A tutorial of using VectorEnv for multiple processes, e.g. the shortest path following example?

[facebookresearch]

A tutorial of using VectorEnv for multiple processes, e.g. the shortest path following example?

[mathfac]

We don't have such tutorial yet.

[gianscarpe]

Hi! Is this still of any interests?

[rpartsey]

Hi @eric-xw @gianscarpe

If you are still interested in using the VectorEnv for shortest path following, this should be pretty easy to implement.

You may define a thin wrapper around of the Env class (let's say, SPFEnv) that should have attributes env (Env instance) and spf (ShortestPathFollower instance). SPFEnv will translate SPFEnv.step(any_action) to Env.step(spf.get_next_action()).

from habitat.tasks.nav.shortest_path_follower import ShortestPathFollower

class SPFEnv(Env):
    def __init__(self, config, goal_radius):
        super().__init__(config)
        self.shortest_path_follower = ShortestPathFollower(
            sim=self.sim,
            goal_radius=goal_radius,
            return_one_hot=False
        )

    def step(self, action):
        best_action = self.shortest_path_follower.get_next_action(
            self.current_episode.goals[0].position
        )
        return super().step(best_action)

Then using VectorEnv you could create as many instances of SPFEnv as you need. For more details on how to use VectorEnv you may check PPOTrainer implementation.

[rpartsey]

Closing due to inactivity. Feel free to re-open the issue, if you still have questions.