Crazyflie LPS Large Volume Formation & Altitude Stability Issues

[amitx64]

Hi,

I am working on a Crazyflie drone formation using the LPS system in a 15m x 15m x 8m volume. My setup includes multiple Crazyflie drones with more than 8 anchors, and I am using TDOA3 positioning with the following firmware configurations:

CONFIG_DECK_LOCO_ALGORITHM_TDOA3=y  
CONFIG_DECK_LOCO_LONGER_RANGE=y  

Additionally, I set:

{ "tdoa3", "stddev", 0.8 }

to handle more than 8 anchors.

Issue:

With this setup, I am experiencing altitude variations more than 0.5 m during flight. I suspect that altitude estimation from the UWB system is inconsistent in a larger volume.

Possible Solution & Questions:

I am exploring the idea of using barometer-based altitude estimation while using UWB for (X, Y) positioning only. I found the following firmware setting:

CONFIG_DECK_LOCO_2D_POSITION=y  
CONFIG_DECK_LOCO_ALGORITHM_TDOA3=y  
CONFIG_DECK_LOCO_LONGER_RANGE=y
KALMAN_USE_BARO_UPDATE=y
CONFIG_CONTROLLER_PID_IMPROVED_BARO_Z_HOLD=y  

My Questions:

1. Does CONFIG_DECK_LOCO_2D_POSITION force the drone to rely on barometer for altitude while using UWB only for X-Y positioning?
2. Is this an effective approach for stabilizing altitude in a large volume (15m x 15m x 8m)?
3. Has anyone tested this configuration, and are there additional tuning parameters needed for better altitude stability?

[ataffanel]

Hi,

In a geometric way, tracking Z in a big space is more challenging indeed.

To answer your questions:

1. 2D_POSITION is intended to be used in a case where Z can be fed as a static value, for example when tracking ground robots in a flat space. So this is not what you are looking for here. The other settings should help though!
2. Yes, adding more measurements should help! One of the best to have would be the Z-ranger or Flow deck since this will give a very accurate Z estimate (as long as the floor is flat of course).
3. We have tested very big spaces before, and the barometer was helping. Adding or moving nodes so that you get better precision geometrically can help as well (we are in a situation very similar to GPS when using the loco system, see: Dilution of Precision on Wikipedia, this explains why X-Y is much better than Z).

[amitx64]

I also flew the Crazyflie with stddev: 0.9 and found that it had low variation in the Z-axis but was highly responsive when someone moved around the drone or stood near a node.

[ataffanel]

This is due to measurements being affected by non line of sight conditions . To fix that you need to either add new measurement like with a z-ranger or flow deck or add nodes to add measurements with line of sight conditions