Farama-Foundation / D4RL

A collection of reference environments for offline reinforcement learning
Apache License 2.0
1.35k stars 285 forks source link

[Bug Report] Relocate demonstration seems to be erroneous #196

Open Aequatio-Space opened 1 year ago

Aequatio-Space commented 1 year ago

Bug Description the "relocate-human-v1" and "relocate-expert-v1" seems to have unrealistic ball position.

Code example

import d4rl
import gym
env = gym.make('relocate-expert-v1')
dataset = env.get_dataset()
i = 0
while i < 5000:
    current_state = {}
    for key in dataset:
        if key.startswith('infos/'):
            current_state[key[len('infos/'):]] = dataset[key][i]
    env.set_env_state(current_state)
    env.mj_render()
    i += 1

when running the above code, a ball floating in the air is clearly visible, which does not appear natural. Other Adroit environment(hammer, door) seems normal under this script. Bug-example

System Info

Checklist