Unable to fly Crazyflies w/ AI deck

[davidjmflores]

Hello! Over the past couple months my team and I have had difficulties getting the Crazyflie 2.0 drones up in the air.

For context, we are using the AI deck and motion capture marker deck (with four reflectors). We are communicating with the Crazyflie over TCP (sending px, py, pz, y) at 60 hz and w/ a Vicon motion capture system for ground-state truth data. Our setup is w/ ROS 2 Foxy, Python3.8, and Ubuntu 20.04. Lastly, the test bed works with other drone systems. We use the CrazySim Gazebo-based simulator for code verification and it's worked well so far.

Unfortunately, when we run on the real CFs we get one of two behaviors. Upon takeoff, the CF will either immediately flip over or it will start hopping up and down off the ground and eventually halting. In the hopping case, we can usually reach an altitude of 0.3m.

So far we've tried incrementally increasing the thrustbase of 36k up through to 50k, adjusting the firmware PID's kvz and kpz, adjusting CF weight distribution, running different CFs, and changing our commands from velocity to positional. Lastly, we've already verified that the coordinate frames are matching. Any advice would be greatly appreciated. I'd be happy to provide any plots necessary for understanding and answer any questions.

[gemenerik]

Sounds like a thrust problem to me. The flipping points at running out of control authority, and the hopping to just not having enough thrust to lift off at all.
On paper your setup should fit within the payload budget, but wear of motors, props and battery can eat into that margin. Can you weigh your drone everything attached? Do you have spare parts to try? It's also worth checking your firmware version against PR #1465. The new thrust compensation keeps thrust steadier across battery voltage but lowers the peak available at high charge, which could push a marginal setup over the edge.