search

PaulDanielML / MuJoCo_RL_UR5

A MuJoCo/Gym environment for robot control using Reinforcement Learning. The task of agents in this environment is pixel-wise prediction of grasp success chances.
MIT License
458 stars 55 forks source link

How to use Offline_RL to train this agent? #24

Open ZhimengChi opened 11 months ago

ZhimengChi commented 11 months ago

Hi, thanks for sharing this amazing project! I have successfully run example.py and example_agent.py. Here are two questions I currently have:

  1. It really confused me that example_agent.py seems only do multiple simulation steps instead of learning to grasp things, according to its code. I'm really new to mujoco and RL, sorry if I have missed some obvious thing here.
  2. How to use the models in Modules.py and Offline_RL to get this agent learning? In train.py it has a load_path ("../xxx.pt"), is it generated from another .py file?

Much appreciate.

guanxiaoyuan commented 10 months ago

Offline RL training should use generate_data.py to generate transactions. However, I encounter a new problem: No such file or directory : '../DQNRESENT;R_0.001_OPTION_ADAM_H_200_W_STEPS_35000_BUFFER_SIZE_2000_BATCH_SIZE_12_SEED_81_9_7_2020_9_52_weights.pt'

guanxiaoyuan commented 10 months ago

Offline RL training should use generate_data.py to generate transactions. However, I encounter a new problem: No such file or directory : '../DQNRESENT;R_0.001_OPTION_ADAM_H_200_W_STEPS_35000_BUFFER_SIZE_2000_BATCH_SIZE_12_SEED_81_9_7_2020_9_52_weights.pt'

reset the LOAD_PATH = None; problem resolved.

lllwin commented 7 months ago

When I ran example_agent.py , I met a problem saying " failed to initialize OpenGL", had you met this problem? And how did you solved that? This confuses me for a long time.

QWelkin commented 7 months ago

When I ran example_agent.py , I met a problem saying " failed to initialize OpenGL", had you met this problem? And how did you solved that? This confuses me for a long time.

I met this problem too. I had looking for solutions for a long time but the problem have not been solved yet. https://github.com/openai/mujoco-py/issues/187 Maybe you can check this out and find some solutions.