Loading...
Checkpoint:
default
Observations:
No gamepad connected (please press button or move axis to connect)
# of Quadrotors
Speed: 1x -
0
suggestion
⬇️ ⬆️
Transform:
Drop RLtools HDF5 checkpoint here!

: A Foundation Policy for Quadrotor Control

Jonas Eschmann, Dario Albani, Giuseppe Loianno

Paper on ArXiV Code on Github
This web app allows you to torture the foundation policy using different quadrotor models, reference trajectories and disturbances. You can reset the episodes by pressing "Sample Initial States". You can simulate step-by-step using "Pause on Reset" and then "Step" throught the episode.
The "X Offset" spreads out the quadrotors (click it to change to Y/Z offset). You can drag the sidebar next to the 3D view to reveal the menu for loading dynamics parameters. You can select one of the presets in the dropdown as well as load parameters from disk, following the l2f json format. The last row allows you to perturb dynamics parameters based on a simple formula (hover over the formula input for a description of the variables).

Reproducing Oscillations Caused by Linear Velocity Feedback Delays

You can actually reproduce the finding of the oscillations on the non-EKF-based quadrotors (Section 2.4.1) by replacing LinearVelocity with LinearVelocityDelayed(2) and loading dynamics parameters of a quadrotor with a thrust-to-weight ratio > 2 (any one except of the Crazyflie).
You can just paste Position.OrientationRotationMatrix.LinearVelocityDelayed(2).AngularVelocityDelayed(0).ActionHistory(1) and hit "Sample Initial States".