3 ms·

> The piloting is pretty damn impressive, though! Wondering why this had to be manually piloted versus automatically tracking, possibly getting direct feedback

by voisin 3y ago

> The piloting is pretty damn impressive, though! Wondering why this had to be manually piloted versus automatically tracking, possibly getting direct feedback from the vehicle about braking and accelerating and turning, so that the video would be smoother and it could fly closer to the vehicle? I didn’t find the piloting to yield a particularly smooth video, always reacting slowly and inconsistent with its distance and angle.

FirmwareBurner 3y ago
>Wondering why this had to be manually piloted versus automatically tracking Lack of skills, time and resources for a prototype project on such a short notice and tight deadline? A team of 3 drone enthusiasts don't have DJIs resources to deliver on-device object tracking computer vison for the exact time Red Bull gives you public access to their new F1 car, considering they were struggling to even get the drone airborne and fly properly in the first place.
- rurban 3y ago
  Unconvincing. I've build road profile sequences (RPS) for F1 race tracks, and their timings stay pretty constant over the years. Such a RPS could be imported into their drone track SW easily, and then you won't need an expert pilot. Every car has that builtin for years. We also simulated driver profiles, so you could adjust them if they wont give you the real data.
  - jcims 3y ago
    And then they run on a rainy day and everything is off. Rockets have an even more predictable trajectory and yet NASA with all its resources can’t create an autonomous tracker that isn’t filled with all sorts of stiff, robotic movements and corrections. I thought the filming was great under the circumstances. And the mix of wider and close shots was very informative and interesting. Just because a more perfect outcome can be imagined doesn’t imply it’s actually better.
  - FirmwareBurner 3y ago
    >Unconvincing. I've build road profile sequences (RPS) for F1 race tracks Nice flex but your argument is unconvincing. Yes the track layout is well known and predictable, but you only get a few GOs with the F1 car on that track and there's a million ways your untested flight and computer vision tracking algos you set up in 3 days can backfire spectacularly, so why not stick to manual flight from a seasoned drone pilot who can handle all the edge cases?
bri3d 3y ago
They'd be starting from scratch, so it would be way too large an undertaking. Automatic tracking of any kind in drones is still not a commodity; most DIY FPV drones like these use flight control software which is ridiculously primitive by control systems standards, just PID rate control with some aggressive gyro filtering (PT1/2/3 and dynamic notch following motor RPM). There's a lot of academic research in this space - GitHub is littered with thesis projects about drone autonomy of all sorts (inside-out, outside-in, inertial based, visual based, basically anything you can imagine), and there's a lot of commercial success (DJI, Skydio, Autel, and Parrot all have decent inside-out visual odometry and follow-me capabilities), but no project has yet emerged that bridges the gap from research one-off to commodity download for custom purposes.
- FirmwareBurner 3y ago
  >and there's a lot of commercial success Yeah and even there, those commercial automatic tracking systems are only meant to follow you mountain biking at 30 kmph, not driving at 10x that speed in your F1 car. F1 themselves use cameras mounted on cables above the track traveling at high speeds for tracking shots. Having drones fly at that speed and accuracy would be a monumental challenge, not to mention super dangerous and expensive if one were to fall down at that speed and hit a car/viewer. It could probably kill someone easily. Hence my point of these having military applications.
mrandish 3y ago
> I didn’t find the piloting to yield a particularly smooth video Considering the context, I thought it as pretty impressive. I'm sure if it were for a feature film it could have been a bit better in aesthetic composition and framing but a movie would have had a lot of rehearsals, section-by-section planned coordination and also not have been shooting a full lap as one continuous shot. Given that this was a Red Bull marketing video vs a movie, I suspect they didn't have Max for very long (hence having to go ahead with the shoot despite the lousy weather). IIRC, there are even F1 rules strictly limiting the number of laps an F1 team driver can do in an F1 car outside of officially sanctioned practice sessions. Interestingly, we'll soon get to see a feature film interpretation of F1 racing. A major movie starring Brad Pitt was shown shooting around the pits last F1 season and Brad Pitt was interviewed about it on one of the F1 pre-race shows. I expect we'll see some dramatic aerial shots when it comes out. As a side note, these kind of drone shots are often done with a wide angle lens which gives some extra leeway for image stabilization and re-framing in post-production. I expect that was probably the case here because I don't think there was enough room in that bullet-shaped drone for a camera gimbal.

by voisin 3y ago

FirmwareBurner 3y ago

>Wondering why this had to be manually piloted versus automatically tracking Lack of skills, time and resources for a prototype project on such a short notice and tight deadline? A team of 3 drone enthusiasts don't have DJIs resources to deliver on-device object tracking computer vison for the exact time Red Bull gives you public access to their new F1 car, considering they were struggling to even get the drone airborne and fly properly in the first place.

rurban 3y ago
Unconvincing. I've build road profile sequences (RPS) for F1 race tracks, and their timings stay pretty constant over the years. Such a RPS could be imported into their drone track SW easily, and then you won't need an expert pilot. Every car has that builtin for years. We also simulated driver profiles, so you could adjust them if they wont give you the real data.
- jcims 3y ago
  And then they run on a rainy day and everything is off. Rockets have an even more predictable trajectory and yet NASA with all its resources can’t create an autonomous tracker that isn’t filled with all sorts of stiff, robotic movements and corrections. I thought the filming was great under the circumstances. And the mix of wider and close shots was very informative and interesting. Just because a more perfect outcome can be imagined doesn’t imply it’s actually better.
- FirmwareBurner 3y ago
  >Unconvincing. I've build road profile sequences (RPS) for F1 race tracks Nice flex but your argument is unconvincing. Yes the track layout is well known and predictable, but you only get a few GOs with the F1 car on that track and there's a million ways your untested flight and computer vision tracking algos you set up in 3 days can backfire spectacularly, so why not stick to manual flight from a seasoned drone pilot who can handle all the edge cases?

bri3d 3y ago

They'd be starting from scratch, so it would be way too large an undertaking. Automatic tracking of any kind in drones is still not a commodity; most DIY FPV drones like these use flight control software which is ridiculously primitive by control systems standards, just PID rate control with some aggressive gyro filtering (PT1/2/3 and dynamic notch following motor RPM). There's a lot of academic research in this space - GitHub is littered with thesis projects about drone autonomy of all sorts (inside-out, outside-in, inertial based, visual based, basically anything you can imagine), and there's a lot of commercial success (DJI, Skydio, Autel, and Parrot all have decent inside-out visual odometry and follow-me capabilities), but no project has yet emerged that bridges the gap from research one-off to commodity download for custom purposes.

FirmwareBurner 3y ago
>and there's a lot of commercial success Yeah and even there, those commercial automatic tracking systems are only meant to follow you mountain biking at 30 kmph, not driving at 10x that speed in your F1 car. F1 themselves use cameras mounted on cables above the track traveling at high speeds for tracking shots. Having drones fly at that speed and accuracy would be a monumental challenge, not to mention super dangerous and expensive if one were to fall down at that speed and hit a car/viewer. It could probably kill someone easily. Hence my point of these having military applications.

mrandish 3y ago

> I didn’t find the piloting to yield a particularly smooth video Considering the context, I thought it as pretty impressive. I'm sure if it were for a feature film it could have been a bit better in aesthetic composition and framing but a movie would have had a lot of rehearsals, section-by-section planned coordination and also not have been shooting a full lap as one continuous shot. Given that this was a Red Bull marketing video vs a movie, I suspect they didn't have Max for very long (hence having to go ahead with the shoot despite the lousy weather). IIRC, there are even F1 rules strictly limiting the number of laps an F1 team driver can do in an F1 car outside of officially sanctioned practice sessions. Interestingly, we'll soon get to see a feature film interpretation of F1 racing. A major movie starring Brad Pitt was shown shooting around the pits last F1 season and Brad Pitt was interviewed about it on one of the F1 pre-race shows. I expect we'll see some dramatic aerial shots when it comes out. As a side note, these kind of drone shots are often done with a wide angle lens which gives some extra leeway for image stabilization and re-framing in post-production. I expect that was probably the case here because I don't think there was enough room in that bullet-shaped drone for a camera gimbal.