Tuesday, September 9, 2008

OCZ NIA in da house

It was here when I got home, so yay!

Initial impressions:
- Very nice packaging
- Decent build quality
- Obviously a work in progress on the hardware side (anytime I see holes in the frame that were obviously used for an external power plug, but lacking a connector behind them, I know things are in their version 1.0 form)
- Software isn't horrible, but definitely needs more options (for instance, the 'Pong' game used to train can only be played with the muscle signal)
- The glance control is abysmal (eye movement detection/EOG)
- Grounding is problematic - I could only use it is I had my hand on the box
- Amplifier overload is waaay too sensitive - if you wiggle out of the detection range, there is a lockout of several seconds
- Needs more and better manual controls
- Needs more and better feedback - depending on how you count them, there are 8 control signals, but you only get an analog trace for muscle activity

All in all, it is pretty much what I expected. Right out of the box it is nearly impossible to control, and you really need to sit with it for a good hour to get a feeling for how to modulate the signals. I will say that it has a decent amount of potential, but from what I can tell, it will not be replacing a mouse anytime soon.

For instance, once I began to get 'good' at the Pong game I started thinking of funny ways to bring up the device in lab meeting. Of course, that made me smile, which blew my control right out the window. When I realized what was happening, I tried to suppress the smile, which must have engaged antagonist muscles because it was just as bad. I had to relax and stare blankly at the screen, which kinda scared me.

Why? Because in order to use this interface, I had to be completely devoid of emotional reaction. Holy Dystopic Futures, Batman! Some nutjob will interpret this as ALL BCIs requiring this 'feature', but that isn't the case. Still, I couldn't help but think of zoned out emo teens staring blankly at their screens, and they disaffectively advance military robots on the other side of the world.

So, there needs alot more work on an adaptive filter for the NIA. That's a problem with most EEG (and less for EMG) devices, and really a requirement when compared to intracortical recording. Basically, EEG is listening in on millions of neurons, and those neurons have 5,000 - 10,000 synapses. The resulting slow wave rhythms are the aggregate signal of all that mess.

People like to use the stadium analogy - that you can tell what is happening in a sports match by listening to the cheer of the crowd and knowing who the home team is (EEG), but you get more information by listening a single person (single unit activity) - but that analogy has one major flaw. It ignores that there are signals of far more diverse frequency and amplitudes, and that inhibitory signals are part of the mix. In essence, you're listening to a stadium at 2,000 feet with an "As seen on TV" sound amplifier, with an unknown distribution of home-to-away fans, of which a random distribution have laryngitis or transplanted dolphin vocal cords, and many of which are watching other games on their portable TVs and cheering more for that game as much or more than the game you are observing. And then there are random nutjobs just running around screaming for completely unknown reasons. (Security must be very lax at this stadium!)

See why single units are better? No nutjobs. Oh wait, no, I mean more info per channel. Pull out 20 random fans and you might be able to figure out things like the score.

Anyhow, enough EEG ranting. If you can filter out all the garbage, which you can't, you could theoretically get a good signal. But, you can't. So devices like the NIA should be looked at purely from a functional standpoint. If I could do a little training, have the cursor follow my eye (for games, this would be the mouse/look control), assign two keys to muscle activity (forward and backward/W and S keys), and alpha waves to left click (shoot), I would be happy. As it stands, I can't. I can get some very basic muscle, poor neural, and essentially no EOG control. That makes me a sad panda.

I'll play with it some more over the week and post regarding any changes. Ah, operantly conditioned association... you will have to be my friend a little longer... (Hats off to the Fetz crowd!)

(I should mention that if I had to decide today between the OCZ Actuator and Phil Kennedy's system, I would go with Phil's. I never posted a review because of some concern regarding the use of lab items for review purpose, which I think is fair, but the Neural Signals device (EOG, EEG, EMG, piezo-switch, light touch switch) was really snazzy. Unfortunately, I can't seem to find a link to it anywhere on their page.)


hydraulic power units said...

If you could show us some picture it would be better.

Erk Subasi said...

haha, that was definitely the best EEG rant ever. you should copyright it ;)

Post a Comment