Hacker News Reader

Comment by Animats | Hacker News Reader

Very nice. But there's no information about what they use as a position reference. GPS with a pseudolite, maybe?[1] If you have a clear line of sight to your local reference transmitter, 1cm repeatability isn't that hard. If not, you need a much more complex system.

They're at my favorite local intersection, Industrial Rd. and Commercial St.

[1] http://gps.stanford.edu/papers/Thesis/StewartCobbThesis97.pd...

View on HN

The precision comes from custom software and hardware we've created, including a robot on the ground that's equipped with a camera and laser rangefinder to guide the drone along a pre-defined path.

That's relative position, which seems adequate for what's exhibited here.

This makes me a little sad to learn that's the method involved.

All the timelapse video now proves is that the drone was in the proper position to flash its light.

Out of position (as measured by the precise laser rangefinder)?

Then don't flash the light yet, drift/adjust, check again, repeat until correct - then flash.

It's not precision as much as it's "eventual precision".

True. But probably as good as it can be given the limitations of GPS.

Still looks very cool though.

The GIF below the video clearly shows that the flight path is not as smooth as the product. It looks like they overshoot and correct every spot.

Have you seen the video of TrackingPoint sniper rifles? Knowing exactly when to fire the weapon means it hits the target every time. i.e. there are some very important tasks for which eventual precision is incredibly powerful.

We're more focused on the "go to this point in 3d space and take a picture" use case, so we don't care as much about the shape of the path in between the points. It is however a totally smooth to the 4 derivative path we are planning and executing to get to that new position, so its quite mathematically pure... Wind and actuator dynamics get in the way of perfect planning though. We did some work on more continuous velocity paths, which are more in line with my computer graphics and cinematography background, but those will have to wait for a future video. =)

But you care a bit about that path if you're up close to cell towers or power lines, surely? You at least care that your path doesn't cross a minimum distance threshold from the object.

Well, sure. I don't really mean we don't care about the path, we care quite a bit about the path, mainly that it gets from point A to point B directly and doesn't hit things, and that it is somewhat as smooth as possible (easy on the actuators to follow). We have a 3d map of the environment we build from the ground robot to directly test against while we are flying to make sure we don't hit things.

Reply on news.ycombinator.com