Seeking advice for max accuracy loiter

Hardware Multirotors
1

I’m putting together my first drone build (quadcopter) and the purpose of this drone is to be a technology demonstration platform. In other words, I’ll be testing some proprietary new tech that’s attached to the drone. I’ve been researching what to do in terms of the drone platform (turnkey is not a good option at all) and it definitely seems like an ardupilot controller is what I need to accomplish what I’m trying to do. Still, I’d like some advice so I make good initial choices for hardware.

The premise is simple, although a bit unusual. The drone is going to lift off, climb to only about 2-3m altitude and translate into a target zone. Then it’s going to loiter in that target zone as accurately as possible for as long as the battery lasts. Realistically, the maximum allowable XY error is about +/- 10cm, but +/-5cm would be a lot better. And we are talking long term absolute position here. Altitude is more forgiving, but it needs to be +/-20cm. Yaw is of the least concern (if it could stay within +/-20 deg that would be sufficient). This will be done primarily indoors, although the ability to work outdoors later would be advantageous.

With GPS out, it seems obvious I’m looking at a quality optical flow and LiDAR sensor. Either combination or separate. The altitude may extend up to 10m in the future, but for the moment, it will not exceed 3m. Based on my cursory research, the ARKflow optical flow sensor looks very good, but it’s also pretty spendy. I am looking for a little bit of hardware advice as well as some implementation advice by anyone with some experience in this realm. I plan to use one of the more barebones PixHawk controllers.

My primary concern is long term accuracy in loiter mode. LiDAR should be self correcting, but I worry about optical flow. Obviously it will error correct in the short term, but I assume it will also have a tendency to drift over time. Manual corrections can be applied, but the longer it can maintain that +/-5cm XY, the better. I can also provide a high contrast ‘target’ on the ground if that would help mitigate drift. Like a cross or bullseye or whatever would work best for the camera. This will be a quadcopter specifically designed just to hover efficiently and our tech payload is lightweight, so we are looking at 30m-1hr of flight time.

I’ve considered other more ‘weird’ options to control position, but they tend to be complex and they aren’t flexible to reconfigure if I want to relocate the testing. I’m hoping some folks out there can give me a reality check before I invest in some hardware. FYI, I’m an electrical engineer with some programming experience, but I am totally new to drones and ardupilot. Thanks in advance.

2

im thinking a couple of posts, with string - excellent XY, yaw and altitude control…
accuracy - couple of mm - including indoors
loiter duration - days

apols - thats not helpful, but it seems you need to break the problem into two parts.

  1. absolute accuracy of you drone/quad whatever
  2. accuracy of your positioning system.

can you get your position from an external indoors system - eg 3d pos calc off drone with telemetry

Your overall accuracy is going to be the sum. Tune the up individually then merge them ?

1 Like
3

First you need to correctly configure and tune a drone using a normal GNSS receiver. The sensors and the own system dynamic need to be “learned” and most sensors “linearized”
Once that is done you can change the GNSS receiver and replace it with whatever positioning system you want and need.

The good news is that there is a software assisted way of achieving all that:

1 Like
4

Thanks for the input. I fully intended to go through a normal tuning procedure to get the vehicle trimmed and controllable. Thanks for the link to MethodicConfigurator, I will check that out for sure.

I guess what I am mainly asking is whether there’s any way my proposed configuration of optical flow +LiDAR can achieve the kind of accuracy I’m talking about and anything I can do about long duration drift in XY beyond manually repositioning with control inputs. There are obviously other ways to achieve this with different methods, but the optical flow and LiDAR seemed like the easiest place to start and it would be highly flexible if it can do it. The next approach would be something like IR-Lock, but that system is really pricey at ~$1200.

5

Why can’t you use RTK GPS?

6

Primary testing will be indoors.

7

ah i missed that part.

8

We did Quick-tune, autotune and systemidentification indoors. All work without GPS, but we did had a GNSS receiver attached anyways.