Roll/pitch rate controller issue.

[jc-bao]

Problem

Cannot control the yaw angle to zero. (green line)

Currently, the lower-level attitude rate controller is running L1 adaptive control.

The yaw angle controller is:

        rpy_drones = geom.quat2rpy(self.quat_drones)
        yaw_drones = rpy_drones[..., [2]]
        vy_target = - yaw_drones * 10.0

(can not use tilt-prioritized controller since no target angle in other dimensions. )

[jc-bao]

check1: attitude rate controller precision

The attitude rate in yaw is relatively more precise than other dimensions.

[jc-bao]

thoughts: can we switch to the attitude controller?

This would bypass the yaw control problem.

[jc-bao]

possible solutions

1. use RL to control the yaw angle. -> more burden in sim2real and RL training
2. implement the attitude controller -> less agile.

[jc-bao]

2D case test

Learned policy: :heavy_check_mark:

[jc-bao]

Run yaw controller at 500Hz