search

jc-bao / policy-adaptation-survey

This repository is for comparing the prevailing adaptive control method in both control and learning communities.
Apache License 2.0
7 stars 1 forks source link

Resolve unstable hovering performance given the attitude rate controller. #26

Closed jc-bao closed 1 year ago

jc-bao commented 1 year ago

@Week23-4

What it is

The learned policy in a new environment cannot hover at a fixed point stably. The drone cannot keep a static orientation and just fly back and force.

Why Important

The upstream task of stabilizing objects in high speed.

Why would this happen

Provide assumptions and possible solutions.

Good exploration Bad exploration
image image

What could be next

Stabilize object

jc-bao commented 1 year ago

Trail: Exploration -> Auto reset.

Result: attitude control become quite precise now. But the learned controller spinning its self up and down, which lead to a large tracking error.

For the action, the thrust is always saturated.

jc-bao commented 1 year ago

Issue resolved at Issue #29