Raw observation usage – Osim

NeurIPS 2019: Learn to Move - Walk Around
#1

Hi,

Is it possible to use osim dict observation, rather than proposed projection?

Thanks!

#2

@scitator Anything is allowed for your training, but the model=3D, difficulty=2, project=True, and obs_as_dict=True (i.e. the projection) will be used in the evaluation of Round 1.

#3

Hello, am I right that 14-keys dictionary isn’t available now even in the training phase? step method returns only 4-keys dict (reset method on the other hand can still give 14-keys dict)

for example,

observation = env.reset(project=False)
print("Observation reset (project=False): ", len(observation))
observation = env.reset(project=True)
print("Observation reset (project=True): ", len(observation))
observation = env.step([0] * 22, project=False)
print("Observation step (project=False): ", len(observation))
observation = env.step([0] * 22, project=True)
print("Observation step (project=True): ", len(observation))

will print

Observation reset (project=False):  14
Observation reset (project=True):  4
Observation step (project=False):  4
Observation step (project=True):  4
#4

Hi @jbr. Try obs_dict, reward, done, info = env.step([0] * 22, project=False)

#5

thanks, smsong, mixed up returned objects, my mistake