switch to mujoco env #381
switch to mujoco env #381
Conversation
There was a problem hiding this comment.
I totally agree in removing the duplicate version of the BatchEnvironment example, but I'm not sure about the design changes in BatchedTask (please see my comments below). In nay case, I'm happy to discuss, as I might be over-engineering things that can be easily handled by other means 👍
| @@ -407,6 +408,11 @@ def __init__( | |||
| self.tasks = [task_classes[0](env=env, sensors=sensors, task_info=task_info, max_steps=max_steps, batch_index=0, **kwargs)] | |||
| self.tasks[0].batch_index = 0 | |||
|
|
|||
| self.frames = None | |||
There was a problem hiding this comment.
I see why you're doing this, but I guess I had a thought on overloading the render() command, so that it produces all the observations we might want to extract from the environment (including, for example, also expert actions). I'm not sure we should be explicit at this level about which type of data are rendered, but I'm happy to discuss it.
|
|
||
| def update_state_views(self): | ||
| for idx, state_view in enumerate(self.state_views): | ||
| updated_state_view = self.tasks[0].env.call(state_view) |
There was a problem hiding this comment.
I know I introduced this notation, but we might prefer to use the env reference in the task sampler rather than that in task 0? I was not sure of what seems cleaner, but I'm currently inclined for the task sampler reference as the better option. Again, happy to discuss what seems best :)
No description provided.