[jselzer]

I worked in computer vision for my MSc three years ago, and based on the state-of-the-art at that time I would be surprised if Natal going to be as robust as the mass market expects. This kind of thing is notoriously hard and error-prone. The information coming in from cameras is often so noisy and incomplete it is very difficult to tell what you are seeing.

When Natal was unveiled at E3, it was having exactly the kind of issues I expected: http://www.youtube.com/watch?v=NKtpJW9pg84

Also the rumour that they have chopped Natal's onboard CPU and moved most of the processing to the Xbox 360 CPU does not fill me with optimism... Prove me wrong, Microsoft, prove me wrong!

[jimbokun]

Makes me think of the Newton hand writing recognition. It was pretty amazing for its time. But the errors made it worse than a keyboard for most people (as decided by the market, at least). Natal could correctly recognize 90+% of all motions correctly, but if it occasionally makes a mistake that causes your character to die or the opponent to score a goal on you, the public might stick with their Wii-motes for now.

[memoryfault]

I got to play the Natal Ricochet game this summer and I must say that it worked flawlessly. The only thing that didn't feel natural was the response time, which they said they were trying to improve.

[jselzer]

Maybe, but performing well in controlled press tour and demo environments is very different than performing well in a million different kinds of living room and lighting configurations. I'm not saying the tech is going to utterly fail, but these things typically do present a lot of problems for computer vision technology.

[jselzer]

Sorry for the obnoxious link, I had sound off when I posted it! Here is something less headache-inducing: http://www.youtube.com/watch?v=DeZep21f32o

[snprbob86]

Yeah, skeletal tracking is an open research problem. It's hard. Very hard. But believe me when I say, the tech has come a long way since that demo...

[gcb]

well, BAM! here it is