, but it might take a long time to work out all the complexities. A more direct solution might be just using an overhead IR or pulley system like you mentioned.

Of course this is all just wild arm-chair speculation. This omni-treadmill thing-a-ma-jigger is a real long-shot anyways. Don't waste too much time on it, because what we all REALLY need are the detailed PR3 build specs!

[PalmerTech]

It would not be enough for running, but it would be plenty for walking at a reasonable speed. The goal would be to keep the center mass as close to the center of the treadmill as possible, and I think 3 foot diameter would be enough (That is larger than the Cybercarpet, for sure).

@Brantlew: I am working on an ever better version right now, I want to make as many improvements as I can before writing it all up. I am going to be laser cutting the shell for this one, so it is going to be as small and lightweight as possible.

EDIT: And yes, I know that walking is not as fun as running. But it makes more sense to see if this even works first! If I did this myself, I would probably order barely enough bearings to make a long, narrow strip. That way, I could make a single direction treadmill out of them, and see how that goes.

[mAchiNE]

A top down IP camera sounds good in theory but how can it determine anything other that which direction you are facing? I'm not sure it will be able to tell if you are walking fowards/backwards or strafing if you are essentially stationary because the "cyber carpet" is keeping you in place as you move around in the virtual world. possibly a custom skeleton tracking setup with 2-4 IR cameras around you and wearing a "suit" that has IR LEDs at each joint would allow for fairly accurate skeleton tracking, just the software to convert the data to WSAD or analogue signals that can be used in game would be the hard part.

EDIT: If 3 foot is enough then a 3foot diameter circle of bearings would require roughly 850 of the ballbearing assebliiles (by square inch calculation/ 1.2) so much more affordable I was worried there for a mi!nute

[mAchiNE]

This typ might be easiear to mount:
http://newzealand.rs-online.com/web/p/b ... s/0687770/" onclick="window.open(this.href);return false;

if they can be found much cheaper

EDIT:
Here's another possible option:
http://newzealand.rs-online.com/web/p/b ... s/0408536/" onclick="window.open(this.href);return false;
Omni roller! much cheaper than the ballbearing units and you would need less of them... (although questionable about how much weight it can handle)

[brantlew]

I was thinking of a top down IR just to detect off-center movement. The carpet can't keep you perfectly in place. There would be some displacement from the center - even if it's shaky. So basically it's similar to the harness-pulley system except it's optical instead of mechanical - a short-cut for determining direction of motion instead of limb analysis.

Other characteristics like gait speed and body direction could be determined by other methods (Kinect, pedometer, IR, etc).

I think if you are going to do limb tracking of any kind, you get a head-start by using the Kinect because it already converts the signals into high level joint objects for you. Going for a lower level IR solution to track joints means you have to start from scratch.

[Synexious]

The CyberCarpet uses 4000 balls and I think it is too small. I would want something bigger. The only affordable way to do this would be to use the same method they did - placing loose balls in a custom grid. To prevent them falling out, a second layer with holes slightly smaller than the diameter of the balls could be placed on top. How much would it cost to have grids cut, out of plastic or metal? Maybe $300 or $400? And the balls themselves should be really cheap. The ball changers might be better but they are too expensive, unless the cost continues to decrease - would 3000 cost half as much as 1500? How about 6000? 9000? 12,000? I would want a surface two or three times as large as the CyberCarpet, at least. I would like to be able to run.

[PalmerTech]

This typ might be easiear to mount:
http://newzealand.rs-online.com/web/p/b ... s/0687770/" onclick="window.open(this.href);return false;

if they can be found much cheaper

http://www.vxb.com/page/bearings/PROD/Kit7760" onclick="window.open(this.href);return false;

@Aphradonis: The problem is, loose balls in a grid do not spin freely, they have quite a bit of friction. These units I linked to rest on a bed of ball bearings, which is why they would work passively. The cybercarpet only works because it has an active treadmill underneath that is bearing most of the weight, instead of the plastic grid.

I mean, for an active solution, the Cybercarpet could work, but it would definitely not work as a passive solution. As for prices, I do not think that it would possibly go under $2 per bearing.

[brantlew]

Hey guys, I think I figured out a way to get really accurate direction and speed tracking - at least for walking. When I first thought about using the Kinect I thought in terms of recognizing different hip and leg "gestures" and translating those to motion commands. That gets complicated because there are so many ways you can move your legs to locomote. I think instead you can just ignore the legs entirely and just concentrate on the feet position. When you walk one foot is always on the ground. So all you have to do is identify which foot is planted and then calculate the velocity and direction of motion between the raised foot and the grounded foot. That will be your motion vector! In normal walking the grounded foot position is constant and all relative motion comes from the raised foot. On a omni-carpet, both feet move but the "relative" motion should still be approximately the same. So walk anyway you like - sideways, bow-legged, angled. It doesn't matter because the relative foot positions are all that matter. It simplifies the problem tremendously. I would have to get a Kinect to prove this, but it seems like a really easy problem now.

[pierreye]

Maybe can consider this pre build ball table. Just ask them not to install the leg.

http://www.alwayse.com/en/products/ball_tables.html" onclick="window.open(this.href);return false;

Or can we consider using the same ball but we build it like a roller skate and build a slightly curve circle using wood to walk in? In this case we don't need that many balls, just enough to cover an average shoe size.

[Synexious]

I thought I read the CyberCarpet can be driven actively or passively. Even actively, wouldn't it be cheaper still? 4000 ball changers would be $8000. And 4000 is barely enough. Still, ten or twelve thousand or so for omnidirectional motion is a bargain compared to $50,000-$100,000+ for an ODT. I hope a CyberCarpet could be done for $5000.

Here's a news article about the CyberWalk ODT, incorrectly called the CyberCarpet. Palmer - you said no ODT could ever perfectly match a free range for VR. Not quite; a 100x100m ODT would be imperceptible from walking on solid ground. At least 6x6m eliminates the feeling of walking on ice. Hopefully in a few decades large chainbelt ODTs will be an affordable investment or DIY project - $15,000 or less. For now a ball carpet is the best (well, only) practical solution.

[cybereality]

@brantlew: Yes, that is genius. The vector of the raised foot from the planted one will indicate which direction and at what velocity you are moving. I imagine the Kinect could do this (line of sight issues not-withstanding) but you could also attach a Wiimote to each shoe, or even some custom IMU type stuff. I do kind of like the idea of modding a sneaker to have the sensors right in there: cyber-shoes!!!! But the Kinect would probably be easier since there is already a lot of open-source code to work with it. However I know there is some lag with the Kinect, but its probably not so bad.

[Synexious]

Maybe can consider this pre build ball table. Just ask them not to install the leg.

http://www.alwayse.com/en/products/ball_tables.html" onclick="window.open(this.href);return false;

Or can we consider using the same ball but we build it like a roller skate and build a slightly curve circle using wood to walk in? In this case we don't need that many balls, just enough to cover an average shoe size.

Excellent ideas, pierreye! So essentially CyberCarpets already exist as commercial products! I had no idea! The largest is 7x4ft (approximately 300mm in 1ft)! That's great. How much do they cost?

I also like the idea of putting the balls on the soles of VR shoes. Like those powered skates but omnidirectional instead of bidirectional. That could be an ultra-cheap, compact solution. Like under $1000 for serviceable omnidirectional motion in place!! Wow!

[cybereality]

YES!!! Putting the balls on the shoes! Another genius!!!

[Synexious]

Here's a ball transfer table for $666. It's way too small and the spacing between the balls looks too big. A large enough table must cost thousands, and finding one with small enough balls might not be possible. Let's keep looking. It seems hard to find info on them. Pierreye's ball shoe idea might be the best way to go.

[PalmerTech]

Yup, those ball tables would be good if the density was higher, I have not been able to find any with closely spaced bearings.

Putting balls on the bottom of the shoes could work, but it seems like it would be akin to walking on rollerskates. You would almost certainly need a harness, hard to know without trying. It seems like it would be pretty slippery!

And yes, an enormous ODT would be better than an open space; But I don't think they are going to become practical anytime soon. Industrial machinery has always been expensive, and there is no real way to cut costs further; As an investment, it could make sense (For people who have the space, or a business), but I am not sure it would ever happen as low as $15,000. Hopefully I am just being pessimistic.

[Synexious]

I would definitely use a harness. I think ball shoes on a shallow dish with a harness is the only affordable option. I meant a small ODT for $15,000, like 6x6m. If they are $50,00 or less today, in ten or twenty years the price could be $15,000 or less.

[mAchiNE]

@Pierreye
Shoe idea so simple, its a great idea

@palmer
assuming you are using the same balls on the shoes as you would have been mounting to the base then it should be the same "slipperiness" I don't see how its any different.

[pierreye]

The only cons I can think off is maybe the shoe might be quite heavy. Or we only mount the balls near the edge of the shoes to cut down the weight.

[brantlew]

There was some discussion of roller shoes at the beginning of this thread. I'm still a bit skeptical about the idea. It just seems so slippery and ungainly - like walking on roller skates. At least it's cheap to try.

If there was a way to add some resistance to the balls then maybe that could help. At least then you could shift some of your weight onto them and feel the mechanics of walking a little better. Maybe then you wouldn't need quite as much harness support either - more like a safety rope instead of a balance assist. The bowl idea could complement this nicely. If the resistance were calibrated so that the shoes were fairly steady on flat ground but would roll on the incline then that might create a stable system.