├── .gitignore ├── LICENSE ├── README.md ├── environment.yml ├── readme_imgs ├── concept_figure.png └── poster.mp4 └── src ├── config └── __init__.py ├── door_lock_verifier.py ├── door_locker.py ├── env ├── __init__.py ├── hand_manipulation_suite │ ├── Adroit │ │ ├── .gitignore │ │ ├── Adroit_hand.xml │ │ ├── Adroit_hand_withOverlay.xml │ │ ├── LICENSE │ │ ├── README.md │ │ ├── gallery │ │ │ ├── news.JPG │ │ │ └── projects.JPG │ │ └── resources │ │ │ ├── assets.xml │ │ │ ├── chain.xml │ │ │ ├── chain1.xml │ │ │ ├── joint_position_actuation.xml │ │ │ ├── meshes │ │ │ ├── F1.stl │ │ │ ├── F2.stl │ │ │ ├── F3.stl │ │ │ ├── TH1_z.stl │ │ │ ├── TH2_z.stl │ │ │ ├── TH3_z.stl │ │ │ ├── arm_base.stl │ │ │ ├── arm_trunk.stl │ │ │ ├── arm_trunk_asmbly.stl │ │ │ ├── distal_ellipsoid.stl │ │ │ ├── elbow_flex.stl │ │ │ ├── elbow_rotate_motor.stl │ │ │ ├── elbow_rotate_muscle.stl │ │ │ ├── forearm_Cy_PlateAsmbly(muscle_cone).stl │ │ │ ├── forearm_Cy_PlateAsmbly.stl │ │ │ ├── forearm_PlateAsmbly.stl │ │ │ ├── forearm_electric.stl │ │ │ ├── forearm_electric_cvx.stl │ │ │ ├── forearm_muscle.stl │ │ │ ├── forearm_simple.stl │ │ │ ├── forearm_simple_cvx.stl │ │ │ ├── forearm_weight.stl │ │ │ ├── hand_base_link.STL │ │ │ ├── hook_0.STL │ │ │ ├── hook_1.STL │ │ │ ├── hook_2.STL │ │ │ ├── hook_3.STL │ │ │ ├── knuckle.stl │ │ │ ├── lfmetacarpal.stl │ │ │ ├── palm.stl │ │ │ ├── upper_arm.stl │ │ │ ├── upper_arm_asmbl_shoulder.stl │ │ │ ├── upper_arm_ass.stl │ │ │ └── wrist.stl │ │ │ └── tendon_torque_actuation.xml │ ├── __init__.py │ ├── assets │ │ ├── DAPG_Adroit.xml │ │ ├── DAPG_assets.xml │ │ ├── DAPG_assets_hook.xml │ │ ├── DAPG_door.xml │ │ ├── DAPG_hammer.xml │ │ ├── DAPG_pen.xml │ │ ├── DAPG_relocate.xml │ │ ├── door_xknob_adroit.xml │ │ ├── door_xknob_adroit_extrageom.xml │ │ ├── door_xknob_hook.xml │ │ └── tasks.jpg │ ├── door_lock_adroit_simple.py │ ├── door_lock_verify_adroit_simple.py │ └── door_v0.py └── robel │ └── dclaw_env.py └── robel_screw_pi_task.py /.gitignore: -------------------------------------------------------------------------------- 1 | # Byte-compiled / optimized / DLL files 2 | __pycache__/ 3 | *.py[cod] 4 | *$py.class 5 | 6 | # C extensions 7 | *.so 8 | 9 | # Distribution / packaging 10 | .Python 11 | build/ 12 | develop-eggs/ 13 | dist/ 14 | downloads/ 15 | eggs/ 16 | .eggs/ 17 | lib/ 18 | lib64/ 19 | parts/ 20 | sdist/ 21 | var/ 22 | wheels/ 23 | pip-wheel-metadata/ 24 | share/python-wheels/ 25 | *.egg-info/ 26 | .installed.cfg 27 | *.egg 28 | MANIFEST 29 | 30 | # PyInstaller 31 | # Usually these files are written by a python script from a template 32 | # before PyInstaller builds the exe, so as to inject date/other infos into it. 33 | *.manifest 34 | *.spec 35 | 36 | # Installer logs 37 | pip-log.txt 38 | pip-delete-this-directory.txt 39 | 40 | # Unit test / coverage reports 41 | htmlcov/ 42 | .tox/ 43 | .nox/ 44 | .coverage 45 | .coverage.* 46 | .cache 47 | nosetests.xml 48 | coverage.xml 49 | *.cover 50 | *.py,cover 51 | .hypothesis/ 52 | .pytest_cache/ 53 | 54 | # Translations 55 | *.mo 56 | *.pot 57 | 58 | # Django stuff: 59 | *.log 60 | local_settings.py 61 | db.sqlite3 62 | db.sqlite3-journal 63 | 64 | # Flask stuff: 65 | instance/ 66 | .webassets-cache 67 | 68 | # Scrapy stuff: 69 | .scrapy 70 | 71 | # Sphinx documentation 72 | docs/_build/ 73 | 74 | # PyBuilder 75 | target/ 76 | 77 | # Jupyter Notebook 78 | .ipynb_checkpoints 79 | 80 | # IPython 81 | profile_default/ 82 | ipython_config.py 83 | 84 | # pyenv 85 | .python-version 86 | 87 | # pipenv 88 | # According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control. 89 | # However, in case of collaboration, if having platform-specific dependencies or dependencies 90 | # having no cross-platform support, pipenv may install dependencies that don't work, or not 91 | # install all needed dependencies. 92 | #Pipfile.lock 93 | 94 | # PEP 582; used by e.g. github.com/David-OConnor/pyflow 95 | __pypackages__/ 96 | 97 | # Celery stuff 98 | celerybeat-schedule 99 | celerybeat.pid 100 | 101 | # SageMath parsed files 102 | *.sage.py 103 | 104 | # Environments 105 | .env 106 | .venv 107 | venv/ 108 | ENV/ 109 | env.bak/ 110 | venv.bak/ 111 | 112 | # Spyder project settings 113 | .spyderproject 114 | .spyproject 115 | 116 | # Rope project settings 117 | .ropeproject 118 | 119 | # mkdocs documentation 120 | /site 121 | 122 | # mypy 123 | .mypy_cache/ 124 | .dmypy.json 125 | dmypy.json 126 | 127 | # Pyre type checker 128 | .pyre/ 129 | 130 | .vscode/ 131 | demos/ 132 | videos/ 133 | .DS_Store 134 | draft.py 135 | checkpoints/ 136 | figures/ 137 | MUJOCO_LOG.TXT 138 | *.zip 139 | *.npy 140 | 141 | nohup.out 142 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # Training Robots to Evaluate Robots: Example-Based Interactive Reward Functions for Policy Learning 2 | 3 | CoRL 2022 (Oral)\ 4 | [Kun Huang](https://www.linkedin.com/in/kun-huang-620034171/), [Edward S. Hu](https://edwardshu.com/), [Dinesh Jayaraman](https://www.seas.upenn.edu/~dineshj/) 5 | #### [[Paper (Openreview)]](https://openreview.net/forum?id=sK2aWU7X9b8) [[Project Website]](https://sites.google.com/view/lirf-corl-2022/) [[Poster]](https://github.com/penn-pal-lab/interactive_reward_functions/blob/main/readme_imgs/poster.mp4) 6 | 7 | 8 |

9 | 10 |

11 | 12 | 13 | Physical interactions can often help reveal information that is not readily apparent. For example, we may tug at a table leg to evaluate whether it is built well, or turn a water bottle upside down to check that it is watertight. We propose to train robots to acquire such interactive behaviors automatically, for the purpose of evaluating the result of an attempted robotic skill execution. These evaluations in turn serve as "interactive reward functions" (IRFs) for training reinforcement learning policies to perform the target skill, such as screwing the table leg tightly. In addition, even after task policies are fully trained, IRFs can serve as verification mechanisms that improve online task execution. For any given task, our IRFs can be conveniently trained using only examples of successful outcomes, and no further specification is needed to train the task policy thereafter. In our evaluations on door locking and weighted block stacking in simulation, and screw tightening on a real robot, IRFs enable large performance improvements, even outperforming baselines with access to demonstrations or carefully engineered rewards. 14 | 15 | If you find this work useful in your research, please cite: 16 | ``` 17 | @inproceedings{ 18 | huang2022training, 19 | title={Training Robots to Evaluate Robots: Example-Based Interactive Reward Functions for Policy Learning}, 20 | author={Kun Huang and Edward S. Hu and Dinesh Jayaraman}, 21 | booktitle={6th Annual Conference on Robot Learning}, 22 | year={2022}, 23 | url={https://openreview.net/forum?id=sK2aWU7X9b8} 24 | } 25 | ``` 26 | 27 | 28 | 29 | ## Install Python Environment 30 | 31 | ### Roboaware 32 | 33 | 1. Run `conda env create -f environment.yml`, then activate this conda environment. 34 | 2. Clone the [`d4rl`](https://github.com/voyager1998/d4rl.git) 35 | 3. cd into the d4rl repo, run `pip install -e .` 36 | 37 | ### ROBEL 38 | 39 | WARNING: Do not install robel package inside `Roboaware` conda env, it's incompatible. 40 | 41 | 1. Clone [ROBEL](https://github.com/voyager1998/robel.git) repo into another folder. 42 | 2. Run `pip install -e .` inside robel. 43 | 3. For the rest, follow exactly the [instructions](https://github.com/google-research/robel). May need to download MuJoCo 2.0 if not installed, but it's fine to have multiple versions of MoJoCo. 44 | 4. Use Dynamixel Wizard to figure out the port etc of the motors. 45 | 5. After modifying ROBEL code, reinstall robel by running 46 | 47 | ```bash 48 | pip uninstall robel 49 | pip install -e . 50 | ``` 51 | 52 | ## Trouble Shooting 53 | 54 | If the error `Failed to initialize OpenGL` appears, refer to [link](https://github.com/openai/mujoco-py/issues/187), and try `unset LD_PRELOAD`. 55 | 56 | If the error `GLEW initialization error: Missing GL version` appears, refer to [link](https://github.com/openai/mujoco-py/issues/408), and try 57 | 58 | ```bash 59 | export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libGLEW.so 60 | ``` 61 | -------------------------------------------------------------------------------- /environment.yml: -------------------------------------------------------------------------------- 1 | name: roboaware 2 | channels: 3 | - conda-forge 4 | - anaconda 5 | - defaults 6 | dependencies: 7 | - _libgcc_mutex=0.1 8 | - _openmp_mutex=4.5 9 | - _pytorch_select=0.1 10 | - _tflow_select=2.3.0 11 | - absl-py=0.13.0 12 | - aiohttp=3.8.1 13 | - aiosignal=1.2.0 14 | - astor=0.8.1 15 | - async-timeout=4.0.1 16 | - asynctest=0.13.0 17 | - attrs=21.2.0 18 | - autopep8=1.5.5 19 | - blas=1.0 20 | - blinker=1.4 21 | - blosc=1.21.0 22 | - bottleneck=1.3.2 23 | - brotli=1.0.9 24 | - brotlipy=0.7.0 25 | - brunsli=0.1 26 | - bzip2=1.0.8 27 | - c-ares=1.17.1 28 | - ca-certificates=2020.10.14 29 | - cachetools=4.2.2 30 | - cairo=1.16.0 31 | - catkin_pkg=0.4.23 32 | - certifi=2020.6.20 33 | - cfitsio=3.470 34 | - charls=2.2.0 35 | - charset-normalizer=2.0.4 36 | - click=8.0.3 37 | - colorlog=4.4.0 38 | - cryptography=3.4.8 39 | - cython=0.29.21 40 | - cytoolz=0.11.0 41 | - dask-core=2021.10.0 42 | - dataclasses=0.8 43 | - distro=1.5.0 44 | - docutils=0.18.1 45 | - ffmpeg=4.0 46 | - fontconfig=2.13.1 47 | - fonttools=4.25.0 48 | - freeglut=3.0.0 49 | - freetype=2.11.0 50 | - frozenlist=1.2.0 51 | - fsspec=2021.10.1 52 | - future=0.18.2 53 | - gast=0.2.2 54 | - giflib=5.2.1 55 | - glib=2.69.1 56 | - google-auth=1.33.0 57 | - google-auth-oauthlib=0.4.4 58 | - google-pasta=0.2.0 59 | - graphite2=1.3.14 60 | - grpcio=1.42.0 61 | - h5py=2.8.0 62 | - harfbuzz=1.8.8 63 | - hdf5=1.10.2 64 | - icu=58.2 65 | - idna=3.3 66 | - imagecodecs=2021.8.26 67 | - imageio=2.9.0 68 | - intel-openmp=2019.4 69 | - jasper=2.0.14 70 | - joblib=1.1.0 71 | - jpeg=9d 72 | - jxrlib=1.1 73 | - keras-applications=1.0.8 74 | - keras-preprocessing=1.1.2 75 | - krb5=1.19.2 76 | - lcms2=2.12 77 | - ld_impl_linux-64=2.35.1 78 | - lerc=3.0 79 | - libaec=1.0.4 80 | - libcurl=7.78.0 81 | - libdeflate=1.8 82 | - libedit=3.1.20210910 83 | - libev=4.33 84 | - libffi=3.3 85 | - libgcc-ng=9.3.0 86 | - libgfortran-ng=7.5.0 87 | - libgfortran4=7.5.0 88 | - libglu=9.0.0 89 | - libgomp=9.3.0 90 | - libmklml=2019.0.5 91 | - libnghttp2=1.46.0 92 | - libopencv=3.4.2 93 | - libopus=1.3.1 94 | - libpng=1.6.37 95 | - libprotobuf=3.17.2 96 | - libssh2=1.9.0 97 | - libstdcxx-ng=9.3.0 98 | - libtiff=4.2.0 99 | - libuuid=1.0.3 100 | - libvpx=1.7.0 101 | - libwebp=1.2.0 102 | - libwebp-base=1.2.0 103 | - libxcb=1.14 104 | - libxml2=2.9.12 105 | - libzopfli=1.0.3 106 | - locket=0.2.1 107 | - lz4-c=1.9.3 108 | - markdown=3.3.4 109 | - matplotlib-base=3.5.0 110 | - mkl=2020.2 111 | - mkl-service=2.3.0 112 | - mkl_fft=1.3.0 113 | - mkl_random=1.1.1 114 | - multidict=5.1.0 115 | - munkres=1.1.4 116 | - ncurses=6.3 117 | - networkx=2.6.3 118 | - ninja=1.10.2 119 | - numexpr=2.7.3 120 | - numpy-base=1.19.2 121 | - oauthlib=3.1.1 122 | - olefile=0.46 123 | - opencv=3.4.2 124 | - openjpeg=2.4.0 125 | - openssl=1.1.1l 126 | - opt_einsum=3.3.0 127 | - packaging=21.3 128 | - pandas=1.3.4 129 | - partd=1.2.0 130 | - pcre=8.45 131 | - pip=21.2.2 132 | - pixman=0.40.0 133 | - py-opencv=3.4.2 134 | - pyasn1=0.4.8 135 | - pyasn1-modules=0.2.8 136 | - pycodestyle=2.8.0 137 | - pycparser=2.21 138 | - pyjwt=2.1.0 139 | - pyopenssl=21.0.0 140 | - pyparsing=3.0.4 141 | - pysocks=1.7.1 142 | - python=3.7.9 143 | - python-dateutil=2.8.2 144 | - python_abi=3.7 145 | - pytorch=1.7.1 146 | - pytz=2021.3 147 | - pywavelets=1.1.1 148 | - quaternion=2020.9.5.14.42.2 149 | - readline=8.1 150 | - requests-oauthlib=1.3.0 151 | - rospkg=1.2.10 152 | - rsa=4.7.2 153 | - scikit-image=0.17.2 154 | - scikit-learn=1.0.1 155 | - scipy=1.6.1 156 | - setuptools=58.0.4 157 | - six=1.16.0 158 | - snappy=1.1.8 159 | - sqlite=3.36.0 160 | - stable-baselines3=1.1.0 161 | - tensorboard=2.4.0 162 | - tensorboard-plugin-wit=1.6.0 163 | - tensorflow=2.1.0 164 | - tensorflow-base=2.1.0 165 | - tensorflow-estimator=2.6.0 166 | - termcolor=1.1.0 167 | - threadpoolctl=2.2.0 168 | - tifffile=2021.7.2 169 | - tk=8.6.11 170 | - toml=0.10.2 171 | - toolz=0.11.2 172 | - torchvision=0.8.2 173 | - typing-extensions=3.10.0.2 174 | - typing_extensions=3.10.0.2 175 | - werkzeug=2.0.2 176 | - wheel=0.37.0 177 | - wrapt=1.13.3 178 | - xorg-fixesproto=5.0 179 | - xorg-kbproto=1.0.7 180 | - xorg-libx11=1.7.2 181 | - xorg-libxcursor=1.2.0 182 | - xorg-libxext=1.3.4 183 | - xorg-libxfixes=5.0.3 184 | - xorg-libxinerama=1.1.4 185 | - xorg-libxrandr=1.5.2 186 | - xorg-libxrender=0.9.10 187 | - xorg-randrproto=1.5.0 188 | - xorg-renderproto=0.11.1 189 | - xorg-xextproto=7.3.0 190 | - xorg-xproto=7.0.31 191 | - xz=5.2.5 192 | - yaml=0.2.5 193 | - yapf=0.31.0 194 | - yarl=1.6.3 195 | - zfp=0.5.5 196 | - zlib=1.2.11 197 | - zstd=1.4.9 198 | - pip: 199 | - appnope==0.1.0 200 | - backcall==0.2.0 201 | - cffi==1.14.2 202 | - cloudpickle==1.3.0 203 | - configparser==5.0.1 204 | - cycler==0.10.0 205 | - docker-pycreds==0.4.0 206 | - fasteners==0.15 207 | - flake8==3.8.3 208 | - gitdb==4.0.5 209 | - gitpython==3.1.11 210 | - glfw==2.3.0 211 | - gym==0.17.2 212 | - importlib-metadata==2.0.0 213 | - ipdb==0.13.3 214 | - ipython==7.18.1 215 | - ipython-genutils==0.2.0 216 | - jedi==0.17.2 217 | - kiwisolver==1.2.0 218 | - matplotlib==3.3.1 219 | - mccabe==0.6.1 220 | - monotonic==1.5 221 | - mujoco-py<2.2,>=2.1 222 | - numpy==1.19.1 223 | - opencv-contrib-python==4.5.1.48 224 | - parso==0.7.1 225 | - pathtools==0.1.2 226 | - pexpect==4.8.0 227 | - pickleshare==0.7.5 228 | - pillow==7.2.0 229 | - promise==2.3 230 | - prompt-toolkit==3.0.7 231 | - protobuf==3.13.0 232 | - psutil==5.7.3 233 | - ptyprocess==0.6.0 234 | - pupil-apriltags==1.0.4 235 | - pyflakes==2.2.0 236 | - pyglet==1.5.0 237 | - pygments==2.6.1 238 | - pyquaternion==0.9.9 239 | - pyyaml==5.3.1 240 | - requests==2.24.0 241 | - sentry-sdk==0.19.2 242 | - shortuuid==1.0.1 243 | - smmap==3.0.4 244 | - subprocess32==3.5.4 245 | - torch==1.6.0 246 | - tqdm==4.48.2 247 | - traitlets==4.3.3 248 | - urllib3 249 | - wandb==0.10.10 250 | - watchdog==0.10.3 251 | - wcwidth==0.2.5 252 | - zipp==3.4.0 253 | -------------------------------------------------------------------------------- /readme_imgs/concept_figure.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/readme_imgs/concept_figure.png -------------------------------------------------------------------------------- /readme_imgs/poster.mp4: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/readme_imgs/poster.mp4 -------------------------------------------------------------------------------- /src/config/__init__.py: -------------------------------------------------------------------------------- 1 | import argparse 2 | from argparse import ArgumentParser 3 | 4 | 5 | def str2bool(v): 6 | return v.lower() == "true" 7 | 8 | 9 | def str2intlist(value): 10 | if not value: 11 | return value 12 | else: 13 | return [int(num) for num in value.split(",")] 14 | 15 | 16 | def str2list(value): 17 | if not value: 18 | return value 19 | else: 20 | return [num for num in value.split(",")] 21 | 22 | 23 | def create_parser(): 24 | """ 25 | Creates the argparser. Use this to add additional arguments 26 | to the parser later. 27 | """ 28 | parser = argparse.ArgumentParser( 29 | "Robot Aware Cost", 30 | formatter_class=argparse.ArgumentDefaultsHelpFormatter, 31 | ) 32 | parser.add_argument("--jobname", type=str, default=None) 33 | parser.add_argument("--log_dir", type=str, default="logs") 34 | parser.add_argument("--wandb", type=str2bool, default=False) 35 | parser.add_argument("--wandb_entity", type=str, default="pal") 36 | parser.add_argument("--wandb_project", type=str, default="roboaware") 37 | parser.add_argument("--wandb_group", type=str, default=None) 38 | parser.add_argument("--wandb_job_type", type=str, default=None) 39 | 40 | parser.add_argument("--visualize_mode", type=str2bool, default=False) 41 | 42 | add_method_arguments(parser) 43 | add_ensemble_arguments(parser) 44 | 45 | return parser 46 | 47 | 48 | def add_method_arguments(parser: ArgumentParser): 49 | # method arguments 50 | parser.add_argument( 51 | "--reward_type", 52 | type=str, 53 | default="gt", 54 | choices=[ 55 | "gt", "success_classifier", "weighted", "dense", "inpaint", 56 | "sparse" 57 | "blackrobot", "inpaint-blur", "eef_inpaint", "dontcare" 58 | ], 59 | ) 60 | # for use with inpaint blur 61 | parser.add_argument("--most_recent_background", 62 | type=str2bool, 63 | default=False) 64 | # inpaint-blur 65 | parser.add_argument("--blur_sigma", type=float, default=10) 66 | parser.add_argument("--unblur_cost_scale", type=float, default=3) 67 | # switch at step L - unblur_timestep 68 | parser.add_argument("--unblur_timestep", type=float, default=1) 69 | 70 | # control algorithm 71 | parser.add_argument( 72 | "--mbrl_algo", 73 | type=str, 74 | default="cem", 75 | choices=["cem"], 76 | ) 77 | 78 | # training 79 | parser.add_argument("--gpu", type=int, default=None) 80 | parser.add_argument("--seed", type=int, default=0) 81 | parser.add_argument("--num_episodes", type=int, default=100) 82 | parser.add_argument("--record_trajectory", type=str2bool, default=False) 83 | parser.add_argument("--record_trajectory_interval", type=int, default=5) 84 | parser.add_argument("--record_video_interval", type=int, default=1) 85 | 86 | # environment 87 | parser.add_argument("--env", type=str, default="FetchPush") 88 | args, unparsed = parser.parse_known_args() 89 | 90 | add_prediction_arguments(parser) 91 | add_dataset_arguments(parser) 92 | add_cost_arguments(parser) 93 | 94 | if args.mbrl_algo == "cem": 95 | add_cem_arguments(parser) 96 | 97 | # env specific args 98 | if args.env == "FetchPush": 99 | add_fetch_push_arguments(parser) 100 | 101 | return parser 102 | 103 | 104 | # Env Hyperparameters 105 | def add_fetch_push_arguments(parser: ArgumentParser): 106 | # override prediction dimension stuff 107 | parser.set_defaults(robot_dim=6, robot_enc_dim=6) 108 | parser.add_argument("--img_dim", type=int, default=64) 109 | parser.add_argument( 110 | "--camera_name", 111 | type=str, 112 | default="external_camera_0", 113 | choices=[ 114 | "head_camera_rgb", "gripper_camera_rgb", "lidar", 115 | "external_camera_0" 116 | ], 117 | ) 118 | parser.add_argument("--multiview", type=str2bool, default=False) 119 | parser.add_argument("--camera_ids", type=str2intlist, default=[0, 4]) 120 | parser.add_argument("--pixels_ob", type=str2bool, default=True) 121 | parser.add_argument("--norobot_pixels_ob", type=str2bool, default=False) 122 | parser.add_argument("--robot_mask_with_obj", type=str2bool, default=False) 123 | parser.add_argument("--inpaint_eef", type=str2bool, default=True) 124 | parser.add_argument("--depth_ob", type=str2bool, default=False) 125 | parser.add_argument("--object_dist_threshold", type=float, default=0.01) 126 | parser.add_argument("--gripper_dist_threshold", type=float, default=0.025) 127 | parser.add_argument("--push_dist", type=float, default=0.2) 128 | parser.add_argument("--max_episode_length", type=int, default=10) 129 | parser.add_argument( 130 | "--robot_goal_distribution", 131 | type=str, 132 | default="random", 133 | choices=["random", "behind_block"], 134 | ) 135 | parser.add_argument("--large_block", type=str2bool, default=False) 136 | parser.add_argument("--red_robot", type=str2bool, default=False) 137 | parser.add_argument("--invisible_demo", type=str2bool, default=False) 138 | parser.add_argument("--demo_dir", type=str, default="demos/fetch_push") 139 | 140 | 141 | def add_prediction_arguments(parser): 142 | parser.add_argument("--lr", 143 | default=0.0003, 144 | type=float, 145 | help="learning rate") 146 | parser.add_argument("--beta1", 147 | default=0.9, 148 | type=float, 149 | help="momentum term for adam") 150 | parser.add_argument("--batch_size", 151 | default=100, 152 | type=int, 153 | help="batch size") 154 | parser.add_argument("--test_batch_size", 155 | default=16, 156 | type=int, 157 | help="test batch size") 158 | parser.add_argument("--optimizer", 159 | default="adam", 160 | help="optimizer to train with") 161 | parser.add_argument("--niter", 162 | type=int, 163 | default=300, 164 | help="number of epochs to train for") 165 | parser.add_argument("--epoch_size", 166 | type=int, 167 | default=600, 168 | help="epoch size") 169 | parser.add_argument("--channels", default=3, type=int) 170 | parser.add_argument("--dataset", 171 | default="smmnist", 172 | help="dataset to train with") 173 | parser.add_argument("--n_past", 174 | type=int, 175 | default=1, 176 | help="number of frames to condition on") 177 | parser.add_argument( 178 | "--n_future", 179 | type=int, 180 | default=9, 181 | help="number of frames to predict during training", 182 | ) 183 | parser.add_argument("--n_eval", 184 | type=int, 185 | default=10, 186 | help="number of frames to predict during eval") 187 | parser.add_argument("--checkpoint_interval", type=int, default=5) 188 | parser.add_argument("--eval_interval", type=int, default=5) 189 | parser.add_argument("--rnn_size", 190 | type=int, 191 | default=256, 192 | help="dimensionality of hidden layer") 193 | parser.add_argument("--prior_rnn_layers", 194 | type=int, 195 | default=2, 196 | help="number of layers") 197 | parser.add_argument("--posterior_rnn_layers", 198 | type=int, 199 | default=2, 200 | help="number of layers") 201 | parser.add_argument("--predictor_rnn_layers", 202 | type=int, 203 | default=2, 204 | help="number of layers") 205 | parser.add_argument("--z_dim", 206 | type=int, 207 | default=10, 208 | help="dimensionality of z_t") 209 | parser.add_argument( 210 | "--g_dim", 211 | type=int, 212 | default=128, 213 | help="dimensionality of encoder output vector and decoder input vector", 214 | ) 215 | parser.add_argument("--action_dim", type=int, default=2) 216 | parser.add_argument("--action_enc_dim", type=int, default=2) 217 | parser.add_argument("--robot_dim", type=int, default=6) 218 | parser.add_argument("--robot_enc_dim", type=int, default=6) 219 | parser.add_argument("--robot_joint_dim", type=int, default=7) 220 | 221 | parser.add_argument("--beta", 222 | type=float, 223 | default=0.0001, 224 | help="weighting on KL to prior") 225 | 226 | parser.add_argument( 227 | "--last_frame_skip", 228 | type=str2bool, 229 | default=False, 230 | help= 231 | "if true, skip connections go between frame t and frame t+t rather than last ground truth frame", 232 | ) 233 | 234 | parser.add_argument("--model", 235 | default="svg", 236 | choices=["svg", "det", "copy"]) 237 | parser.add_argument("--model_use_mask", type=str2bool, default=True) 238 | parser.add_argument("--model_use_robot_state", type=str2bool, default=True) 239 | parser.add_argument("--reconstruction_loss", 240 | default="mse", 241 | choices=["mse", "l1", "dontcare_mse"]) 242 | parser.add_argument("--scheduled_sampling", type=str2bool, default=False) 243 | parser.add_argument("--robot_pixel_weight", 244 | type=float, 245 | default=0, 246 | help="weighting on robot pixels") 247 | 248 | parser.add_argument("--learned_robot_model", type=str2bool, default=False) 249 | parser.add_argument("--robot_model_ckpt", type=str, default=None) 250 | parser.add_argument("--use_xy_channel", type=str2bool, default=False) 251 | parser.add_argument("--load_checkpoint", type=str, default=None) 252 | 253 | 254 | def add_dataset_arguments(parser): 255 | parser.add_argument("--data_threads", 256 | type=int, 257 | default=5, 258 | help="number of data loading threads") 259 | parser.add_argument("--data_root", 260 | default="data", 261 | help="root directory for data") 262 | parser.add_argument("--train_val_split", type=float, default=0.8) 263 | # data collection policy arguments 264 | parser.add_argument("--temporal_beta", type=float, default=1) 265 | parser.add_argument("--demo_length", type=int, default=12) 266 | parser.add_argument("--action_noise", type=float, default=0) 267 | parser.add_argument( 268 | "--video_type", 269 | default="object_inpaint_demo", 270 | choices=["object_inpaint_demo", "robot_demo", "object_only_demo"], 271 | ) 272 | # robonet video prediction dataset arguments 273 | parser.add_argument( 274 | "--video_length", 275 | type=int, 276 | default=31, 277 | help="max length of the video, used for evaluation dataloader") 278 | parser.add_argument("--impute_autograsp_action", 279 | type=str2bool, 280 | default=True) 281 | parser.add_argument("--preload_ram", type=str2bool, default=False) 282 | parser.add_argument("--training_regime", 283 | type=str, 284 | choices=[ 285 | "multirobot", "singlerobot", "finetune", 286 | "train_sawyer_multiview", "finetune_sawyer_view", 287 | "finetune_widowx" 288 | ], 289 | default="multirobot") 290 | parser.add_argument( 291 | "--preprocess_action", 292 | type=str, 293 | choices=["raw", "camera_raw", "state_infer", "camera_state_infer"], 294 | default="raw") 295 | parser.add_argument("--img_augmentation", type=str2bool, default=False) 296 | parser.add_argument("--color_jitter_range", type=float, default=0.1) 297 | parser.add_argument("--random_crop_size", type=int, default=59) 298 | parser.add_argument("--dropout", type=float, default=None) 299 | parser.add_argument("--world_error_dict", type=str, default=None) 300 | parser.add_argument("--finetune_num_train", type=int, default=400) 301 | parser.add_argument("--finetune_num_test", type=int, default=100) 302 | parser.add_argument("--num_train", type=int, default=400) 303 | parser.add_argument("--num_test", type=int, default=100) 304 | 305 | 306 | # CEM Hyperparameters 307 | def add_cem_arguments(parser): 308 | parser.add_argument("--horizon", type=int, default=3) 309 | parser.add_argument("--opt_iter", type=int, default=10) 310 | parser.add_argument("--action_candidates", type=int, default=30) 311 | parser.add_argument("--topk", type=int, default=5) 312 | parser.add_argument("--replan_every", type=int, default=1) 313 | parser.add_argument("--dynamics_model_ckpt", type=str, default=None) 314 | parser.add_argument("--candidates_batch_size", type=int, default=200) 315 | parser.add_argument("--use_env_dynamics", type=str2bool, default=False) 316 | parser.add_argument("--debug_trajectory_path", type=str, default=None) 317 | parser.add_argument("--debug_cem", type=str2bool, default=False) 318 | parser.add_argument("--object_demo_dir", type=str, default=None) 319 | parser.add_argument("--subgoal_start", type=int, default=0) 320 | parser.add_argument("--sequential_subgoal", type=str2bool, default=True) 321 | parser.add_argument("--demo_cost", type=str2bool, default=False) 322 | parser.add_argument("--demo_timescale", type=int, default=1) 323 | parser.add_argument("--action_repeat", type=int, default=1) 324 | parser.add_argument( 325 | "--demo_type", 326 | default="object_only_demo", 327 | choices=["object_inpaint_demo", "object_only_demo", "robot_demo"], 328 | ) 329 | parser.add_argument("--cem_init_std", type=float, default=1) 330 | parser.add_argument("--sparse_cost", type=str2bool, default=False) 331 | 332 | 333 | # Cost Fn Hyperparameters 334 | 335 | 336 | def add_cost_arguments(parser): 337 | # cost thresholds for determining goal success 338 | parser.add_argument("--world_cost_success", type=float, default=4000) 339 | parser.add_argument("--robot_cost_success", type=float, default=0.01) 340 | # weight of the costs 341 | parser.add_argument("--robot_cost_weight", type=float, default=0) 342 | parser.add_argument("--world_cost_weight", type=float, default=1) 343 | # checks if pixel diff > threshold before counting it 344 | parser.add_argument("--img_cost_threshold", type=float, default=None) 345 | # only used by img don't care cost, divide by number of world pixels 346 | parser.add_argument("--img_cost_world_norm", type=str2bool, default=True) 347 | 348 | 349 | def add_ensemble_arguments(parser): 350 | parser.add_argument("--load_traj_length", type=int, default=10) 351 | parser.add_argument("--num_ensembles", type=int, default=10) 352 | parser.add_argument("--train_data_per_ensemble", type=int, default=500) 353 | 354 | 355 | def argparser(): 356 | """ Directly parses the arguments. """ 357 | parser = create_parser() 358 | args, unparsed = parser.parse_known_args() 359 | assert len(unparsed) == 0, unparsed 360 | return args, unparsed 361 | -------------------------------------------------------------------------------- /src/door_lock_verifier.py: -------------------------------------------------------------------------------- 1 | import os 2 | import time 3 | from stable_baselines3 import SAC 4 | 5 | from src.env.hand_manipulation_suite.door_v0 import DoorEnvV0 6 | from src.env.hand_manipulation_suite.door_lock_verify_adroit_simple import DoorLockVerifyEnv 7 | 8 | if __name__ == "__main__": 9 | from src.config import argparser 10 | 11 | config, _ = argparser() 12 | config.jobname = "door_lock_verifier" 13 | checkpoint_path = os.path.join("checkpoints", config.jobname) 14 | os.makedirs(checkpoint_path, exist_ok=True) 15 | 16 | env = DoorLockVerifyEnv() 17 | env.reset() 18 | 19 | model = SAC("MlpPolicy", env, verbose=1) 20 | 21 | # Training/Continue training 22 | total_timesteps = 2000000 23 | save_per_timesteps = 10000 24 | timestep = [] 25 | success_rates = [] 26 | rewards = [] 27 | continue_training = 2000000 28 | for i in range(int(total_timesteps / save_per_timesteps)): 29 | print("======================================") 30 | start = time.time() 31 | model.learn(total_timesteps=save_per_timesteps) 32 | 33 | # Save model 34 | model.save(checkpoint_path + "/mfrl_sc_" + 35 | str((i + 1) * save_per_timesteps + continue_training)) 36 | print("Model saved after", 37 | (i + 1) * save_per_timesteps + continue_training, "timesteps") 38 | -------------------------------------------------------------------------------- /src/door_locker.py: -------------------------------------------------------------------------------- 1 | import os 2 | import time 3 | from stable_baselines3 import SAC 4 | 5 | from src.env.hand_manipulation_suite.door_v0 import DoorEnvV0 6 | from src.env.hand_manipulation_suite.door_lock_adroit_simple import DoorLockEnvAdroitSimple 7 | 8 | if __name__ == "__main__": 9 | from src.config import argparser 10 | 11 | config, _ = argparser() 12 | config.jobname = "door_locker" 13 | checkpoint_path = os.path.join("checkpoints", config.jobname) 14 | os.makedirs(checkpoint_path, exist_ok=True) 15 | 16 | env = DoorLockEnvAdroitSimple() 17 | env.reset() 18 | 19 | model = SAC( 20 | "MlpPolicy", 21 | env, 22 | verbose=1, 23 | ) 24 | 25 | # Training/Continue training 26 | total_timesteps = 4000000 27 | save_per_timesteps = 10000 28 | timestep = [] 29 | success_rates = [] 30 | door_closed_rates = [] 31 | rewards = [] 32 | continue_training = 0 33 | 34 | for i in range(int(total_timesteps / save_per_timesteps)): 35 | print("======================================") 36 | start = time.time() 37 | model.learn(total_timesteps=save_per_timesteps) 38 | 39 | # Save model 40 | model.save(checkpoint_path + "/mfrl_sc_" + 41 | str((i + 1) * save_per_timesteps + continue_training)) 42 | print("Model saved after", 43 | (i + 1) * save_per_timesteps + continue_training, "timesteps") 44 | -------------------------------------------------------------------------------- /src/env/__init__.py: -------------------------------------------------------------------------------- 1 | # from src.env.robotics.fetch_env import FetchEnv 2 | # from src.env.robotics.fetch_push import FetchPushEnv 3 | def get_env(name): 4 | if name == "FetchPush": 5 | from src.env.robotics.fetch_push import FetchPushEnv 6 | 7 | return FetchPushEnv 8 | -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/.gitignore: -------------------------------------------------------------------------------- 1 | *.DS_Store 2 | -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/Adroit_hand.xml: -------------------------------------------------------------------------------- 1 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 | 39 | 40 | 41 | 42 | 43 | 44 | 45 | 46 | 47 | 48 | 49 | 50 | 51 | 52 | 53 | 54 | 55 | 56 | 57 | 58 | 59 | -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/Adroit_hand_withOverlay.xml: -------------------------------------------------------------------------------- 1 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 | 39 | 40 | 41 | 42 | 43 | 44 | 45 | 46 | 47 | 48 | 49 | 50 | 51 | 52 | 53 | 54 | 55 | 56 | 57 | 58 | -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/LICENSE: -------------------------------------------------------------------------------- 1 | Apache License 2 | Version 2.0, January 2004 3 | http://www.apache.org/licenses/ 4 | 5 | TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION 6 | 7 | 1. Definitions. 8 | 9 | "License" shall mean the terms and conditions for use, reproduction, 10 | and distribution as defined by Sections 1 through 9 of this document. 11 | 12 | "Licensor" shall mean the copyright owner or entity authorized by 13 | the copyright owner that is granting the License. 14 | 15 | "Legal Entity" shall mean the union of the acting entity and all 16 | other entities that control, are controlled by, or are under common 17 | control with that entity. For the purposes of this definition, 18 | "control" means (i) the power, direct or indirect, to cause the 19 | direction or management of such entity, whether by contract or 20 | otherwise, or (ii) ownership of fifty percent (50%) or more of the 21 | outstanding shares, or (iii) beneficial ownership of such entity. 22 | 23 | "You" (or "Your") shall mean an individual or Legal Entity 24 | exercising permissions granted by this License. 25 | 26 | "Source" form shall mean the preferred form for making modifications, 27 | including but not limited to software source code, documentation 28 | source, and configuration files. 29 | 30 | "Object" form shall mean any form resulting from mechanical 31 | transformation or translation of a Source form, including but 32 | not limited to compiled object code, generated documentation, 33 | and conversions to other media types. 34 | 35 | "Work" shall mean the work of authorship, whether in Source or 36 | Object form, made available under the License, as indicated by a 37 | copyright notice that is included in or attached to the work 38 | (an example is provided in the Appendix below). 39 | 40 | "Derivative Works" shall mean any work, whether in Source or Object 41 | form, that is based on (or derived from) the Work and for which the 42 | editorial revisions, annotations, elaborations, or other modifications 43 | represent, as a whole, an original work of authorship. For the purposes 44 | of this License, Derivative Works shall not include works that remain 45 | separable from, or merely link (or bind by name) to the interfaces of, 46 | the Work and Derivative Works thereof. 47 | 48 | "Contribution" shall mean any work of authorship, including 49 | the original version of the Work and any modifications or additions 50 | to that Work or Derivative Works thereof, that is intentionally 51 | submitted to Licensor for inclusion in the Work by the copyright owner 52 | or by an individual or Legal Entity authorized to submit on behalf of 53 | the copyright owner. For the purposes of this definition, "submitted" 54 | means any form of electronic, verbal, or written communication sent 55 | to the Licensor or its representatives, including but not limited to 56 | communication on electronic mailing lists, source code control systems, 57 | and issue tracking systems that are managed by, or on behalf of, the 58 | Licensor for the purpose of discussing and improving the Work, but 59 | excluding communication that is conspicuously marked or otherwise 60 | designated in writing by the copyright owner as "Not a Contribution." 61 | 62 | "Contributor" shall mean Licensor and any individual or Legal Entity 63 | on behalf of whom a Contribution has been received by Licensor and 64 | subsequently incorporated within the Work. 65 | 66 | 2. Grant of Copyright License. Subject to the terms and conditions of 67 | this License, each Contributor hereby grants to You a perpetual, 68 | worldwide, non-exclusive, no-charge, royalty-free, irrevocable 69 | copyright license to reproduce, prepare Derivative Works of, 70 | publicly display, publicly perform, sublicense, and distribute the 71 | Work and such Derivative Works in Source or Object form. 72 | 73 | 3. Grant of Patent License. Subject to the terms and conditions of 74 | this License, each Contributor hereby grants to You a perpetual, 75 | worldwide, non-exclusive, no-charge, royalty-free, irrevocable 76 | (except as stated in this section) patent license to make, have made, 77 | use, offer to sell, sell, import, and otherwise transfer the Work, 78 | where such license applies only to those patent claims licensable 79 | by such Contributor that are necessarily infringed by their 80 | Contribution(s) alone or by combination of their Contribution(s) 81 | with the Work to which such Contribution(s) was submitted. If You 82 | institute patent litigation against any entity (including a 83 | cross-claim or counterclaim in a lawsuit) alleging that the Work 84 | or a Contribution incorporated within the Work constitutes direct 85 | or contributory patent infringement, then any patent licenses 86 | granted to You under this License for that Work shall terminate 87 | as of the date such litigation is filed. 88 | 89 | 4. Redistribution. You may reproduce and distribute copies of the 90 | Work or Derivative Works thereof in any medium, with or without 91 | modifications, and in Source or Object form, provided that You 92 | meet the following conditions: 93 | 94 | (a) You must give any other recipients of the Work or 95 | Derivative Works a copy of this License; and 96 | 97 | (b) You must cause any modified files to carry prominent notices 98 | stating that You changed the files; and 99 | 100 | (c) You must retain, in the Source form of any Derivative Works 101 | that You distribute, all copyright, patent, trademark, and 102 | attribution notices from the Source form of the Work, 103 | excluding those notices that do not pertain to any part of 104 | the Derivative Works; and 105 | 106 | (d) If the Work includes a "NOTICE" text file as part of its 107 | distribution, then any Derivative Works that You distribute must 108 | include a readable copy of the attribution notices contained 109 | within such NOTICE file, excluding those notices that do not 110 | pertain to any part of the Derivative Works, in at least one 111 | of the following places: within a NOTICE text file distributed 112 | as part of the Derivative Works; within the Source form or 113 | documentation, if provided along with the Derivative Works; or, 114 | within a display generated by the Derivative Works, if and 115 | wherever such third-party notices normally appear. The contents 116 | of the NOTICE file are for informational purposes only and 117 | do not modify the License. You may add Your own attribution 118 | notices within Derivative Works that You distribute, alongside 119 | or as an addendum to the NOTICE text from the Work, provided 120 | that such additional attribution notices cannot be construed 121 | as modifying the License. 122 | 123 | You may add Your own copyright statement to Your modifications and 124 | may provide additional or different license terms and conditions 125 | for use, reproduction, or distribution of Your modifications, or 126 | for any such Derivative Works as a whole, provided Your use, 127 | reproduction, and distribution of the Work otherwise complies with 128 | the conditions stated in this License. 129 | 130 | 5. Submission of Contributions. Unless You explicitly state otherwise, 131 | any Contribution intentionally submitted for inclusion in the Work 132 | by You to the Licensor shall be under the terms and conditions of 133 | this License, without any additional terms or conditions. 134 | Notwithstanding the above, nothing herein shall supersede or modify 135 | the terms of any separate license agreement you may have executed 136 | with Licensor regarding such Contributions. 137 | 138 | 6. Trademarks. This License does not grant permission to use the trade 139 | names, trademarks, service marks, or product names of the Licensor, 140 | except as required for reasonable and customary use in describing the 141 | origin of the Work and reproducing the content of the NOTICE file. 142 | 143 | 7. Disclaimer of Warranty. Unless required by applicable law or 144 | agreed to in writing, Licensor provides the Work (and each 145 | Contributor provides its Contributions) on an "AS IS" BASIS, 146 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 147 | implied, including, without limitation, any warranties or conditions 148 | of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A 149 | PARTICULAR PURPOSE. You are solely responsible for determining the 150 | appropriateness of using or redistributing the Work and assume any 151 | risks associated with Your exercise of permissions under this License. 152 | 153 | 8. Limitation of Liability. In no event and under no legal theory, 154 | whether in tort (including negligence), contract, or otherwise, 155 | unless required by applicable law (such as deliberate and grossly 156 | negligent acts) or agreed to in writing, shall any Contributor be 157 | liable to You for damages, including any direct, indirect, special, 158 | incidental, or consequential damages of any character arising as a 159 | result of this License or out of the use or inability to use the 160 | Work (including but not limited to damages for loss of goodwill, 161 | work stoppage, computer failure or malfunction, or any and all 162 | other commercial damages or losses), even if such Contributor 163 | has been advised of the possibility of such damages. 164 | 165 | 9. Accepting Warranty or Additional Liability. While redistributing 166 | the Work or Derivative Works thereof, You may choose to offer, 167 | and charge a fee for, acceptance of support, warranty, indemnity, 168 | or other liability obligations and/or rights consistent with this 169 | License. However, in accepting such obligations, You may act only 170 | on Your own behalf and on Your sole responsibility, not on behalf 171 | of any other Contributor, and only if You agree to indemnify, 172 | defend, and hold each Contributor harmless for any liability 173 | incurred by, or claims asserted against, such Contributor by reason 174 | of your accepting any such warranty or additional liability. 175 | 176 | END OF TERMS AND CONDITIONS 177 | 178 | APPENDIX: How to apply the Apache License to your work. 179 | 180 | To apply the Apache License to your work, attach the following 181 | boilerplate notice, with the fields enclosed by brackets "[]" 182 | replaced with your own identifying information. (Don't include 183 | the brackets!) The text should be enclosed in the appropriate 184 | comment syntax for the file format. We also recommend that a 185 | file or class name and description of purpose be included on the 186 | same "printed page" as the copyright notice for easier 187 | identification within third-party archives. 188 | 189 | Copyright [yyyy] [name of copyright owner] 190 | 191 | Licensed under the Apache License, Version 2.0 (the "License"); 192 | you may not use this file except in compliance with the License. 193 | You may obtain a copy of the License at 194 | 195 | http://www.apache.org/licenses/LICENSE-2.0 196 | 197 | Unless required by applicable law or agreed to in writing, software 198 | distributed under the License is distributed on an "AS IS" BASIS, 199 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 200 | See the License for the specific language governing permissions and 201 | limitations under the License. 202 | -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/README.md: -------------------------------------------------------------------------------- 1 | # Adroit Manipulation Platform 2 | 3 | Adroit manipulation platform is reconfigurable, tendon-driven, pneumatically-actuated platform designed and developed by [Vikash Kumar](https://vikashplus.github.io/) during this Ph.D. ([Thesis: Manipulators and Manipulation in high dimensional spaces](https://digital.lib.washington.edu/researchworks/handle/1773/38104)) to study dynamic dexterous manipulation. Adroit is comprised of the [Shadow Hand](https://www.shadowrobot.com/products/dexterous-hand/) skeleton (developed by [Shadow Robot company](https://www.shadowrobot.com/)) and a custom arm, and is powered by a custom actuation sysem. This custom actuation system allows Adroit to move the ShadowHand skeleton faster than a human hand (70 msec limit-to-limit movement, 30 msec overall reflex latency), generate sufficient forces (40 N at each finger tendon, 125N at each wrist tendon), and achieve high compliance on the mechanism level (6 grams of external force at the fingertip displaces the finger when the system is powered.) This combination of speed, force, and compliance is a prerequisite for dexterous manipulation, yet it has never before been achieved with a tendon-driven system, let alone a system with 24 degrees of freedom and 40 tendons. 4 | 5 | ## Mujoco Model 6 | Adroit is a 28 degree of freedom system which consists of a 24 degrees of freedom **ShadowHand** and a 4 degree of freedom arm. This repository contains the Mujoco Models of the system developed with extreme care and great attention to the details. 7 | 8 | 9 | ## In Projects 10 | Adroit has been used in a wide variety of project. A small list is appended below. Details of these projects can be found [here](https://vikashplus.github.io/). 11 | [![projects](https://github.com/vikashplus/Adroit/blob/master/gallery/projects.JPG)](https://vikashplus.github.io/) 12 | ## In News and Media 13 | Adroit has found quite some attention in the world media. Details can be found [here](https://vikashplus.github.io/news.html) 14 | 15 | [![News](https://github.com/vikashplus/Adroit/blob/master/gallery/news.JPG)](https://vikashplus.github.io/news.html) 16 | 17 | 18 | ## Citation 19 | If the contents of this repo helped you, please consider citing 20 | 21 | ``` 22 | @phdthesis{Kumar2016thesis, 23 | title = {Manipulators and Manipulation in high dimensional spaces}, 24 | school = {University of Washington, Seattle}, 25 | author = {Kumar, Vikash}, 26 | year = {2016}, 27 | url = {https://digital.lib.washington.edu/researchworks/handle/1773/38104} 28 | } 29 | ``` 30 | -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/gallery/news.JPG: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/gallery/news.JPG -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/gallery/projects.JPG: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/gallery/projects.JPG -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/assets.xml: -------------------------------------------------------------------------------- 1 | 18 | 19 | 20 | 21 | 22 | 30 | 31 | 36 | 37 | 38 | 39 | 40 | 41 | 42 | 43 | 44 | 45 | 46 | 47 | 48 | 49 | 50 | 51 | 52 | 53 | 54 | 55 | 56 | 57 | 58 | 59 | 61 | 62 | 63 | 64 | 65 | 66 | 67 | 68 | 69 | 70 | 71 | 72 | 73 | 74 | 75 | 76 | 77 | 78 | 79 | 80 | 81 | 82 | 83 | 84 | 85 | 86 | 87 | 88 | 89 | 90 | 91 | 92 | 93 | 94 | 95 | 96 | 97 | 98 | 99 | 100 | 101 | 102 | 103 | 104 | 105 | 106 | 107 | 108 | 109 | 110 | 111 | 112 | 113 | 114 | 115 | 116 | 117 | 118 | 119 | 120 | 121 | 122 | 123 | 124 | 125 | 126 | 127 | 128 | 129 | 130 | 131 | 132 | 133 | 134 | 135 | 136 | 144 | 145 | 146 | 147 | 148 | 149 | 150 | 151 | 152 | 153 | 154 | 155 | 156 | 157 | 158 | 159 | 160 | 161 | 162 | 163 | 164 | 165 | 166 | 167 | 168 | 169 | 170 | 171 | 172 | 173 | 174 | 175 | 176 | 177 | 178 | 179 | 180 | 181 | 182 | 183 | 184 | 185 | 186 | 187 | 188 | 189 | 190 | 191 | 192 | 193 | 194 | 195 | 196 | 197 | 198 | 199 | 200 | 201 | 202 | 203 | 204 | 205 | 206 | 207 | 208 | 209 | 210 | 211 | 212 | 213 | 214 | 215 | 216 | 217 | 218 | 219 | 220 | 221 | 222 | 223 | 224 | 225 | 226 | 227 | 228 | 229 | 230 | 231 | 232 | 233 | 234 | 235 | 236 | 237 | 238 | 239 | 240 | 241 | 242 | 243 | 244 | 245 | 246 | 247 | 248 | 249 | 250 | 251 | 252 | 253 | 254 | 255 | 256 | 257 | 258 | 259 | 260 | 261 | 262 | 263 | 264 | 265 | 266 | 267 | 268 | 269 | 270 | 271 | 272 | 273 | 274 | 275 | 276 | 277 | 278 | 279 | 280 | 281 | 282 | 283 | 284 | 285 | 286 | 287 | 288 | 289 | 290 | 291 | 292 | 293 | 294 | 295 | 296 | 297 | 298 | 299 | 300 | 301 | 302 | 303 | 304 | 305 | 306 | 307 | 308 | 309 | 310 | 311 | 312 | 313 | 314 | 315 | 316 | 317 | 318 | 319 | 320 | 321 | 322 | 323 | 324 | 325 | 326 | 327 | 328 | 329 | 330 | 331 | 332 | 333 | 334 | 335 | 336 | 337 | 338 | 339 | 340 | 341 | 342 | 343 | 344 | 345 | 346 | -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/chain.xml: -------------------------------------------------------------------------------- 1 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 39 | 40 | 41 | 42 | 43 | 44 | 45 | 46 | 47 | 48 | 49 | 50 | 51 | 52 | 53 | 54 | 55 | 56 | 57 | 58 | 59 | 60 | 61 | 62 | 63 | 64 | 65 | 66 | 67 | 68 | 69 | 70 | 71 | 72 | 73 | 74 | 75 | 76 | 77 | 78 | 79 | 80 | 81 | 82 | 83 | 84 | 85 | 86 | 87 | 88 | 89 | 90 | 91 | 92 | 93 | 94 | 95 | 96 | 97 | 98 | 99 | 100 | 101 | 102 | 103 | 104 | 105 | 106 | 107 | 108 | 109 | 110 | 111 | 112 | 113 | 114 | 115 | 116 | 117 | 118 | 119 | 120 | 121 | 122 | 123 | 124 | 125 | 126 | 127 | 128 | 129 | 130 | 131 | 132 | 133 | 134 | 135 | 136 | 137 | 138 | 139 | 140 | 141 | 142 | 143 | 144 | 145 | 146 | 147 | 148 | 149 | 150 | 151 | 152 | 153 | 154 | 155 | 156 | 157 | 158 | 159 | 160 | 161 | 162 | 163 | 164 | 165 | 166 | 167 | 168 | 169 | 170 | 171 | 172 | 173 | 174 | 175 | 176 | 177 | 178 | 179 | 180 | 181 | 182 | 183 | 184 | 185 | 186 | 187 | 188 | 189 | 190 | 191 | 192 | 193 | 194 | 195 | 196 | 197 | 198 | 199 | 200 | 201 | 202 | 203 | 204 | 205 | 206 | 207 | 208 | 209 | 210 | 211 | 212 | 213 | 214 | 215 | 216 | 217 | 218 | 219 | 220 | 221 | 222 | 223 | 224 | 225 | 226 | 227 | -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/chain1.xml: -------------------------------------------------------------------------------- 1 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 39 | 40 | 41 | 42 | 43 | 44 | 45 | 46 | 47 | 48 | 49 | 50 | 51 | 52 | 53 | 54 | 55 | 56 | 57 | 58 | 59 | 60 | 61 | 62 | 63 | 64 | 65 | 66 | 67 | 68 | 69 | 70 | 71 | 72 | 73 | 74 | 75 | 76 | 77 | 78 | 79 | 80 | 81 | 82 | 83 | 84 | 85 | 86 | 87 | 88 | 89 | 90 | 91 | 92 | 93 | 94 | 95 | 96 | 97 | 98 | 99 | 100 | 101 | 102 | 103 | 104 | 105 | 106 | 107 | 108 | 109 | 110 | 111 | 112 | 113 | 114 | 115 | 116 | 117 | 118 | 119 | 120 | 121 | 122 | 123 | 124 | 125 | 126 | 127 | 128 | 129 | 130 | 131 | 132 | 133 | 134 | 135 | 136 | 137 | 138 | 139 | 140 | 141 | 142 | 143 | 144 | 145 | 146 | 147 | 148 | 149 | 150 | 151 | 152 | 153 | 154 | 155 | 156 | 157 | 158 | 159 | 160 | 161 | 162 | 163 | 164 | 165 | 166 | 167 | 168 | 169 | 170 | 171 | 172 | 173 | 174 | 175 | 176 | 177 | 178 | 179 | 180 | 181 | 182 | 183 | 184 | 185 | 186 | 187 | 188 | 189 | 190 | 191 | 192 | 193 | 194 | 195 | 196 | 197 | 198 | 199 | 200 | 201 | 202 | 203 | 204 | 205 | 206 | 207 | 208 | 209 | 210 | 211 | 212 | 213 | 214 | 215 | 216 | 217 | 218 | 219 | 220 | 221 | 222 | 223 | 224 | 225 | 226 | 227 | 228 | -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/joint_position_actuation.xml: -------------------------------------------------------------------------------- 1 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 | 39 | 40 | 41 | 42 | 43 | 44 | 45 | 46 | 47 | -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/F1.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/F1.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/F2.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/F2.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/F3.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/F3.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/TH1_z.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/TH1_z.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/TH2_z.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/TH2_z.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/TH3_z.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/TH3_z.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/arm_base.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/arm_base.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/arm_trunk.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/arm_trunk.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/arm_trunk_asmbly.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/arm_trunk_asmbly.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/distal_ellipsoid.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/distal_ellipsoid.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/elbow_flex.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/elbow_flex.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/elbow_rotate_motor.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/elbow_rotate_motor.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/elbow_rotate_muscle.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/elbow_rotate_muscle.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/forearm_Cy_PlateAsmbly(muscle_cone).stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/forearm_Cy_PlateAsmbly(muscle_cone).stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/forearm_Cy_PlateAsmbly.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/forearm_Cy_PlateAsmbly.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/forearm_PlateAsmbly.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/forearm_PlateAsmbly.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/forearm_electric.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/forearm_electric.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/forearm_electric_cvx.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/forearm_electric_cvx.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/forearm_muscle.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/forearm_muscle.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/forearm_simple.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/forearm_simple.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/forearm_simple_cvx.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/forearm_simple_cvx.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/forearm_weight.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/forearm_weight.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/hand_base_link.STL: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/hand_base_link.STL -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/hook_0.STL: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/hook_0.STL -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/hook_1.STL: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/hook_1.STL -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/hook_2.STL: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/hook_2.STL -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/hook_3.STL: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/hook_3.STL -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/knuckle.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/knuckle.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/lfmetacarpal.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/lfmetacarpal.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/palm.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/palm.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/upper_arm.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/upper_arm.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/upper_arm_asmbl_shoulder.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/upper_arm_asmbl_shoulder.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/upper_arm_ass.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/upper_arm_ass.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/meshes/wrist.stl: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/Adroit/resources/meshes/wrist.stl -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/Adroit/resources/tendon_torque_actuation.xml: -------------------------------------------------------------------------------- 1 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 | 39 | 40 | 41 | 42 | 43 | 44 | 45 | 46 | 47 | 48 | 49 | 50 | 51 | 52 | 53 | 54 | 55 | 56 | 57 | 58 | 59 | 60 | 61 | 62 | 63 | 64 | 65 | 66 | 67 | 68 | 69 | 70 | 71 | 72 | 73 | 74 | 75 | 76 | 77 | 78 | 79 | 80 | 81 | 82 | 83 | 84 | 85 | 86 | 87 | 88 | 89 | 90 | 91 | 92 | 93 | 94 | 95 | 96 | 97 | 98 | 99 | 100 | 101 | 102 | 103 | 104 | 105 | 106 | 107 | 108 | 109 | 110 | 111 | 112 | 113 | 114 | 115 | 116 | 117 | 118 | 119 | 120 | 121 | 122 | 123 | 124 | -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/__init__.py -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/assets/DAPG_Adroit.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 | 39 | 40 | 41 | 42 | 43 | 44 | 45 | 46 | 47 | 48 | 49 | 50 | 51 | 52 | 53 | 54 | 55 | 56 | 57 | 58 | 59 | 60 | 61 | 62 | 63 | 64 | 65 | 66 | 67 | 68 | 69 | 70 | 71 | 72 | 73 | 74 | 75 | 76 | 77 | 78 | 79 | 80 | 81 | 82 | 83 | 84 | 85 | 86 | 87 | 88 | 89 | 90 | 91 | 92 | 93 | 94 | 95 | 96 | 97 | 98 | 99 | 100 | 101 | 102 | 103 | 104 | 105 | 106 | 107 | 108 | 109 | 110 | 111 | 112 | 113 | 114 | 115 | 116 | 117 | 118 | 119 | 120 | 121 | 122 | 123 | 124 | 125 | 126 | 127 | 128 | 129 | 130 | 131 | 132 | 133 | 134 | 135 | 136 | 137 | 138 | 139 | 140 | 141 | 142 | 143 | 144 | 145 | 146 | 147 | 148 | 149 | 150 | 151 | 152 | 153 | 154 | 155 | 156 | 157 | 158 | 159 | 160 | 161 | 162 | 163 | 164 | 165 | 166 | 167 | 168 | 169 | 170 | -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/assets/DAPG_assets_hook.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 81 | -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/assets/DAPG_door.xml: -------------------------------------------------------------------------------- 1 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 | 39 | 40 | 41 | 42 | 43 | 44 | 45 | 46 | 47 | 48 | 49 | 50 | 51 | 52 | 53 | 54 | 55 | 56 | 57 | 58 | 59 | 60 | 61 | 62 | 63 | 64 | 65 | 66 | 67 | 68 | 69 | 70 | 71 | 72 | 73 | 74 | 75 | 76 | 77 | 78 | 79 | 80 | 81 | 82 | 83 | 84 | 85 | 86 | 87 | 88 | 89 | 90 | 91 | 92 | 93 | -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/assets/DAPG_hammer.xml: -------------------------------------------------------------------------------- 1 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 | 39 | 40 | 41 | 42 | 43 | 44 | 45 | 46 | 47 | 48 | 49 | 50 | 51 | 52 | 53 | 54 | 55 | 56 | 57 | 58 | 59 | 60 | 61 | 62 | 63 | 64 | 65 | 66 | 67 | 68 | 69 | 70 | 71 | 72 | 73 | 74 | 75 | 76 | 77 | 78 | 79 | 80 | 81 | 82 | 83 | 84 | 85 | 86 | 87 | 88 | 89 | 90 | 91 | 92 | 93 | 94 | 95 | 96 | 97 | 98 | 99 | 100 | 101 | 102 | 103 | 104 | 105 | 106 | 107 | 108 | 109 | 110 | 111 | 112 | -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/assets/DAPG_pen.xml: -------------------------------------------------------------------------------- 1 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 | 39 | 40 | 41 | 42 | 43 | 44 | 45 | 46 | 47 | 48 | 49 | 50 | 51 | 52 | 53 | 54 | 55 | 56 | 57 | 58 | 59 | 60 | 61 | 62 | 63 | 64 | 65 | 66 | 67 | 68 | 69 | 70 | 71 | 72 | 73 | 74 | 75 | 76 | 77 | 78 | 79 | 80 | 81 | 82 | 83 | 84 | 85 | 86 | 87 | 88 | 89 | 90 | 91 | -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/assets/DAPG_relocate.xml: -------------------------------------------------------------------------------- 1 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 | 39 | 40 | 41 | 42 | 43 | 44 | 45 | 46 | 47 | 48 | 49 | 50 | 51 | 52 | 53 | 54 | 55 | 56 | 57 | 58 | 59 | 60 | 61 | 62 | 63 | 64 | 65 | 66 | 67 | 68 | 69 | 70 | 71 | 72 | 73 | 74 | 75 | 76 | 77 | 78 | 79 | 80 | 81 | 82 | 83 | 84 | 85 | 86 | 87 | 88 | 89 | -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/assets/door_xknob_adroit.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 | 39 | 40 | 41 | 42 | 43 | 44 | 45 | 46 | 47 | 48 | 49 | 50 | 51 | 52 | 53 | 54 | 55 | 56 | 57 | 58 | 59 | 60 | 61 | 62 | 63 | 64 | 65 | 66 | 67 | 68 | 69 | 70 | 71 | 72 | 73 | 74 | 75 | 76 | 77 | -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/assets/door_xknob_adroit_extrageom.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 | 39 | 40 | 41 | 42 | 43 | 44 | 45 | 46 | 47 | 48 | 49 | 50 | 51 | 52 | 53 | 54 | 55 | 56 | 57 | 58 | 59 | 60 | 61 | 62 | 63 | 64 | 65 | 66 | 67 | 68 | 69 | 70 | 71 | 72 | 73 | 74 | 75 | 76 | 77 | 78 | 79 | 80 | -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/assets/door_xknob_hook.xml: -------------------------------------------------------------------------------- 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 | 13 | 14 | 15 | 16 | 17 | 18 | 19 | 20 | 21 | 22 | 23 | 24 | 25 | 26 | 27 | 28 | 29 | 30 | 31 | 32 | 33 | 34 | 35 | 36 | 37 | 38 | 39 | 40 | 41 | 42 | 43 | 44 | 45 | 46 | 47 | 48 | 49 | 50 | 51 | 52 | 53 | 54 | 55 | 56 | 57 | 58 | 59 | 60 | 61 | 62 | 63 | 64 | 65 | 66 | 67 | 68 | 69 | 70 | 71 | 72 | 73 | 74 | 75 | 76 | 77 | 78 | 79 | 80 | 81 | 82 | 83 | 84 | 85 | 86 | 87 | 88 | 89 | 90 | -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/assets/tasks.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/penn-pal-lab/interactive_reward_functions/645618a29c2e3f266794e15c6f27231c20fd01b6/src/env/hand_manipulation_suite/assets/tasks.jpg -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/door_lock_adroit_simple.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | from gym import utils 3 | from gym import spaces 4 | from mjrl.envs import mujoco_env 5 | from mujoco_py import MjViewer 6 | from d4rl import offline_env 7 | import os 8 | import time 9 | from stable_baselines3 import SAC 10 | 11 | 12 | def get_aliased_angle(pose): 13 | aliased_pose = pose 14 | while aliased_pose > 1.57 / 2.0: 15 | aliased_pose -= 1.57 16 | while aliased_pose < -1.57 / 2.0: 17 | aliased_pose += 1.57 18 | return aliased_pose 19 | 20 | 21 | def get_aliased_angle_ndarray(poses): 22 | # change the poses to aliased poses in place 23 | while True: 24 | over_poses = poses > 1.57 / 2.0 25 | if not np.any(over_poses): 26 | break 27 | else: 28 | poses[over_poses] -= 1.57 29 | while True: 30 | under_poses = poses < -1.57 / 2.0 31 | if not np.any(under_poses): 32 | break 33 | else: 34 | poses[under_poses] += 1.57 35 | 36 | 37 | class DoorLockEnvAdroitSimple(mujoco_env.MujocoEnv, utils.EzPickle, 38 | offline_env.OfflineEnv): 39 | def __init__(self, **kwargs): 40 | offline_env.OfflineEnv.__init__(self, **kwargs) 41 | self.door_hinge_did = 0 42 | self.door_bid = 0 43 | self.grasp_sid = 0 44 | self.handle_sid = 0 45 | self.knob_sid = 0 46 | self.extra_knob_sid = 0 47 | self.action_dim_simple = 5 48 | # Override action_space to -1, 1 49 | self.action_space = spaces.Box(low=-1.0, 50 | high=1.0, 51 | dtype=np.float32, 52 | shape=(self.action_dim_simple, )) 53 | 54 | # Use a sequence of states as observation 55 | self.obs_horizon = 1 56 | self.latch_obs_list = [] 57 | self.use_whole_state_obs = False 58 | 59 | self.t = 0 60 | self.max_traj_length = 100 61 | 62 | curr_dir = os.path.dirname(os.path.abspath(__file__)) 63 | self.sim = mujoco_env.get_sim( 64 | curr_dir + '/assets/door_xknob_adroit_extrageom.xml') 65 | self.data = self.sim.data 66 | self.model = self.sim.model 67 | 68 | self.frame_skip = 5 69 | self.metadata = { 70 | 'render.modes': ['human', 'rgb_array'], 71 | 'video.frames_per_second': int(np.round(1.0 / self.dt)) 72 | } 73 | self.mujoco_render_frames = False 74 | 75 | self.init_qpos = self.data.qpos.ravel().copy() 76 | self.init_qvel = self.data.qvel.ravel().copy() 77 | observation = self.get_obs(hist_len=self.obs_horizon) 78 | self.obs_dim = np.sum([ 79 | o.size for o in observation 80 | ]) if type(observation) is tuple else observation.size 81 | 82 | high = np.inf * np.ones(self.obs_dim) 83 | low = -high 84 | self.observation_space = spaces.Box(low, high, dtype=np.float32) 85 | 86 | self.seed() 87 | 88 | # change actuator sensitivity 89 | self.sim.model.actuator_gainprm[self.sim.model.actuator_name2id( 90 | 'A_WRJ1'):self.sim.model.actuator_name2id('A_WRJ0') + 91 | 1, :3] = np.array([10, 0, 0]) 92 | self.sim.model.actuator_gainprm[self.sim.model.actuator_name2id( 93 | 'A_FFJ3'):self.sim.model.actuator_name2id('A_THJ0') + 94 | 1, :3] = np.array([1, 0, 0]) 95 | self.sim.model.actuator_biasprm[self.sim.model.actuator_name2id( 96 | 'A_WRJ1'):self.sim.model.actuator_name2id('A_WRJ0') + 97 | 1, :3] = np.array([0, -10, 0]) 98 | self.sim.model.actuator_biasprm[self.sim.model.actuator_name2id( 99 | 'A_FFJ3'):self.sim.model.actuator_name2id('A_THJ0') + 100 | 1, :3] = np.array([0, -1, 0]) 101 | 102 | utils.EzPickle.__init__(self) 103 | self.act_mid = np.mean(self.model.actuator_ctrlrange, axis=1) 104 | self.act_rng = 0.5 * (self.model.actuator_ctrlrange[:, 1] - 105 | self.model.actuator_ctrlrange[:, 0]) 106 | self.door_hinge_did = self.model.jnt_dofadr[self.model.joint_name2id( 107 | 'door_hinge')] 108 | self.grasp_sid = self.model.site_name2id('S_grasp') 109 | self.handle_sid = self.model.site_name2id('S_handle') 110 | self.door_bid = self.model.body_name2id('frame') 111 | self.knob_sid = self.model.site_name2id('knob_hinge') 112 | self.extra_knob_sid = self.model.site_name2id('extra_knob') 113 | 114 | self.use_latch_pose_reward = False 115 | self.use_dense_reward = True 116 | self.use_engineered_reward = False 117 | self.use_verification_reward = False 118 | # TODO(for users): train an IRF policy first and put the checkpoint path below 119 | self.verifier_checkpoint_path = "checkpoints/door_lock_verifier" 120 | if self.use_verification_reward: 121 | self.pi_v = SAC.load(self.verifier_checkpoint_path) 122 | self.latch_random_reset = True 123 | self.terminate_early = False 124 | 125 | def step(self, a): 126 | a = np.clip(a, -1.0, 1.0) 127 | whole_action = np.zeros(28) 128 | whole_action[:4] = np.copy(a[:4]) 129 | whole_action[4:] = np.ones(24) * a[4] 130 | 131 | try: 132 | whole_action = self.act_mid + whole_action * \ 133 | self.act_rng # mean center and scale 134 | except: 135 | whole_action = whole_action # only for the initialization phase 136 | self.do_simulation(whole_action, self.frame_skip) 137 | 138 | done = False 139 | if self.t >= self.max_traj_length: 140 | done = True 141 | 142 | ob = self.get_obs(hist_len=self.obs_horizon) 143 | door_pose = self.data.qpos[self.door_hinge_did] 144 | 145 | reward = 0.0 146 | if self.use_dense_reward: 147 | palm_pos = self.data.site_xpos[self.grasp_sid].ravel() 148 | knob_pos = self.data.site_xpos[self.knob_sid].ravel() 149 | # close door 150 | reward += -0.1 * (door_pose - 0.0)**2 151 | reward += -1.0 * np.linalg.norm(palm_pos - knob_pos) 152 | 153 | door_closed = (door_pose <= 0.05) 154 | latch_pose = self.data.get_joint_qpos("latch") 155 | goal_achieved = True if door_closed and -0.7 < latch_pose < 0.7 else False 156 | if self.terminate_early and goal_achieved: 157 | done = True 158 | if self.use_latch_pose_reward: 159 | reward += 3.0 * door_closed 160 | reward += 10.0 * goal_achieved 161 | else: 162 | reward += 3.0 * door_closed 163 | 164 | if self.use_verification_reward and done: 165 | is_door_locked = self.verify_by_pi_v(num_steps=100) 166 | reward += 1000.0 * is_door_locked 167 | 168 | if self.use_engineered_reward and door_closed: 169 | latch_v = self.data.get_joint_qvel("latch") 170 | reward += (-latch_v) 171 | 172 | self.t += 1 173 | 174 | self.sim.model.site_rgba[self.knob_sid] = [1, 0, 0, 1] 175 | self.sim.model.site_rgba[self.extra_knob_sid] = [0, 1, 0, 0] 176 | 177 | return ob, reward, done, dict(goal_achieved=goal_achieved, 178 | door_closed=door_closed) 179 | 180 | def get_obs(self, hist_len=1): 181 | adroit_qpos = self.data.qpos.ravel()[ 182 | 1:-2] # last 2 dimensions are about door hinge and latch 183 | palm_pos = self.data.site_xpos[self.grasp_sid].ravel() 184 | door_pose = np.array([self.data.qpos[self.door_hinge_did]]) 185 | knob_pos = self.data.site_xpos[self.knob_sid].ravel() 186 | 187 | latch_pose = self.data.get_joint_qpos("latch") 188 | latch_visual_pose = get_aliased_angle(latch_pose) 189 | if len(self.latch_obs_list) < self.obs_horizon: 190 | self.latch_obs_list = [latch_visual_pose] * self.obs_horizon 191 | else: 192 | self.latch_obs_list.pop(0) 193 | self.latch_obs_list.append(latch_visual_pose) 194 | 195 | door_close = 1.0 if door_pose < 0.05 else -1.0 196 | 197 | if self.use_whole_state_obs: 198 | obs = np.concatenate([ 199 | adroit_qpos, palm_pos, door_pose, knob_pos, [latch_pose], 200 | [door_close] 201 | ]) 202 | else: 203 | obs = np.concatenate([ 204 | adroit_qpos, palm_pos, door_pose, knob_pos, 205 | self.latch_obs_list[-hist_len:], [door_close] 206 | ]) 207 | 208 | return obs 209 | 210 | def reset_model(self): 211 | qp = self.init_qpos.copy() 212 | qv = self.init_qvel.copy() 213 | self.set_state(qp, qv) 214 | 215 | self.model.body_pos[self.door_bid, 216 | 0] = self.np_random.uniform(low=-0.3, high=-0.2) 217 | self.model.body_pos[self.door_bid, 218 | 1] = self.np_random.uniform(low=0.25, high=0.35) 219 | self.model.body_pos[self.door_bid, 220 | 2] = self.np_random.uniform(low=0.252, high=0.35) 221 | 222 | self.sim.data.set_joint_qpos("latch", 1.57) 223 | self.sim.data.set_joint_qpos( 224 | "door_hinge", self.np_random.uniform(low=0.25, high=0.4)) 225 | if self.latch_random_reset: 226 | self.sim.data.set_joint_qpos( 227 | "latch", self.np_random.uniform(low=1.57, high=3.14)) 228 | self.sim.data.set_joint_qpos( 229 | "door_hinge", self.np_random.uniform(low=0.25, high=1.3)) 230 | 231 | self.sim.forward() 232 | 233 | self.t = 0 234 | return self.get_obs(hist_len=self.obs_horizon) 235 | 236 | def verify_by_pi_v(self, num_steps=100): 237 | # Reset hand pose first 238 | # Necessary to prevent Pi_t from exploiting Pi_v 239 | qp = self.init_qpos.copy() 240 | qv = self.init_qvel.copy() 241 | curr_qp = self.data.qpos.ravel().copy() 242 | curr_qp[0:-2] = qp[0:-2] 243 | self.set_state(curr_qp, qv) 244 | self.sim.forward() 245 | 246 | self.sim.model.site_rgba[self.knob_sid] = [1, 0, 0, 0] 247 | self.sim.model.site_rgba[self.extra_knob_sid] = [0, 1, 0, 1] 248 | 249 | is_door_locked = True 250 | for t in range(num_steps): 251 | self.mj_render() 252 | 253 | obs = self.get_obs(hist_len=1) 254 | action, _states = self.pi_v.predict(obs, deterministic=True) 255 | 256 | action = np.clip(action, -1.0, 1.0) 257 | whole_action = np.zeros(28) 258 | whole_action[:4] = np.copy(action[:4]) 259 | whole_action[4:] = np.ones(24) * action[4] 260 | 261 | try: 262 | whole_action = self.act_mid + whole_action * self.act_rng 263 | except: 264 | whole_action = whole_action # only for the initialization phase 265 | self.do_simulation(whole_action, self.frame_skip) 266 | 267 | door_pose = np.array([self.data.qpos[self.door_hinge_did]]) 268 | if door_pose > 0.1: 269 | is_door_locked = False 270 | break 271 | return is_door_locked 272 | 273 | def get_env_state(self): 274 | """ 275 | Get state of hand as well as objects and targets in the scene 276 | """ 277 | qp = self.data.qpos.ravel().copy() 278 | qv = self.data.qvel.ravel().copy() 279 | door_body_pos = self.model.body_pos[self.door_bid].ravel().copy() 280 | return dict(qpos=qp, qvel=qv, door_body_pos=door_body_pos) 281 | 282 | def set_env_state(self, state_dict): 283 | """ 284 | Set the state which includes hand as well as objects and targets in the scene 285 | """ 286 | qp = state_dict['qpos'] 287 | qv = state_dict['qvel'] 288 | self.set_state(qp, qv) 289 | self.model.body_pos[self.door_bid] = state_dict['door_body_pos'] 290 | self.sim.forward() 291 | 292 | def mj_viewer_setup(self): 293 | self.viewer = MjViewer(self.sim) 294 | self.viewer.cam.azimuth = 90 295 | self.sim.forward() 296 | self.viewer.cam.distance = 1.5 297 | 298 | def evaluate_success(self, paths): 299 | num_success = 0 300 | num_paths = len(paths) 301 | # success if door open for 25 steps 302 | for path in paths: 303 | if np.sum(path['env_infos']['goal_achieved']) > 25: 304 | num_success += 1 305 | success_percentage = num_success * 100.0 / num_paths 306 | return success_percentage 307 | 308 | 309 | if __name__ == "__main__": 310 | import time 311 | 312 | env = DoorLockEnvAdroitSimple() 313 | env.reset() 314 | env.mj_render() 315 | for i in range(1000): 316 | env.reset() 317 | for t in range(1000): 318 | action = env.action_space.sample() 319 | obs, reward, done, info = env.step(action) 320 | env.mj_render() 321 | if done: 322 | break 323 | -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/door_lock_verify_adroit_simple.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | from gym import utils 3 | from gym import spaces 4 | from mjrl.envs import mujoco_env 5 | from mujoco_py import MjViewer 6 | from d4rl import offline_env 7 | import os 8 | 9 | from src.env.hand_manipulation_suite.door_lock_adroit_simple import get_aliased_angle 10 | 11 | USE_EXTRA_KNOB_ENV = True 12 | 13 | 14 | class DoorLockVerifyEnv(mujoco_env.MujocoEnv, utils.EzPickle, 15 | offline_env.OfflineEnv): 16 | def __init__(self, **kwargs): 17 | offline_env.OfflineEnv.__init__(self, **kwargs) 18 | self.door_hinge_did = 0 19 | self.door_bid = 0 20 | self.grasp_sid = 0 21 | self.handle_sid = 0 22 | self.knob_sid = 0 23 | self.extra_knob_sid = 0 24 | self.is_init_locked = 0 25 | self.action_dim_simple = 5 26 | # Override action_space to -1, 1 27 | self.action_space = spaces.Box(low=-1.0, 28 | high=1.0, 29 | dtype=np.float32, 30 | shape=(self.action_dim_simple, )) 31 | 32 | self.t = 0 33 | self.max_traj_length = 100 34 | 35 | curr_dir = os.path.dirname(os.path.abspath(__file__)) 36 | self.sim = mujoco_env.get_sim( 37 | curr_dir + '/assets/door_xknob_adroit_extrageom.xml') 38 | self.data = self.sim.data 39 | self.model = self.sim.model 40 | 41 | self.frame_skip = 5 42 | self.metadata = { 43 | 'render.modes': ['human', 'rgb_array'], 44 | 'video.frames_per_second': int(np.round(1.0 / self.dt)) 45 | } 46 | self.mujoco_render_frames = False 47 | 48 | self.init_qpos = self.data.qpos.ravel().copy() 49 | self.init_qvel = self.data.qvel.ravel().copy() 50 | observation = self.get_obs() 51 | self.obs_dim = np.sum([ 52 | o.size for o in observation 53 | ]) if type(observation) is tuple else observation.size 54 | 55 | high = np.inf * np.ones(self.obs_dim) 56 | low = -high 57 | self.observation_space = spaces.Box(low, high, dtype=np.float32) 58 | 59 | self.seed() 60 | 61 | # change actuator sensitivity 62 | self.sim.model.actuator_gainprm[self.sim.model.actuator_name2id( 63 | 'A_WRJ1'):self.sim.model.actuator_name2id('A_WRJ0') + 64 | 1, :3] = np.array([10, 0, 0]) 65 | self.sim.model.actuator_gainprm[self.sim.model.actuator_name2id( 66 | 'A_FFJ3'):self.sim.model.actuator_name2id('A_THJ0') + 67 | 1, :3] = np.array([1, 0, 0]) 68 | self.sim.model.actuator_biasprm[self.sim.model.actuator_name2id( 69 | 'A_WRJ1'):self.sim.model.actuator_name2id('A_WRJ0') + 70 | 1, :3] = np.array([0, -10, 0]) 71 | self.sim.model.actuator_biasprm[self.sim.model.actuator_name2id( 72 | 'A_FFJ3'):self.sim.model.actuator_name2id('A_THJ0') + 73 | 1, :3] = np.array([0, -1, 0]) 74 | 75 | utils.EzPickle.__init__(self) 76 | ob = self.reset_model() 77 | self.act_mid = np.mean(self.model.actuator_ctrlrange, axis=1) 78 | self.act_rng = 0.5 * (self.model.actuator_ctrlrange[:, 1] - 79 | self.model.actuator_ctrlrange[:, 0]) 80 | self.door_hinge_did = self.model.jnt_dofadr[self.model.joint_name2id( 81 | 'door_hinge')] 82 | self.grasp_sid = self.model.site_name2id('S_grasp') 83 | self.handle_sid = self.model.site_name2id('S_handle') 84 | self.door_bid = self.model.body_name2id('frame') 85 | self.knob_sid = self.model.site_name2id('knob_hinge') 86 | if USE_EXTRA_KNOB_ENV: 87 | self.extra_knob_sid = self.model.site_name2id('extra_knob') 88 | 89 | def step(self, a): 90 | a = np.clip(a, -1.0, 1.0) 91 | whole_action = np.zeros(28) 92 | whole_action[:4] = np.copy(a[:4]) 93 | whole_action[4:] = np.ones(24) * a[4] 94 | 95 | try: 96 | whole_action = self.act_mid + whole_action * self.act_rng # mean center and scale 97 | except: 98 | whole_action = whole_action # only for the initialization phase 99 | self.do_simulation(whole_action, self.frame_skip) 100 | 101 | done = False 102 | if self.t >= self.max_traj_length: 103 | done = True 104 | 105 | ob = self.get_obs() 106 | door_pose = self.data.qpos[self.door_hinge_did] 107 | knob_pos = self.data.site_xpos[self.knob_sid].ravel() 108 | if USE_EXTRA_KNOB_ENV: 109 | knob_pos = self.data.site_xpos[self.extra_knob_sid].ravel() 110 | palm_pos = self.data.site_xpos[self.grasp_sid].ravel() 111 | 112 | reward = 0.0 113 | # dense reward for approaching knob 114 | reward += -4.0 * np.linalg.norm(palm_pos - knob_pos) 115 | # Decrease the rew scale to increase exploration 116 | # Investigate whether SAC subtracts the baseline reward 117 | if self.is_init_locked: 118 | # door should be closed after verification 119 | reward += -0.1 * (door_pose - 0.0)**2 120 | 121 | goal_achieved = True if door_pose <= 0.1 else False 122 | reward += 1.0 * goal_achieved 123 | 124 | if door_pose > 0.2: 125 | reward -= 3.0 126 | else: 127 | # door should be open after verification 128 | reward += -0.1 * (door_pose - 1.57)**2 129 | 130 | goal_achieved = True if door_pose >= 0.2 else False 131 | reward += 10.0 * goal_achieved 132 | 133 | if door_pose < 0.1: 134 | reward -= 3.0 135 | 136 | self.t += 1 137 | 138 | self.sim.model.site_rgba[self.knob_sid] = [1, 0, 0, 0] 139 | self.sim.model.site_rgba[self.extra_knob_sid] = [0, 1, 0, 1] 140 | 141 | return ob, reward, done, dict(goal_achieved=goal_achieved) 142 | 143 | def get_obs(self): 144 | adroit_qpos = self.data.qpos.ravel()[ 145 | 1:-2] # last 2 dimensions are about door hinge and latch 146 | palm_pos = self.data.site_xpos[self.grasp_sid].ravel() 147 | door_pose = np.array([self.data.qpos[self.door_hinge_did]]) 148 | knob_pos = self.data.site_xpos[self.knob_sid].ravel() 149 | if USE_EXTRA_KNOB_ENV: 150 | knob_pos = self.data.site_xpos[self.extra_knob_sid].ravel() 151 | 152 | latch_pose = self.data.get_joint_qpos("latch") 153 | latch_visual_pose = get_aliased_angle(latch_pose) 154 | 155 | door_close = 1.0 if door_pose < 0.05 else -1.0 156 | return np.concatenate([ 157 | adroit_qpos, palm_pos, door_pose, knob_pos, [latch_visual_pose], 158 | [door_close] 159 | ]) 160 | 161 | def reset_model(self): 162 | qp = self.init_qpos.copy() 163 | qv = self.init_qvel.copy() 164 | self.set_state(qp, qv) 165 | 166 | self.model.body_pos[self.door_bid, 167 | 0] = self.np_random.uniform(low=-0.3, high=-0.2) 168 | self.model.body_pos[self.door_bid, 169 | 1] = self.np_random.uniform(low=0.25, high=0.35) 170 | self.model.body_pos[self.door_bid, 171 | 2] = self.np_random.uniform(low=0.252, high=0.35) 172 | 173 | pos_or_neg_example = self.np_random.choice(np.array([0, 1])) 174 | if pos_or_neg_example < 0.5: 175 | # init as a positive example: door locked 176 | latch_init_pose = self.np_random.uniform(low=-0.7, high=1.0) 177 | self.is_init_locked = True 178 | else: 179 | # init as a negative example: example generated by policy 180 | # TODO (for user): here the latch_init_pose should load from pre-collected 181 | # policy rollouts 182 | latch_init_pose = self.np_random.uniform(low=1.0, high=3.1) 183 | self.is_init_locked = False 184 | 185 | self.sim.data.set_joint_qpos("latch", latch_init_pose) 186 | 187 | self.sim.forward() 188 | 189 | self.t = 0 190 | return self.get_obs() 191 | 192 | def get_env_state(self): 193 | """ 194 | Get state of hand as well as objects and targets in the scene 195 | """ 196 | qp = self.data.qpos.ravel().copy() 197 | qv = self.data.qvel.ravel().copy() 198 | door_body_pos = self.model.body_pos[self.door_bid].ravel().copy() 199 | return dict(qpos=qp, qvel=qv, door_body_pos=door_body_pos) 200 | 201 | def set_env_state(self, state_dict): 202 | """ 203 | Set the state which includes hand as well as objects and targets in the scene 204 | """ 205 | qp = state_dict['qpos'] 206 | qv = state_dict['qvel'] 207 | self.set_state(qp, qv) 208 | self.model.body_pos[self.door_bid] = state_dict['door_body_pos'] 209 | self.sim.forward() 210 | 211 | def mj_viewer_setup(self): 212 | self.viewer = MjViewer(self.sim) 213 | self.viewer.cam.azimuth = 90 214 | self.sim.forward() 215 | self.viewer.cam.distance = 1.5 216 | 217 | def evaluate_success(self, paths): 218 | num_success = 0 219 | num_paths = len(paths) 220 | # success if door open for 25 steps 221 | for path in paths: 222 | if np.sum(path['env_infos']['goal_achieved']) > 25: 223 | num_success += 1 224 | success_percentage = num_success * 100.0 / num_paths 225 | return success_percentage 226 | 227 | 228 | if __name__ == "__main__": 229 | env = DoorLockVerifyEnv() 230 | env.reset() 231 | env.mj_render() 232 | for i in range(1000): 233 | obs = env.reset() 234 | for t in range(50): 235 | action = np.zeros(env.action_space.sample().shape) 236 | obs, reward, done, info = env.step(action) 237 | env.mj_render() 238 | -------------------------------------------------------------------------------- /src/env/hand_manipulation_suite/door_v0.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | from gym import utils 3 | from gym import spaces 4 | from mjrl.envs import mujoco_env 5 | from mujoco_py import MjViewer 6 | from d4rl import offline_env 7 | import os 8 | 9 | ADD_BONUS_REWARDS = True 10 | 11 | 12 | class DoorEnvV0(mujoco_env.MujocoEnv, utils.EzPickle, offline_env.OfflineEnv): 13 | def __init__(self, **kwargs): 14 | offline_env.OfflineEnv.__init__(self, **kwargs) 15 | self.door_hinge_did = 0 16 | self.door_bid = 0 17 | self.grasp_sid = 0 18 | self.handle_sid = 0 19 | curr_dir = os.path.dirname(os.path.abspath(__file__)) 20 | mujoco_env.MujocoEnv.__init__(self, curr_dir + '/assets/DAPG_door.xml', 21 | 5) 22 | 23 | # Override action_space to -1, 1 24 | self.action_space = spaces.Box(low=-1.0, 25 | high=1.0, 26 | dtype=np.float32, 27 | shape=self.action_space.shape) 28 | 29 | # change actuator sensitivity 30 | self.sim.model.actuator_gainprm[self.sim.model.actuator_name2id( 31 | 'A_WRJ1'):self.sim.model.actuator_name2id('A_WRJ0') + 32 | 1, :3] = np.array([10, 0, 0]) 33 | self.sim.model.actuator_gainprm[self.sim.model.actuator_name2id( 34 | 'A_FFJ3'):self.sim.model.actuator_name2id('A_THJ0') + 35 | 1, :3] = np.array([1, 0, 0]) 36 | self.sim.model.actuator_biasprm[self.sim.model.actuator_name2id( 37 | 'A_WRJ1'):self.sim.model.actuator_name2id('A_WRJ0') + 38 | 1, :3] = np.array([0, -10, 0]) 39 | self.sim.model.actuator_biasprm[self.sim.model.actuator_name2id( 40 | 'A_FFJ3'):self.sim.model.actuator_name2id('A_THJ0') + 41 | 1, :3] = np.array([0, -1, 0]) 42 | 43 | utils.EzPickle.__init__(self) 44 | ob = self.reset_model() 45 | self.act_mid = np.mean(self.model.actuator_ctrlrange, axis=1) 46 | self.act_rng = 0.5 * (self.model.actuator_ctrlrange[:, 1] - 47 | self.model.actuator_ctrlrange[:, 0]) 48 | self.door_hinge_did = self.model.jnt_dofadr[self.model.joint_name2id( 49 | 'door_hinge')] 50 | self.grasp_sid = self.model.site_name2id('S_grasp') 51 | self.handle_sid = self.model.site_name2id('S_handle') 52 | self.door_bid = self.model.body_name2id('frame') 53 | 54 | def step(self, a): 55 | a = np.clip(a, -1.0, 1.0) 56 | try: 57 | a = self.act_mid + a * self.act_rng # mean center and scale 58 | except: 59 | a = a # only for the initialization phase 60 | self.do_simulation(a, self.frame_skip) 61 | ob = self.get_obs() 62 | handle_pos = self.data.site_xpos[self.handle_sid].ravel() 63 | palm_pos = self.data.site_xpos[self.grasp_sid].ravel() 64 | door_pos = self.data.qpos[self.door_hinge_did] 65 | 66 | # get to handle 67 | reward = -0.1 * np.linalg.norm(palm_pos - handle_pos) 68 | # open door 69 | reward += -0.1 * (door_pos - 1.57) * (door_pos - 1.57) 70 | # velocity cost 71 | reward += -1e-5 * np.sum(self.data.qvel**2) 72 | 73 | if ADD_BONUS_REWARDS: 74 | # Bonus 75 | if door_pos > 0.2: 76 | reward += 2 77 | if door_pos > 1.0: 78 | reward += 8 79 | if door_pos > 1.35: 80 | reward += 10 81 | 82 | goal_achieved = True if door_pos >= 1.35 else False 83 | 84 | return ob, reward, False, dict(goal_achieved=goal_achieved) 85 | 86 | def get_obs(self): 87 | # qpos for hand 88 | # xpos for obj 89 | # xpos for target 90 | qp = self.data.qpos.ravel() 91 | handle_pos = self.data.site_xpos[self.handle_sid].ravel() 92 | palm_pos = self.data.site_xpos[self.grasp_sid].ravel() 93 | door_pos = np.array([self.data.qpos[self.door_hinge_did]]) 94 | if door_pos > 1.0: 95 | door_open = 1.0 96 | else: 97 | door_open = -1.0 98 | latch_pos = qp[-1] 99 | return np.concatenate([ 100 | qp[1:-2], [latch_pos], door_pos, palm_pos, handle_pos, 101 | palm_pos - handle_pos, [door_open] 102 | ]) 103 | 104 | def reset_model(self): 105 | qp = self.init_qpos.copy() 106 | qv = self.init_qvel.copy() 107 | self.set_state(qp, qv) 108 | 109 | self.model.body_pos[self.door_bid, 110 | 0] = self.np_random.uniform(low=-0.3, high=-0.2) 111 | self.model.body_pos[self.door_bid, 112 | 1] = self.np_random.uniform(low=0.25, high=0.35) 113 | self.model.body_pos[self.door_bid, 114 | 2] = self.np_random.uniform(low=0.252, high=0.35) 115 | self.sim.forward() 116 | return self.get_obs() 117 | 118 | def get_env_state(self): 119 | """ 120 | Get state of hand as well as objects and targets in the scene 121 | """ 122 | qp = self.data.qpos.ravel().copy() 123 | qv = self.data.qvel.ravel().copy() 124 | door_body_pos = self.model.body_pos[self.door_bid].ravel().copy() 125 | return dict(qpos=qp, qvel=qv, door_body_pos=door_body_pos) 126 | 127 | def set_env_state(self, state_dict): 128 | """ 129 | Set the state which includes hand as well as objects and targets in the scene 130 | """ 131 | qp = state_dict['qpos'] 132 | qv = state_dict['qvel'] 133 | self.set_state(qp, qv) 134 | self.model.body_pos[self.door_bid] = state_dict['door_body_pos'] 135 | self.sim.forward() 136 | 137 | def mj_viewer_setup(self): 138 | self.viewer = MjViewer(self.sim) 139 | self.viewer.cam.azimuth = 90 140 | self.sim.forward() 141 | self.viewer.cam.distance = 1.5 142 | 143 | def evaluate_success(self, paths): 144 | num_success = 0 145 | num_paths = len(paths) 146 | # success if door open for 25 steps 147 | for path in paths: 148 | if np.sum(path['env_infos']['goal_achieved']) > 25: 149 | num_success += 1 150 | success_percentage = num_success * 100.0 / num_paths 151 | return success_percentage 152 | 153 | 154 | if __name__ == "__main__": 155 | import time 156 | 157 | env = DoorEnvV0() 158 | env.reset() 159 | env.mj_render() 160 | 161 | for i in range(1000): 162 | env.reset() 163 | for t in range(50): 164 | action = np.zeros(env.action_space.sample().shape) 165 | action[4:] = -1.5 166 | # action = env.action_space.sample() 167 | obs, reward, done, info = env.step(action) 168 | env.mj_render() 169 | time.sleep(0.02) 170 | 171 | for t in range(50): 172 | action = np.zeros(env.action_space.sample().shape) 173 | action[4:] = 1.5 174 | # action = env.action_space.sample() 175 | obs, reward, done, info = env.step(action) 176 | env.mj_render() 177 | time.sleep(0.02) 178 | -------------------------------------------------------------------------------- /src/robel_screw_pi_task.py: -------------------------------------------------------------------------------- 1 | import os 2 | import time 3 | 4 | from stable_baselines3 import SAC 5 | 6 | from robel.dclaw.turn import (DCLAW3_ASSET_PATH, DCLAW4_ASSET_PATH) 7 | from src.env.robel.dclaw_env import DClawScrewTask 8 | 9 | if __name__ == "__main__": 10 | from src.config import argparser 11 | 12 | config, _ = argparser() 13 | config.jobname = "real_robel_screw_valve4" 14 | figures_path = os.path.join("figures", config.jobname) 15 | os.makedirs(figures_path, exist_ok=True) 16 | checkpoint_path = os.path.join("checkpoints", config.jobname) 17 | os.makedirs(checkpoint_path, exist_ok=True) 18 | 19 | env = DClawScrewTask( 20 | asset_path=DCLAW4_ASSET_PATH, 21 | # frame_skip=80, 22 | verification_mode=False, 23 | use_verification_reward=True, 24 | # device_path='/dev/ttyUSB0', 25 | # action_mode="fixed_last_joint", 26 | use_engineered_rew=False, 27 | ) 28 | env.use_hist_obs = True 29 | env.reset() 30 | 31 | model = SAC( 32 | "MlpPolicy", 33 | env, 34 | verbose=1, 35 | gamma=0.99, 36 | batch_size=1024, 37 | target_entropy=-3.0, 38 | ) 39 | 40 | # Training/Continue training 41 | total_timesteps = 4000000 42 | save_per_timesteps = 10000 43 | 44 | for i in range(int(total_timesteps / save_per_timesteps)): 45 | print("======================================") 46 | start = time.time() 47 | model.learn(total_timesteps=save_per_timesteps) 48 | 49 | # Save model 50 | model.save(checkpoint_path + "/mfrl_sc_" + 51 | str((i + 1) * save_per_timesteps)) 52 | print("Model saved after", (i + 1) * save_per_timesteps, "timesteps") 53 | --------------------------------------------------------------------------------