View Code? Open in Web Editor
NEW
The Official Code for Cross-Modal Domain Adaptation for Cost-Efficient Visual Reinforcement Learning (NeurIPS'21))
Home Page: https://proceedings.neurips.cc/paper/2021/hash/68264bdb65b97eeae6788aa3348e553c-Abstract.html
License: MIT License
Python 86.76%
Jupyter Notebook 13.24%
codas's People
codas's Issues
It seems to be an error that InvertedDoublePendulmEnv-v4 is used instead of -v5, for I cannot find its definition anywhere in this code base.
Related snippets include:
|
elif isinstance(self.env.env.env, InvertedDoublePendulumEnvv4): |
|
qpos1 = ob[0:1] |
|
qpos2 = np.arcsin(ob[1:raw_env.model.nq]) |
|
qpos = np.concatenate([qpos1, qpos2]) |
|
qvel = ob[raw_env.model.nq + raw_env.model.nq - 1:raw_env.model.nq + raw_env.model.nq - 1 + raw_env.model.nv] |
|
elif isinstance(self.env.env.env, InvertedDoublePendulumEnvv5): |
|
qpos1 = ob[0:1] |
|
qpos2 = np.arcsin(ob[1:raw_env.model.nq]) |
|
qpos = np.concatenate([qpos1, qpos2]) |
|
qvel = ob[raw_env.model.nq + raw_env.model.nq - 1:raw_env.model.nq + raw_env.model.nq - 1 + raw_env.model.nv] |
|
register( |
|
id='InvertedDouble-v4', |
|
max_episode_steps=1000, |
|
entry_point='codas.reset_able_mj_env.inverted_double_pendulum_v4:InvertedDoublePendulumEnv', |
|
) |
- id and the entry point should be -v5?
I just come over this work but found that the code is not even (trying to) released.
This may lead to some reproducibility concerns as baseline studies and also degrade the impact of your work.
Any idea @xionghuichen ?