"davidwyandgg" wrote: In my experience, Carmack was correct. At least with positional tracking, the Hydra is not accurate in its readings, but it is consistent. This means that if I place a Hydra controller into the exact same location relative to the base station I will get the same result (the consistency) but if I compare two positions relative to each other they will not match their real world relative positions (low accuracy).
- Dave
That... Is an interesting situation to deal with.
It kind of makes sense if you think about it, given that you're reading a pulsed magnetic field, unless something else in the environment changes, that field will remain consistent, but it might not be uniform in nature.
What this suggests is that you could calibrate for the field if you were inclined to. (move the controllers through a large area, and compare the actual position to the reported position.)
if the data is consistent, (as you say), then you should be able to build up enough of it to correctly calculate the offsets, and correctly report the position.
Of course, the downside here is that if you move the base station, the magnetic field could change (especially relative to any background magnetic fields), and you'd have to start the whole process all over again.
Also, that kind of calibration process is incredibly fiddly and time consuming.
place a desk in a empty room, calibrate the four corners by having someone place their head on each corner? Or simply have them place their nose in each corner of the room?
The other thoughts I have had is...could 1-2 cameras do this? I've seen some amazing head tracking software in stuff, so I think it could be done. But its all about latency too. Wonder if you could use video card cores to do the math fast enough?
If you're going down the route of optical tracking, the playstation method is much easier to pull off, and faster. (Or alternately, adapt the system the wii uses.)
The PS version would require sticking some markers to the HMD, and the Wii method is similar, but you'd need IR leds instead.
Tracking with these can be pretty good, but their weakness is if you lose line of sight to the camera doing the tracking, you lose your position information...
"davidwyandgg" wrote: The base station is on their head so they can take their hand tracking with them. Otherwise, they are limited to length of the Hydra's cables.
Ok, I can't imagine what the positional quality will be like. So I'll just have to wait and try when the rift+new hydra arrives.
But there are definitely two steps here:
1) Get a good control scheme and body movements sitting down or standing up without walking. 2) Full holodeck thing.
To achieve 1) you just need a third head sensor added to the hydra on the head and lots of hard- and soft- ware tweaking.
The walking will be a button or a stick, there's no way around it. I think for better immersion you can add a torso sensor and wireless+battery to the gear so you can rotate with body movements only (no TF2 though). But walking will always be button/stick.
2) Will happen when we jack into the brain. Consumer omnidirectional treadmills with good immersion won't happen this lifetime unfortunately.
"It's like Homeworld in first person." Disable Aero and vSync for a completely simulator sickness free experience with 2xHz FPS.
Keep the config utility open for tracking to work.
"greywar777" wrote: place a desk in a empty room, calibrate the four corners by having someone place their head on each corner? Or simply have them place their nose in each corner of the room?
The other thoughts I have had is...could 1-2 cameras do this? I've seen some amazing head tracking software in stuff, so I think it could be done. But its all about latency too. Wonder if you could use video card cores to do the math fast enough?
Whats the latency on the Kinect?
I know I kind of already replied here... But anyway... that calibration method would already rely on having good positional tracking.
If not, you'd need far more data than that to work with (the hydra method for instance... you would almost need a map of points through most of the environment for that...)
As for the kinect... I've heard some pretty bad things about it - some claim it has latency as high as 350 ms or more depending on what you're using it for. My guess though is the really high latency is to do with trying to perform full-body skeletal tracking. I assume you can get better results than this with it depending on what you're doing, but those kind of latency figures do make me wary of trying to use a kinect for head tracking for instance...
The holodeck guys use a PSMove for positional tracking because it's fast and reasonably accurate. Keep something in mind: it's not about *one* sensor type. So you don't rely only on the camera looking at a glowy ball... you also have linear accelerometers.
His thesis covers a lot of issues and mentions techniques like using a ball with a "constellation" which can provide absolute orientation (permitting recalibration of orientation against drift). By all accounts, orientation of the Rift is fantastic, which is really good since that's the most important! Thanks Palmer and Co. for deciding to make that better!
I'm a bit partial to having a camera rather than magnetic fields. Magnetic fields *could* be wonderful, but they're so finicky. While processing a general image with uncontrolled lighting, like Kinect... is expensive (latency) and extremely inaccurate. Inertial sensors are fast but provide a acceleration which can be integrated to return position but with drift.
To help calibrate absolute position I think a camera assisted by bright points is good. I'm planning to put two little LEDs (different colors) on the top corners of my Rift; maybe another (third color) on the back so there are always two points visible unless occluded. We know what points we're looking at (color), and we know how far apart they are physically, and we also know orientation (Rift)... so given a camera image with two of these points we can accurately determine position (x,y,z) relative to camera. This is more accurate than the PSMove because it's almost like having a globe the diameter of the rift. Also, if you use a higher resolution camera you can increase accuracy. However... I think moderate accuracy is sufficient when combined with the linear accelerometers.
That is my son in the first video. I hooked up the hydra to the tuscany demo to play with head tracking.
Back in December, when I thought I was getting the Rift, I started playing with the Hydra to get ready to develop for it. At that time i was playing with head tracking on a regular monitor, to create a virtual "window".
The hydra was attached to the phone in this case and a custom frustrum was created from the hydra's position relative to the monitor to make it look like you were looking into the monitor to a virtual world. Once I had the Tuscany demo and the Rift's orientation it was pretty easy to merge the two ideas.
I was getting pretty motion sick from testing the rift and a lot of people at work were as well. When I hook up the positional tracking I get almost no VR sickness unless I spin the camera significantly. I am taking the rig, with hydra, back to work tomorrow and test some of the people who got especially motion sick to see if the added positional tracking helps them as well.
While it's not perfect, it works pretty well, especially since you head doesn't do nearly the wide range of motion, nor the quick motion, that you hands can do. RoadtoVR.com is supposed to be putting up a tutorial I gave them about how I hooked up the hydra and also have a copy of the executable for people to download and play with. Paul said it was supposed to be today but it must have been delayed. Check RoadtoVR.com, it should be there soon.
Tony
A
Anonymous
April 12, 2013
Or you can just use the camera on the PC as I demo here. Not as precise but more people have webcams. "Demo - Girl Robot Car PCCam Head Tracking More NSFW" viewtopic.php?f=29&t=370
Optical tracking methods introduce some serious latency.
I'm not sure if that matters as much for positional tracking as it does for rotational tracking, but it's something to consider.
In my experience though, the average PC webcam is particularly bad at this, given that many of them struggle to maintain even 30 fps. (what the latency is for a frame is another question again.)
Meanwhile, the PSmove api atavener mentions benefits from the PS Eye camera, which, with appropriate drivers, can function as a webcam, but, critically, can handle framerates of 75 up to 187 fps depending on the resolution.
However, reading through the thesis itself, it still mentions that the latency for tracking a move controller with a PS eye camera (fairly simple by optical tracking standards, and with what amounts to a high-speed camera) still had a latency in the region of 63 ms...
Then again, by comparison, I've heard claims of worst-case latencies for kinect-based tracking in the region of 300-350 ms... So... I don't know really.
I still suspect the Wii might be quicker for this, if you use IR leds mounted to the headset, because the IR camera already does all the work of determining the position of the LED's, so it's only the calculations needed for depth information that you have to worry about.
But I guess it all depends on the relative amount of latency involved in these methods, and how critical that actually is...
