We are developing a technology for automated landing navigation without GPS or better to say with higher accuracy than standard GPS signal. I think, there can be a lot of various utilisations of such technology in dyidrones world. After a year of development, when we tried many different approaches, we developed current prototype, which is offering some reasonable results.
In video you can see our prototype drone (DJI F450 with E300, 3DR APM2.6). Which is navigated by our system in full automatic mode (no GPS lock or RC). System consists of ground unit (small paper box with black tape on the ground, touched by drone) and onboard unit, which provides signals for APM. The onboard unit weight is less than 30g and power consumption is less than 50mA.
The technology is not yet bulletproof and stable. It requires a lot of time and effort to tune it up. But our resources for development are shortening. So with this post we are requesting some feedback from the community. If there is demand for such technology, please write a comment, how much you expect from the price or technology accuracy itself.
Thanks
Comments
@Juraj,
I've been working on this for a while now and have finally managed to get a first functional prototype, using the solution suggested by Randy (camera and onboard computer). The sensor works fine with pretty good accuracy. Currently I'm testing it in real conditions and working on the integration with ArduCopter: the driver is already implemented and tested, and a custom flight mode is now under development. The first target system is APM2.
I'd be happy to help with anything you need. And I'd also be happy to get more details on your solution!
@Juraj,
There's some documentation of the code on the developer wiki. Maybe the section, "Understanding the ArduCopter code" might help although it's not really extensive.
The best way to get more information on integration is by emailing drones-discuss@googlegroups.com.
ArduCopter does use inertial navigation (mixed with GPS) even on the APM2 but I agree that trying to blend camera (or whatever you're using) data plus inertial data would be very very difficult with the limited CPU on the APM2. Pixhawk or using a companion computer (like the Odroid or Intel Edison) is really the way to go.
I'm pretty sure there will be at least two more attempts to develop precision landing in ArduCopter within the next 6 months and they'll all face the same issue on how to integrate with the flight controller code so I'm sure we can come up with a method that works for everyone. Basically I think it needs to integrate with the landing code that's used in Land, Auto and RTL flight modes.
@Randy - we are trying to develop it also on fork of Arducopter code. Possibly for the future, we want to share this code with community as a new type of flying mode (kind of alternative to RTL) using our sensor technology.I tried to to find some relevant Artupilot/Arducopter project architecture description to do it right way. But I am missing this kind of stuff, or I am unable to find it. Are you the right person to communicate about this?
Doing this and provide our sensors through kind of kickstarter project we can let other people find maybe better solutions or approaches, because as I said the technology has good results, but definitely we can not say now that it is stable.
Inertial navigation would be really useful as additional information to tune up the landing procedure, but according to discussions found on diydrones APM2.6 has not enough processing power to handle it. Our goal was to use as least additional HW as possible. We developed kind of new type of sensor, that is providing navigation information for landing.
How does it work if it doesn't use a camera system? Inertial navigation?
Precision landing will definitely very useful so it's very worthy goal your pursuing. As a minimum it'll stop people from having to reposition their vehicle's slightly during the final descent of RTL but there are lots of other reason (like landing on a charging station) that will make use of this ability.
If I were to do it, I would use a webcam + companion computer (like an Odroid or Intel Edison) but there are multiple solutions I think.
I can see this being valuable, particularly if the target is something everyone has. If the target is too specialized the value is limited, but if I can set an iphone or iPad on a table with a particular .jpg image on the screen and have the drone recognize it, then I can send anyone the image and expect the drone to land precisely where they want it to, or deliver something exactly to a particular spot on a sheet of paper printed by the recipient.
I'm not sure what is in your paper box or how the copter recognizes it, so I'm not sure how applicable your approach is to the real world.
No camera signal. We have tried it also this way, but there were issues with camera stabilisation and picture recognising has high demand on CPU power.
So its based on camera signal?
-
1
-
2
of 2 Next