Hi, this is my first time posting in the DIY Drones forum, but I have been a previous user of Ardupilot based drones.
I've been working on an open source project called the Thanics Halo drone, and I'm looking to see if electronics/drone/programming hobbyists would be interested in having a modern drone platform to put their circuits/modules on. Ardupilot is great for getting a drone to fly, but that's pretty much it. I'm trying to make a drone that is very immersive using VR live streaming (with a 360 camera), and spherical obstacle avoidance (using stereo computer vision and laser rangefinders), and of course very hackable, both software and hardware wise.
It is important to note that this project is far from complete, and not fully documented, but here's our GitHub repos (including ECAD/MCAD files): https://github.com/ThanicsRobotics. You can also check out our website thanics.com. The files and pictures on the website depict the V2 drone, and we are now working on V3 (which we will be publishing PCB/MCAD designs soon), which is far more efficient, smaller, and 3D printable.
I'm envisioning having multiple modular "bays" with a suite of interfaces (UART/I2C/SPI/USB) to the host processor (Raspberry Pi CM3), so that hobbyists can easily design their own boards to interface with the drone. This would certainly be more expensive than the Ardupilot based boards, but I'm looking to make a platform for hobbyists to take their drone project a bit further to be comparable (and in some areas better) than the flagship commercial drones (like DJI Phantom/Mavic). Please let me know what you think.
Replies
Well, the in between is that you want both. You want to be able to have a non-realtime 360 experience later, while also having a realtime 360 experience while flying. It is inefficient to strap on a 360 camera and have a gimbaled regular camera. Since each frame transmitted in a 360 video stream is a full 360 image, even with head-jerk like viewing (like a FPV racer would do, or a fighter pilot), they could get a smooth, fast view. This is something a gimbal just could not do, it's latency and mechanical spin is too slow. Sorry if I am being unclear, I am really appreciating your feedback.
Yes, I got that. But that sort of non-realtime 360 experience been easy to do for year with things like Kodak 360 (see our video from two years ago below). And real-time is easy to do with a gimbal. So perhaps there's something in between that I'm missing
Andrew Loomis said:
We are integrating already made real-time 360 stitching cameras like the Ricoh Theta V, and adding the long range, high power live streaming capability for the pilot (for VR-like FPV).
So when you look at your flight after you've recorded your GIMBAL headtracked flight, all you see in the replay is where you were looking during the flight. When you replay your 360 camera flight, you can look in any direction, no matter where you were looking during the flight. You can kind of have a different flight experience each time you replay the flight.
Ah -- got it. I haven't followed that particular application. Not sure I'm totally understanding why that's better than a gimbled camera that follows your head tracker. If I'm right and non-real time 360 capture is a solved problem (with existing cameras, whether app-driving or with dedicated cameras like GoPro Fusion 360 or Kodak), and real-time 360 is better provided by gimbal/headtracker connection, what's the advantage of your solution? This isn't my interest area, so apols if I'm misinformed.
That kind of 360 is not what we are talking about, we are talking about simultaneous spherical video capture. Being able to capture a simultaneous spherical experience of your flight, and even live stream it to the pilot as FPV, it's pretty much a extremely low latency VR experience in the sky (no gimbal lag, just lag of IMU headtracking in smartphone, which is very small). Check out this video as an example (drag the screen to see 360) - https://youtu.be/rLDl0cNZf0w (This is a test we did a while ago with an old prototype, the new camera we're using is the Theta V, 4K instead of 1080p, you'll also want to turn down your volume...) Flight actually starts at 1 minute. Our gimbal also wasn't working that well then either...
360 is usually provided by the app (DJI has that and we did with the Solo). Just pause and rotate, and all the stitching is done on the ground. I'm not quite seeing the need for a specialized drone for that.
Chris, I definitely agree that this is an ambitious project, but I'm interested to know if you think it is a useful endeavor, and why 360 photography on drones (especially consumer drones) has seemed to be overlooked in the market. There are few in the world with more experience in drones than yourself, I'd really appreciate the feedback.
Chris Anderson said:
Wow, thanks for your reply Chris! We're definitely leveraging other open source projects in our project's creation. As for Skydio, their approach is different to ours. Since our drone has a 360 camera, it doesn't matter which way the drone is facing when viewing FPV, so we only need computer vision obstacle avoidance on one side (the side in forward motion), instead of 6 sides like Skydio. We are still using low cost laser rangefinders on the other axis though, for added security. We're going for a lower cost approach (that doesn't need 13 cameras and a NVIDIA supercomputer), that would really only work for a platform using a 360 camera.
Chris Anderson said:
That's an ambitious project! Skydio spent $80m and six years doing something like that. Out of curiosity, why not start with a mature flight controller software stack like ArduPilot or PX4, rather than starting over from scratch?