Hacker News Reader

Comment by alexcroox | Hacker News Reader

This is the big reason people don't buy these types of drones (this one certainly isn't the first of it's kind).

It doesn't have obstacle avoidance or it will make a massive deal of it.

Intel last keynote demoed very capable drones. They're showing it off, it won't be long until it hits the market. They named the components RealSense, a miniature kinect I guess. Nice demo at 4:55:

https://www.youtube.com/watch?v=Us0BqJvsF9k&t=295

Now shipping in mobile phone size - http://www.pcworld.com/article/2907352/intel-shrinks-realsen...

Shame the dev kit only has Windows support.

Confirms my thoughts, it really is a miniature Kinect-like kit.

RealSense uses projected IR and thus doesn't work outdoors.

(edit: because short-range and little ambient IR are required)

Thanks for the tip. Are there any portable technology that would work outdoors ?

Vision works for animals, and a lot of people are working very hard to do it for computers. Google "UAV visual SLAM" for examples.

LIDAR works already but is heavy, power hungry and expensive. Not good for these small vehicles.

Yeah, I pictures a Google Drone laughing.

stereo vision, structure from motion, animals use both

Except the end of that video shows it working in a forest... https://youtu.be/Us0BqJvsF9k?t=6m14s

OK, that's cool. I stand corrected. Clearly it can work outdoors in some circumstances. IR texture projectors get washed out by ambient IR from the sun. Thus the choice of a shady forest for the very nice demo. People have also used Kinect outdoors in the evening, etc.

But in the general case you can't rely on this approach because sun.

> because sun.

That's a fair point. So RealSense will probably work at night, in a shady forest (like the demo) or maybe the shady side of a mountain, but as soon as the sun is involved IR becomes unusable. I don't suppose you could switch the other side of the visible spectrum and do UV cameras, could you? Having your data reduced to one dimension per pixel really makes the AI easier.

There's a lot of UV in sunlight too. Hence sunscreen. Many birds can see it. Some birds have patterns on them that are only visible in UV. Fun facts.

I probably have no clue what I am talking about, but wouldn't it be sorta like steering forces?

Seek A Avoid B

In your case, a ton of Bs in terms of leaves, branches and etc.

I guess it would need a wide scan to detect if there is a path either vertically or horizontally.

Well, (A) has a tracking device. All the (B)s don't. If they did, then yeah, this would work.

Right?

Yeah but I am thinking it has some sort of an scanner, sorta like self-parking cars do :).

So once it Avoids (Steers away) B it can continue to Seek (Steers toward) A with the tracking device.

Obviously depends on how well the scanner to pickup obstacles work. You would need some sort of a heat map to detect leaves or branches.

It is a hard robotics problem. Saying that "it would need a wide scan" is a vast underestimation. Try looking up some papers on obstacle avoidance. Also, consider the costs of sticking the additional sensors on the robot, processing that data in realtime, and still powering the whole thing.

I am guessing the biggest problem is the noise. Yeah, it's easy to figure out if there is a giant wall standing in front of you, but what about snow that constantly blocks its lens? Some leaves may be attached to tree so it might be required to circumvent, but some might not even be worth circumventing if it's just a piece of leaf falling.

maybe someone can pull it off with ultra sound? Those can be pretty cheap.

Probably won't have much resolution for leaves. I've used ultrasonic transducers for a college class project almost 15 years ago. Ultrasound worked great for hard surfaces like walls, but softer materials like clothing reduced its range of detection. Snow covered trees, I think will have a similar signature that would be hard to detect except at close range. With a UAV, it could be done, but it'd have to be moving pretty slow I think. You'll probably have better luck with some sort of optical system.

Reply on news.ycombinator.com