Fix environment reset for full-body mode

[chernyadev]

Closes #37 🕺

[xuai05]

Hello, I noticed that you have uploaded a new branch, fix_full_body_reset.

I have downloaded this branch and reinstalled Bigym within it.
However, I am still only getting upper-body observations.

Is there any additional step I might have missed? Or does this issue still persist?
Thank you for your help!

[chernyadev]

Hi @xuai05, thanks for trying out this branch! Could you share the code you're using to launch BiGym in full-body mode?

Here's a working code sample for this branch:

from bigym.action_modes import JointPositionActionMode
from bigym.envs.reach_target import ReachTarget

env = ReachTarget(
    action_mode=JointPositionActionMode(floating_base=False),
    render_mode="human",
)

print("Observation Space:")
print(env.observation_space)
print("Action Space:")
print(env.action_space)

env.reset()
for i in range(1000):
    obs, reward, terminated, truncated, info = env.step(env.action_space.sample())
    img = env.render()
    if i % 100 == 0:
        env.reset()
env.close()

Expected output:

Observation Space:
Dict('proprioception': Box(-inf, inf, (74,), float32), 'proprioception_grippers': Box(0.0, 1.0, (2,), float32))
Action Space:
Box([-0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1
 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1  0.   0. ], [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1
 0.1 0.1 0.1 1.  1. ], (23,), float32)

Could you try running this and let me know if it doesn't work?

[xuai05]

Hi Cherniadev, Thank you so much for your prompt response and for providing the working code sample. I have successfully obtained the expected results as you demonstrated. However, while visualizing the environment, I encountered an issue where the rendered image appears as follows: I am unsure whether this is due to a potential camera angle configuration in the current code or if it is related to the execution of my script. frombigym.action_modesimportJointPositionActionMode frombigym.bigym_envimportBiGymEnv fromPILimportImage env_class=BiGymEnv action_mode_class=JointPositionActionMode env=env_class( action_mode=action_mode_class(floating_base=False), render_mode="rgb_array" ) img=env.render() print(f"img {type(img)}, shape is {img.shape}") obs, _=env.reset() foriinrange(50): obs, rew, term, trunc, info=env.step(env.action_space.sample()) print(f"obs typs{type(obs)}") img=env.render() image=Image.fromarray(img) image.save(f"output_image_{i}.png") env.close() I would really appreciate your insights on this and look forward to your guidance on resolving the issue.

…

-----原始邮件----- 发件人:"Nikita Cherniadev" ***@***.***> 发送时间:2025-03-16 17:08:39 (星期日) 收件人: chernyadev/bigym ***@***.***> 抄送: xuai05 ***@***.***>, Mention ***@***.***> 主题: Re: [chernyadev/bigym] Fix environment reset for full-body mode (PR #38) Hi @xuai05, thanks for trying out this branch! Could you share the code you're using to launch BiGym in full-body mode? Here's a working code sample for this branch: from bigym.action_modes import JointPositionActionMode from bigym.envs.reach_target import ReachTarget env = ReachTarget( action_mode=JointPositionActionMode(floating_base=False), render_mode="human", ) print("Observation Space:") print(env.observation_space) print("Action Space:") print(env.action_space) env.reset() for i in range(1000): obs, reward, terminated, truncated, info = env.step(env.action_space.sample()) img = env.render() if i % 100 == 0: env.reset() env.close() Expected output: Observation Space: Dict('proprioception': Box(-inf, inf, (74,), float32), 'proprioception_grippers': Box(0.0, 1.0, (2,), float32)) Action Space: Box([-0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 0. 0. ], [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 1. 1. ], (23,), float32) Could you try running this and let me know if it doesn't work? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***> chernyadev left a comment (NeuracoreAI/bigym#38) Hi @xuai05, thanks for trying out this branch! Could you share the code you're using to launch BiGym in full-body mode? Here's a working code sample for this branch: from bigym.action_modes import JointPositionActionMode from bigym.envs.reach_target import ReachTarget env = ReachTarget( action_mode=JointPositionActionMode(floating_base=False), render_mode="human", ) print("Observation Space:") print(env.observation_space) print("Action Space:") print(env.action_space) env.reset() for i in range(1000): obs, reward, terminated, truncated, info = env.step(env.action_space.sample()) img = env.render() if i % 100 == 0: env.reset() env.close() Expected output: Observation Space: Dict('proprioception': Box(-inf, inf, (74,), float32), 'proprioception_grippers': Box(0.0, 1.0, (2,), float32)) Action Space: Box([-0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 -0.1 0. 0. ], [0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 0.1 1. 1. ], (23,), float32) Could you try running this and let me know if it doesn't work? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: ***@***.***>

[chernyadev]

@xuai05 Could you please validate the code you're using? We can't reproduce this behavior on this branch. Let me know if the issue still occurs. Thanks!