Question
Hi, Sir,
There seem to be two ways to load a dataset in d4rl:
1.
env = gym.make(name)
dataset = d4rl.qlearning_dataset(env)
2.
env = gym.make(name)
dataset = env.get_dataset()
In fact, if we use the second method for dataset loading, We can use for k in ['observations', 'actions', 'rewards', 'terminals', 'infos/goal'] to get information about the goal in antmaze. However, this method cannot index with 'next_observations' as the key directly. If I use the first method, then I can't get the goal directly through 'infos/goal'. Is there a way to get the goal directly from the state in the trajectory?
Question
Hi, Sir,
There seem to be two ways to load a dataset in d4rl:
1.
env = gym.make(name)dataset = d4rl.qlearning_dataset(env)2.
env = gym.make(name)dataset = env.get_dataset()In fact, if we use the second method for dataset loading, We can use
for k in ['observations', 'actions', 'rewards', 'terminals', 'infos/goal']to get information about the goal in antmaze. However, this method cannot index with'next_observations'as the key directly. If I use the first method, then I can't get the goal directly through'infos/goal'.Is there a way to get the goal directly from the state in the trajectory?