For my training and executing I use the list instead of the dict, and I didn’t manage to find a provided way to convert the observation dictionary to a list (as it’s required in submission). So I just made my own method for doing so with consistent ordering: https://gist.github.com/Ryan-Amaral/960c291d7cafe928a725f60f36e1f549, hopefully it is of use to someone else.
Was there actually a provided way to do this that I missed? Oh well, it didn’t take too long either way.
Not sure if the order you have would be the same order the assessor server would provide. The gist at [1] is a copy of the same order from the L2M2019Env.get_observations() method.
Links
[1] https://gist.github.com/cmoxiv/a4d1671e28efcf783326755508cbc9ee
It all looks the same to me, I also tested it to make sure the results are the same as using project=True and obs_as_dict=False. Unless I’m missing something, the only difference I can see is the first 3 velocities of the pelvis being divided by self.LENGTH0, not sure what that is. But at the very least your method is much more clean and concise.
FYI, self.LENGTH0 is there to normalize the velocities to model size (i.e. leg length) but is not relevant for the competition as it is set to self.LENGTH0 = 1.
Thanks. Come to think of it, I believe the order does not actually does not matter. In all cases we get the the obstacle_as_dict from which we can pick any filed to use in our controller. It all depends on what we feed to the controller.