Closed jzilly closed 6 years ago
Hi Maxime, The wrapper does indeed seem like a much better idea. Will rewrite to make it a wrapper.
With h5py viewer installed, the data will be similarly viewable and better structured. The main reason for h5 would be that it compresses the data and we do expect to generate enormous amount of data with the simulator. For now I would prefer to stick with this option.
What do you mean, an enormous amount of data? 30 FPS 10 minutes 60 seconds = 18K samples
I would much prefer to minimize the number of dependencies. The more dependencies the simulator has, the more likely it is that something will break.
Two things:
I would much prefer that you log the data as CSV or JSON. That way it's introspectable by humans, and we avoid adding a dependency on h5py.
Can you implement this funcitonality as a Gym environment wrapper? That way we can better separate concerns and avoid adding more code to the simulator.
On gym wrappers: https://github.com/openai/gym/tree/master/gym/wrappers https://github.com/openai/gym/blob/master/gym/core.py#L238
Wrappers go into gym_duckietown/wrappers.py