Episode vs timesteps about gym-pybullet-drones

[JacopoPan]

Hi again @nbenave,

"episodes" are 5 seconds (in simulation clock) flights, in most aviaries in this repo. Of course that is arbitrary and application dependent. Once the episode is over, env.step() will return done=True.

The stopTrainingRewardThreshold set at 0 is an upperbound that, in the current implementation cannot be achieved (but again, that is application/implementation/reward shaping dependent as well).

W.r.t. 5 and 6, that depends on how trajectories/batches are used by each RL algorithm. You should look into the implementations of SAC, PPO, etc. in stable-baselines3.

[sunjeet-jena]

Is there a way we can increase (in simulation clock) time ? I need more samples near the goal position.

[JacopoPan]

Do you mean the episodes length in seconds, @sunjeet-jena?
There is attribute self.EPISODE_LEN_SEC in both BaseSingeleAgentAviary and BaseMultiagentAviary classes.