Are you sure you want to delete this task? Once this task is deleted, it cannot be recovered.
Zed-Wu 0890cf8398 | 1 year ago | |
---|---|---|
configs | 1 year ago | |
example_mani_skill_data | 1 year ago | |
full_mani_skill_data | 1 year ago | |
full_mani_skill_state_data | 1 year ago | |
mani_skill_learn | 1 year ago | |
scripts | 1 year ago | |
scripts10/full_mani_skill_example | 1 year ago | |
submission_example | 1 year ago | |
tools | 1 year ago | |
work_dirs | 1 year ago | |
.gitignore | 1 year ago | |
LICENSE | 1 year ago | |
P20_matrix_256.npy | 1 year ago | |
P_matrix_256.npy | 1 year ago | |
README.md | 1 year ago | |
create_Pmatrix.py | 1 year ago | |
data_download_urls.txt | 1 year ago | |
environment.yml | 1 year ago | |
requirements.txt | 1 year ago | |
setup.py | 1 year ago | |
user_solution.py | 1 year ago |
This is the offical repository of team Fattonny for the No Interaction Track in SAPIEN Open-Source Manipulation Skill Challenge.
Our technical report is published on ICLR 2022 Workshop on Generalizable Policy Learning in Physical World.
Our repository is based on the repository ManiSkill-Learn, which is a framework for training agents on SAPIEN Open-Source Manipulation Skill Challenge, a physics-rich generalizable manipulation skill benchmark over diverse objects with large-scale demonstrations.
ManiSkill-Learn requires the python version to be >= 3.6
and torch version to be >= 1.5.0
. We suggest users to use python 3.8
and pytorch 1.9.0
with cuda 11.1
. The evaluation system of ManiSkill challenge uses python 3.8.10
.
To create our anaconda environment, run
cd ManiSkill-Learn-main
conda env create -f environment.yml
Then please follow the instructions in ManiSkill-Learn to install ManiSkill-Learn and ManiSkill.
We used the same tool provided in ManiSkill-Learn to train our ensemble models.
To change the permission of the scripts, run
chmod 777 ./scripts10/full_mani_skill_example/*sh
To change the gpu id used, modify the --gpu-ids=xx
in sh.
For drawer tasks, we train 20 models (10_0 -> 10_19) and combine them together to one ensemble model:
./scripts10/full_mani_skill_example/train_pnt_bc_drawer10_0.sh
./scripts10/full_mani_skill_example/train_pnt_bc_drawer10_1.sh
./scripts10/full_mani_skill_example/train_pnt_bc_drawer10_x.sh
...
./scripts10/full_mani_skill_example/train_pnt_bc_drawer10_19.sh
For door tasks, we train 20 models (10_0 -> 10_19) and combine them together to one ensemble model:
./scripts10/full_mani_skill_example/train_pnt_bc_door10_0.sh
./scripts10/full_mani_skill_example/train_pnt_bc_door10_1.sh
./scripts10/full_mani_skill_example/train_pnt_bc_door10_x.sh
...
./scripts10/full_mani_skill_example/train_pnt_bc_door10_19.sh
For chair tasks, we train 20 models (10_0 -> 10_19) and combine them together to one ensemble model:
./scripts10/full_mani_skill_example/train_pnt_bc_chair10_0.sh
./scripts10/full_mani_skill_example/train_pnt_bc_chair10_1.sh
./scripts10/full_mani_skill_example/train_pnt_bc_chair10_x.sh
...
./scripts10/full_mani_skill_example/train_pnt_bc_chair10_19.sh
For bucket tasks, we train 20 models (10_0 -> 10_19) and combine them together to one ensemble model:
./scripts10/full_mani_skill_example/train_pnt_bc_bucket10_0.sh
./scripts10/full_mani_skill_example/train_pnt_bc_bucket10_1.sh
./scripts10/full_mani_skill_example/train_pnt_bc_bucket10_x.sh
...
./scripts10/full_mani_skill_example/train_pnt_bc_bucket10_19.sh
For drawer task: https://drive.google.com/file/d/13JHUN-HJTwRQBng-pKIFOc13nEUFKDeE/view?usp=sharing
For door task: https://drive.google.com/file/d/1KReHkxBKHF-m4rCmQT-bv_I3-jSHpDW8/view?usp=sharing
For chair task: https://drive.google.com/file/d/1UIFKqM54-yEMjIHQER7-PGA6clNiUEOs/view?usp=sharing
For bucket task: https://drive.google.com/file/d/1frUiuV8CHTA0BvED7bw4FTM_P9uspSrC/view?usp=sharing
After downloading the checkpoints, unzip and place them into the folder ./work_dirs/
.
We evaluate the models following ManiSkill Benchmark Repo.
Please follow the steps to run our solution:
user_solution.py
will run our solution. Please do not change its name.PYTHONPATH=YOUR_SOLUTION_DIRECTORY:$PYTHONPATH python mani_skill/tools/evaluate_policy.py --env ENV_NAME
YOUR_SOLUTION_DIRECTORY
is the directory containing your user_solution.py
--level-range 100-200
user_solution.py
before running the script./eval_results.csv
@inproceedings{wu2022minimalist,
title={A Minimalist Ensemble Method for Generalizable Offline Deep Reinforcement Learning},
author={Wu, Kun and Zhao, Yinuo and Xu, Zhiyuan and Zhao, Zhen and Ren, Pei and Che, Zhengping and Liu, Chi Harold and Feng, Feifei and Tang, Jian},
booktitle={ICLR 2022 Workshop on Generalizable Policy Learning in Physical World}
}
No Description
Python Shell Cuda C++ Text other
Dear OpenI User
Thank you for your continuous support to the Openl Qizhi Community AI Collaboration Platform. In order to protect your usage rights and ensure network security, we updated the Openl Qizhi Community AI Collaboration Platform Usage Agreement in January 2024. The updated agreement specifies that users are prohibited from using intranet penetration tools. After you click "Agree and continue", you can continue to use our services. Thank you for your cooperation and understanding.
For more agreement content, please refer to the《Openl Qizhi Community AI Collaboration Platform Usage Agreement》