Envs difference about gym-pybullet-drones

[azamkamranian]

@JacopoPan thank you for the explanation. yes, I got it.

[JacopoPan]

Hi @azamkamranian,

There are essentially 3 layers of implementation:

• BaseAviary implements all is needed to simulate one or more quadcopter (but not RL interfaces)
• Subclasses BaseSingleAgentAviary and BaseMultiagentAviary extend it with Gym-style action and observation spaces
• Finally, subsubclasses like HoverAviary implement the task-specific methods of each Gym (reward, done, info) and other tweaks like normalization and clipping of the observation, obstacles, etc.

Did that answer your question?

[azamkamranian]

@JacopoPan I know there is so much work but I think a wiki really helps. not only for envs, but a little explaining about the quadcopter physics and ctrl stuff. It really helps to understand functions and ctrl methods for those who are new to drones.

[JacopoPan]

I am typically against over complicating documentations but I admit that this repo ballooned a bit w.r.t. to its initial README.
I'm currently working on a new repo with a couple of simple single agent control tasks that should make it easier to approach RL with a controls background and vice versa. I'll revisit this one once the other is released.

[azamkamranian]

That is so great. I want to use new envs for implementing RL algorithms. So, I am looking forward to the new rep too.

[xiaomaohouhou]

gym.error.UnregisteredEnv: No registered env with id: takeoff-aviary-v0

The terminal reminds me of this error. Can you help me? Thank you!

[JacopoPan]

Hi @xiaomaohouhou

did you pip3 install -e . in the main folder? The registration of the environment is in gym-pybullet-drones/gym_pybullet_drones/__init__.py.