Help with rendering of quadrotor example

[itsLydt]

Hello, just getting familiarized with the suite but unfortunately having some issues with the quadrotor example from the readme.

I ran the commands
cd ./examples/ # Navigate to the examples folder
python3 pid/pid_experiment.py --algo pid --task quadrotor --overrides ./pid/config_overrides/quadrotor_3D/quadrotor_3D_tracking.yaml

after following the installation instructions. I expected to see the same animation as in the readme. The path is drawn in red, but then disappears. The simulator UI flashes several times and the quadrotor seems to glitch and teleport around randomly. After that, it flies around but doesn't seem to perform the figure 8 pattern I expected, although it is somewhat difficult to tell as the red path has disappeared.

Is there possibly some problem with my installation or something I am missing? I did not encounter any errors while following the installation instructions.

My system is:
Linux Ubuntu-20 5.15.0-122-generic #132~20.04.1-Ubuntu SMP Fri Aug 30 15:50:07 UTC 2024 x86_64 x86_64 x86_64 GNU/Linux

python --version
Python 3.10.16

Attached is the output of the example
sim.txt

Thanks in advance for any help!

[Federico-PizarroBejarano]

Hi @itsLydt, thanks for using our gym! I think an easier way to run the examples is to cd examples/pid and then run ./pid_experiment.sh. I now realize that the README does it in this other way, but by looking at and modifying pid_experiment.sh you should have an easier time. However, both methods of running it are actually equivalent.

I also realized the animation in the README is from some older configurations, which should still be possible but is not the trajectory we are running in the PID example. The PID example currently is much more aggressive. The red line disappears and the simulation flashes a ton of times because we reset the environment in multiple places (probably too many), but that is to be expected. Finally the drone teleports to the real starting location and flies around, but I agree it is hard to tell if it is following the trajectory. This should be fixable but honestly I don't use the GUI very often, but I will take a look. In summary, what you're seeing is correct, is running a different trajectory than the README (which maybe we should fix) but could be visualized better (which I will look into).

Let me know if you have any other questions!

[Federico-PizarroBejarano]

No worries! Honestly we don't have much documentation apart from the docstrings on the files and functions, but I am happy to assist you with any questions you may have. As you may have seen, the examples generally follow a structure:

1. A shell script (like rl_experiment.sh) which has all of the args for running an experiment. Easy to modify to switch environments or other arguments/overrides.
2. An experiment file (like rl_experiment.py) which actually runs the experiment.
   • Usually starts with a config factory which creates the config from the config files and overrides
   • Then creates the environment and controller objects (and safety filter if necessary)
   • Then runs the experiment using the BaseExperiment class. This outputs the results in a standard format
   • Typically includes some custom plotting but this is just interpreting the results from the BaseExperiment class. The plotting can be changed for your use case.
3. Sometimes there is a training script (like train_rl_model.sh) which trains the RL agents.

Once you are comfortable with the shell and experiment script, and check out the structure of the output from the BaseExperiment class, you should be ready to run your own experiments. Let me know if you have any other questions! Also happy to hop on a Zoom call if that would be useful.

[itsLydt]

Thanks, I am trying to run a series of experiments in which I add input and estimation disturbances. Is there an example I can look at to see how this is done and what types of disturbances are available?

Edit - found the list of disturbance types in envs/disturbances.py.

[Federico-PizarroBejarano]

Happy you found it! You can see it being used in this YAML file:

safe-control-gym/examples/no_controller/verbose_api.yaml

Line 69 in 0e5f4f1

disturbances:

[itsLydt]

ah thank you so much. I assume the observation disturbance would be noise in the state readings? What is the difference between a dynamics disburbance vs an action disturbance?

[Federico-PizarroBejarano]

As far as I remember the action disturbance modifies actions before they're applied, while the dynamics disturbance directly pushes around the robot. Is this correct @adamhall ?