Meant to be Seen

http://www.youtube.com/watch?v=CFBo-W8HB5k

Here is a flex sensor I am testing out to be used in a DIY data-glove.

These are supposedly made by the same company that made the sensors in the PowerGlove.

Seems to work good enough for me.

Interesting that these increase in resistance when flexed. The >1 million bends life cycle seems great compared to what I've read about resistive ink sensors, but several posts around the 'Net suggest the base resistance of this sensor seems to spontaneously change over time, perhaps as a function of ambient temperature? Or are people just not applying the necessary strain relief and wrecking their sensors? Cybereality, if you could record a few week's worth of baseline readings and report back it would definitely help me make a purchasing decision.

Even if these Spectra Symbol units prove to be fairly reliable, though, $12.95 is downright painful for something you're going to be buying in groups of five. I'm sure the price would be much less if you were buying in industrial quantities, but I think the tiny current world market for these things demands they be fairly expensive at any quantity. When I have some free time later this month I was planning to post some of my own experiments - Velostat-based bend sensors definitely work, and I've managed to get the materials cost down to a few pennies, but I'm still not sure about their durability, temperature sensitivity, and the severity of their non-linear output graph. I bought a sizable brick of Velostat bags, so if anyone out there wants to help me figure out the limitations of this approach, please PM me. I'd be very willing to mail off some raw materials and my current best attempt at a functional sensor in exchange for some feedback.

Nice work Cyber, I know you've been looking into this for a while now, good luck with the project. Quick question, what in your opinion does this technology have over strapping a leap motion or other 3d camera to the front of the rift to track your fingers?

shent1080 wrote:Nice work Cyber, I know you've been looking into this for a while now, good luck with the project. Quick question, what in your opinion does this technology have over strapping a leap motion or other 3d camera to the front of the rift to track your fingers?

Occlusion is a huge problem for camera based solutions. There is lots of interesting research in the area but none of it is really ready for prime time yet. It takes a lot of compute time to do free form optical tracking with no markers so you get high latency to go with your poor tracking.

Leap, the tracking is terrible if it's not pointing off into empty space with your hands within the kind of position you'd have your hands over a keyboard (and even then it only tracks the tips of fingers reasonably, overall hand position tracking is always poor). It has excellent tracking properties within the limited subset it's designed to work with, that's how it stays fast (but it's still cpu intensive). However, that subset makes it a pretty poor choice for tracking from a moving position at an angle it's not designed for.

Flex sensors are cheap, don't suffer from occlusion and take a negligible amount of time to capture data from.

Makes sense, anything to lower the cpu/gpu usage is essential as the amount of items we'll all have soon is going to mean pc upgrades galore. I imagine this is going to be paired with an IMU of some sort to track where the hands are? Such a shame something like the leap won't work well for VR, nice easy solution.

Optical can be great for certain applications, but there are also many problems with it. Occlusion is a big one, and hasn't really been solved. I want to use this with a 360 degree setup (ie Omni, etc.) which means a single stationary camera won't work. You could mount the camera of the headset, but then the field of view is limited to what you can see. You could use multiple cameras, but the processing for this is going to get really high, which will directly effect latency. Using physical sensors seems to be the easier, and actually more robust method.

cybereality wrote:Optical can be great for certain applications, but there are also many problems with it. Occlusion is a big one, and hasn't really be solved. I want to use this with a 360 degree setup (ie Omni, etc.) which means a single stationary camera won't work. You could mount the camera of the headset, but then the field of view is limited to what you can see. You could use multiple cameras, but the processing for this is going to get really high, which will directly effect latency. Using physical sensors seems to be the easier, and actually more robust method.

I would defiantly say that it is the more robust method. I imagine this is for what everyone wanted when they first saw Lawnmower man for the first time, to see your own hands in 3d and interact with a virtual world.

As i see it, at the moment you would need 10 flex sensors to track your finger movements, and 2 imu's to track hand position.

Isn't there a decent camera on the market that you could strap to your rift to track anything in front of you, off screen your hands wouldn't generate a virtual movement but i would imagine that when you haven't got REAL spacial awareness or physical objects to pick up, throw, weild, you would always look at where your hands were anyway to make sure you had grabbed the object.

there are a couple of instances where you would need to look away from your hands which could be solved by a couple of imu's but you would have the advantage of a complete first person perspective with the ability to move your feet and other limbs and actually see them move.

Yes, there are ways to make any method work given enough constraints.

I am just following one type of design to see where it leads.

Fair enough, looking forward to the next vid, good luck.

For making the full glove, take a look at glove liners. They are thin and fit snugly. They are a bit warm, but I have an old pair that I plan to make into a dataglove eventually.

nateight wrote:Interesting that these increase in resistance when flexed. The >1 million bends life cycle seems great compared to what I've read about resistive ink sensors, but several posts around the 'Net suggest the base resistance of this sensor seems to spontaneously change over time, perhaps as a function of ambient temperature?

You can make a calibration program for the sensors. You can even calibrate the homebutton of the iPhone, so there is no shame in this

Ziggurat wrote:You can make a calibration program for the sensors. You can even calibrate the homebutton of the iPhone, so there is no shame in this

You are entirely correct. The winner of the friendly little dataglove race we have going here is not likely to be the person with the best tech, but rather the person with the best calibration algorithms. Industrially produced bend sensors and homemade Velostat units alike are going to have some problems that will be best solved by grabbing a baseline every time the glove is used. "Make a tight fist. *bing* Extend your fingers as far above the back of your hand as possible. *brooing*" Something similar will exist for positional and rotational tracking, something like Hydra shoulder touching. If, after a very brief calibration routine, your glove can then produce data that is significantly superior to what you get out of some other glove, you/we all win.