51 comments

[ 2.6 ms ] story [ 101 ms ] thread
The video is too stylized for me. I would love for a more steady camera showcasing the drone avoiding obstacles as it flies around. The jump-cuts and music detract from the cool tech I want to learn about.
Some lo tech drone filming from afar would have been so much better. I can understand the excitement translating in the filming style but come on.
(comment deleted)
I just don't understand asking a human to run the camera drone.

Presumably there's more than one of these, have one follow and rely on the pace drone for positioning.

It was pretty hard to watch, but really watching sensing and localization applied in a 3D space is extremely cool.

maybe a camera drone would see the one in front as an obstacle and constantly try to turn away from it :)
To do that reliably, I think both vehicles would need to know their absolute positions with high resolution (spatial and time). Could be a separate research project given how small their planes are.
Many of the fragments in the video would be a lot more interesting if they lasted longer than a second. This is edited to seem exciting, not to show what the actual capabilities and limitations are.
Hey, birds manage do to this without LIDAR and Kinnect so you know it was theoretically possible.
> Barry’s realization was that, at the fast speeds ... the world simply does not change much ...

So is c the speed where nothing changes ? cold.

I guess I was expecting something more like this goshawk flying through a forest, but maybe it's too much to ask for :)

https://www.youtube.com/watch?v=p-_RHRAzUHM&start=100

Nice link. Makes you wonder what biomimicry might bring to drones in a couple of generations. Maybe in addition to a set of propeller motors, you could have a couple of servos running small deployable "fan-like fins" (or is that fin-like fans?) for fast maneuvering where just changing prop speed won't quite cut it.
I would describe that as "Drone dodges, ducks, dips, dives and... dodges".
Fancy camera work, but the obstacle avoidance is not convincing. The drone basically passes by one tree, and that's mostly it. Compare that to any bird that lives its life in the brush. Human built flying machines are fast and can carry huge loads but compared to animals we still lag far behind in control and finesse.
So does the brain and AI. What, I'm off topic?
Is... is the video intended to be a parody of something? It's like they took two shots of a drone turning slightly to avoid a tree in the middle of a field, and decided to hit it with a nuclear bomb of "MAKE IT FEEL MORE EPIC" in editing.
Yeah... I was expecting a drone to be weaving through a forest of trees. Not flying towards a single tree and maneuvering around it.

Still cool, but that music and editing was over the top.

The insight into only needing the differentials in the next 10 meters makes a load of sense and clearly worked.

Why is everyone hating on the video? I thought it was light-hearted and fun.

yeah, only i wished the insight would have been an algo that determines the necessary distance based on the drone's speed
The Navy really wants to get rid of pilots as it seems.
This does go a long way to explaining speeder bikes -- the human rider controls the gross movement, but the bike itself does obstacle avoidance. Otherwise, there's no way a regular stormtrooper could fly those through the thick forests of Endor.
In video games, there are a lot of clever tricks they use to make the player more likely to hit enemies. E.g. the cursor slows down when you get close to an enemy, and bullets bend towards them. See this video: https://www.youtube.com/watch?v=UL1_ht1EdAU

I wonder if tricks like these would work on real life vehicles.

>In video games

in console games

console games are video games too
Isn't console games and video games equivalent expressions? As opposed to computer games? (that is: I agree with you)
Computer games are also video games. That said, I don't think a correction to the original statement was really necessary.
I think the distinction was between console games and PC games, which don't have auto-aim because the player is unlikely to be using a controller.
It's not a useful distinction since many kinds of games use player UX "cheats" like slight warping to physics algorithms, RNG weighting, etc. Most of these features act to take emphasis off of precise, low-level execution, or the pain of a string of unlucky results, in favor of higher-level strategy concerns, fast decision making, and expected values over time. There is an occasional chorus of "simulation purism" from some folks, but really, simulation elements are bulldozed in games constantly and consciously. The actual results are subtle and people don't take much notice when rules are bent, they just experience "fluid control" or "fair outcomes."

Auto-aim is only well-known because it's visually obvious when it occurs - the gun did not point in the direction of the camera.

Maybe you don't get to pilot a speeder bike unless you're somewhat Force-sensitive. I'm sure there's a continuum of Force-sensitivity between ysalamiri and Jedi Masters, with some humans just having great reflexes.
LOLZ @ this comment, most amusing thing I have read all day :)
Nice idea. This is the advantage of having a huge fanbase. You can count on the fans to plug all the plot holes and patch up whatever poor writing the studio throws out.
Whats great is that the immenentizing of the eschaton in this case proves that the Force is, indeed, real.
I think this algorithm would be great for "oh shit a tree" scenarios, but awful for regular flying, since it basically seems to have two modes, blithe disinterest up to a given depth, and panicked evasion thereafter. It would make for a bumpy ride.
Im reminded of the movie Real Genius. This is most applicable as an autonomous weapon. Unavoidable and its going to suck.
The way the trees glow white in the plane view at 1:09 -- could they be using near infrared cameras or filters to help specifically with tree detection? Nowhere in the color video do the leaves look that bright. The paper doesn't seem to mention it though.

http://groups.csail.mit.edu/robotics-center/public_papers/Ba...

It definitely looks like infrared to me. The paper mentions that they use a pair of Point Grey Firefly MV cameras, and the reference manual for those cameras[1] says that "in monochrome models, the IR filter is replaced with a transparent piece of glass."

[1]: http://www.ptgrey.com/support/downloads/10116/

I wish I could be more excited about this but the most probable use for this technology is making it easier for drones to spy on people or for cruise missiles to be able to better fly into someone's window. :-(
could it patrol the city for crime?
I wonder how well this generalizes to other environments, especially cities with dense blocks.
Very impressive. I wonder if this is the beginning of the end of FPV racing as a sport?

If you're unsure what FPV racing is, see http://fpvracing.tv

This is a very disappointing movie, it shows the drone dodging the same tree (or just branch actually) over an over again, there is no indication of its speed, it dodges only one obstacle every time before a cut/edit is done. Is there a reason this was not tried in a forest (as the title would suggest)?
The sad thing about this is as the technology advances, the likely first movers on this tech will be for military purposes.

Do enjoy what appears to be VR piloting at 1.28 however. First great useage of it I have seen.

I instantly thought about Interstellar beginning scene...