You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
During the simulation, the robot arm exhibits object penetration, passing through objects instead of interacting with them correctly through physical collisions (e.g., pushing or making contact as expected
num_up_409600_pt_-60.85.mp4
Code example
import os
import gymnasium as gym
os.environ["MUJOCO_GL"] = "egl"
env = gym.make('Pusher-v5', render_mode="rgb_array", width=1280, height=720)
observation,_ = env.reset()
obs = observation
k = 0
frames = []
for_in range(1000):
with torch.no_grad():
obs_array = np.array(obs)
obs_tensor = torch.tensor(obs_array, dtype=torch.float32).unsqueeze(0).to(device)
action, logprob, _, value = agent.get_action_and_value(obs_tensor)
next_obs, reward, terminated,ter , info = env.step(action[0].cpu().numpy())
k+=reward
obs=next_obs
img = env.render()
frames.append(img)
# Fix for multi-env outputsif terminated or ter :break
video= f"num_up_{tong_ptv}_pt_{round(k, 2)}.mp4"
imageio.mimsave(video, frames, fps=30)
Describe the bug
During the simulation, the robot arm exhibits object penetration, passing through objects instead of interacting with them correctly through physical collisions (e.g., pushing or making contact as expected
num_up_409600_pt_-60.85.mp4
Code example
System info
!pip install gymnasium_robotics[mujoco-py]
gym.version == '1.2.0'
https://www.kaggle.com/
python --version == Python 3.12.12
Additional context
No response
Checklist