Hey folks - my beast of an full R1 in-depth video - and text review is still inbound! Trying to edit down what is currently a 30-minute long monster review full of multiple sport tests and much more. Working to get it closer to 15 minutes or so. Thus, I figured I'd upload this unedited continuous clip in the meantime. Enjoy!
Me too! I've been pushing for it! Obviously, it chews up a ton more storage space, but it's not a massive amount if you're prepared to offload soon thereafter.
I can imagine Apple / Amazon utilising this technology for next generation Maps. Imagine 3D views of footpaths, trees, cars etc. Basically a next-gen, higher resolution maps for most likely 2019 - 2020
I'm gonna go out on a limb here and say DJI is probably 2-4 years behind. The hardware DJI uses to do object tracking is much simpler and doesn't use deep neural nets. Right now there's basically 2 platforms for doing DNN on a mobile device. There's Nvidia with their GPU powered board and there's Intel. Here's the thing, the hardware isn't really the magic sauce for what skydio R1 does. The magic is the model they produced with DNN. To do that you really need to be an expert at AI and DNN. That's not something a typical engineer can master in a few months or even a few years. A good senior level engineer can get competent in 1-2 years, but that's far from mastery. To master DNN in a specific discipline like obstacle avoidance takes more than 5 years. Geoffrey Hinton the godfather of DNN has been at this for more than 20 years. Just about everyone in the AI field would agree geoffrey hinton is the master of DNN right now. Only way DJI can catch up is to build a system that uses virtual training. You'd need some pretty expensive hardware with hundreds of GPU, which no one can get their hand on today due to crypto currency miners. Nvidia says supply will remain short for 2018 and likely into 2019. That means if DJI were to start the project today, they couldn't even get the hardware they need to train a DNN. They could use Amazon AWS deep learning instances, but that is also expensive. Until geoffrey hinton gets capsule networks to the point where it's better than convolutional and recurrent networks, Skydio has a 2 year advantage.
Hey folks - my beast of an full R1 in-depth video - and text review is still inbound! Trying to edit down what is currently a 30-minute long monster review full of multiple sport tests and much more. Working to get it closer to 15 minutes or so. Thus, I figured I'd upload this unedited continuous clip in the meantime. Enjoy!
the pole at the end of the path is a stroke of genius for testing!
Shockingly best example review of actually using the R1
This is exactly what I imagine the TV show COPS will look like in 3 years.
Very cool overlay! Thanks for sharing.
Well done Ray, hope you enjoy Freo! Keep it coming!
Really impressive. Thanks the the unique look Ray!
Incredibly impressive.
Love this viewpoint. I hope that they add it as a feature in the future.
Me too! I've been pushing for it! Obviously, it chews up a ton more storage space, but it's not a massive amount if you're prepared to offload soon thereafter.
Fantastic!!!
Impressive!
I have liquid maths coming out of my ears while watching this. Amazing code work I'm sure!
Great video! Here is the future of human technology. I believe that in years the price will drop.
wow - this is stunning - amazing algorithms!
I can imagine Apple / Amazon utilising this technology for next generation Maps. Imagine 3D views of footpaths, trees, cars etc.
Basically a next-gen, higher resolution maps for most likely 2019 - 2020
Scary smart. If it mated with Sony Aibo dog, you’ve got an actual Black Mirror episode right there!
Everyone keeps talking about this Black Mirror thing...one day I'll have to check it out!
DC Rainmaker it’s great. Four (short) seasons on Netflix. All future science focused. Every episode very different. Good stuff.
Very cool!
Awesome!
awesome
that’s leaps and bounds over the mavic pro tracking features you reviewed a year or two ago.
Hey, that is the South Fremantle power station!
So it has 13 cameras for avoidance.. what if there are other drones in the air? How will it react? Duck, dodge, shoot to kill?
I kept thinking what if the drone went down on the other side of that tall barbed wire topped fence.
Wow
DJI has serious competition
Amazing! Now the interesting thing is how will DJI respond to this!
How is the noise level compared to other high end drones?
It’s in the same ballpark, but still a bit higher than the Mavic Air. Not massively so, but a little bit.
I'm gonna go out on a limb here and say DJI is probably 2-4 years behind. The hardware DJI uses to do object tracking is much simpler and doesn't use deep neural nets. Right now there's basically 2 platforms for doing DNN on a mobile device. There's Nvidia with their GPU powered board and there's Intel. Here's the thing, the hardware isn't really the magic sauce for what skydio R1 does. The magic is the model they produced with DNN. To do that you really need to be an expert at AI and DNN. That's not something a typical engineer can master in a few months or even a few years. A good senior level engineer can get competent in 1-2 years, but that's far from mastery. To master DNN in a specific discipline like obstacle avoidance takes more than 5 years. Geoffrey Hinton the godfather of DNN has been at this for more than 20 years. Just about everyone in the AI field would agree geoffrey hinton is the master of DNN right now. Only way DJI can catch up is to build a system that uses virtual training. You'd need some pretty expensive hardware with hundreds of GPU, which no one can get their hand on today due to crypto currency miners. Nvidia says supply will remain short for 2018 and likely into 2019. That means if DJI were to start the project today, they couldn't even get the hardware they need to train a DNN. They could use Amazon AWS deep learning instances, but that is also expensive. Until geoffrey hinton gets capsule networks to the point where it's better than convolutional and recurrent networks, Skydio has a 2 year advantage.
Interesting info!
Want.
I can't believe this was 5 yrs ago. Mini 3 pro active track sucks balls and it's only a yr old.
Purple - life, red - death :D
MINECRAFT IRL
Real life GTA!!!
very noisey...