dodox/fancy_gym
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Removed old manual registration of mp-envs (port to fancy registry)

Browse Source
This commit is contained in:
Dominik Moritz Roth 2023-07-23 10:25:28 +02:00
parent 9ba3fa9dbc
commit a069aaac65
1 changed files with 79 additions and 189 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

266
fancy_gym/envs/__init__.py
View File

@ -15,6 +15,8 @@ from .mujoco.reacher.reacher import ReacherEnv, MAX_EPISODE_STEPS_REACHER
from .mujoco.reacher.mp_wrapper import MPWrapper as MPWrapper_Reacher from .mujoco.reacher.mp_wrapper import MPWrapper as MPWrapper_Reacher
from .mujoco.ant_jump.ant_jump import MAX_EPISODE_STEPS_ANTJUMP from .mujoco.ant_jump.ant_jump import MAX_EPISODE_STEPS_ANTJUMP
from .mujoco.beerpong.beerpong import MAX_EPISODE_STEPS_BEERPONG, FIXED_RELEASE_STEP from .mujoco.beerpong.beerpong import MAX_EPISODE_STEPS_BEERPONG, FIXED_RELEASE_STEP
from .mujoco.beerpong.mp_wrapper import MPWrapper as MPWrapper_Beerpong
from .mujoco.beerpong.mp_wrapper import MPWrapper_FixedRelease as MPWrapper_Beerpong_FixedRelease
from .mujoco.half_cheetah_jump.half_cheetah_jump import MAX_EPISODE_STEPS_HALFCHEETAHJUMP from .mujoco.half_cheetah_jump.half_cheetah_jump import MAX_EPISODE_STEPS_HALFCHEETAHJUMP
from .mujoco.hopper_jump.hopper_jump import MAX_EPISODE_STEPS_HOPPERJUMP from .mujoco.hopper_jump.hopper_jump import MAX_EPISODE_STEPS_HOPPERJUMP
from .mujoco.hopper_jump.hopper_jump_on_box import MAX_EPISODE_STEPS_HOPPERJUMPONBOX from .mujoco.hopper_jump.hopper_jump_on_box import MAX_EPISODE_STEPS_HOPPERJUMPONBOX
@ -25,6 +27,10 @@ from .mujoco.box_pushing.box_pushing_env import BoxPushingDense, BoxPushingTempo
BoxPushingTemporalSpatialSparse, MAX_EPISODE_STEPS_BOX_PUSHING BoxPushingTemporalSpatialSparse, MAX_EPISODE_STEPS_BOX_PUSHING
from .mujoco.table_tennis.table_tennis_env import TableTennisEnv, TableTennisWind, TableTennisGoalSwitching, \ from .mujoco.table_tennis.table_tennis_env import TableTennisEnv, TableTennisWind, TableTennisGoalSwitching, \
MAX_EPISODE_STEPS_TABLE_TENNIS MAX_EPISODE_STEPS_TABLE_TENNIS
from .mujoco.table_tennis.mp_wrapper import TT_MPWrapper as MPWrapper_TableTennis
from .mujoco.table_tennis.mp_wrapper import TT_MPWrapper_Replan as MPWrapper_TableTennis_Replan
from .mujoco.table_tennis.mp_wrapper import TTVelObs_MPWrapper as MPWrapper_TableTennis_VelObs
from .mujoco.table_tennis.mp_wrapper import TTVelObs_MPWrapper_Replan as MPWrapper_TableTennis_VelObs_Replan
# Classic Control # Classic Control
# Simple Reacher # Simple Reacher
@ -129,46 +135,73 @@ register(
} }
) )
gym_register( # TODO: Add [MPs] later when finished (old TODO I moved here during refactor)
register(
id='AntJump-v0', id='AntJump-v0',
entry_point='fancy_gym.envs.mujoco:AntJumpEnv', entry_point='fancy_gym.envs.mujoco:AntJumpEnv',
max_episode_steps=MAX_EPISODE_STEPS_ANTJUMP, max_episode_steps=MAX_EPISODE_STEPS_ANTJUMP,
add_mp_types=[],
) )
gym_register( register(
id='HalfCheetahJump-v0', id='HalfCheetahJump-v0',
entry_point='fancy_gym.envs.mujoco:HalfCheetahJumpEnv', entry_point='fancy_gym.envs.mujoco:HalfCheetahJumpEnv',
max_episode_steps=MAX_EPISODE_STEPS_HALFCHEETAHJUMP, max_episode_steps=MAX_EPISODE_STEPS_HALFCHEETAHJUMP,
add_mp_types=[],
) )
gym_register( register(
id='HopperJumpOnBox-v0', id='HopperJumpOnBox-v0',
entry_point='fancy_gym.envs.mujoco:HopperJumpOnBoxEnv', entry_point='fancy_gym.envs.mujoco:HopperJumpOnBoxEnv',
max_episode_steps=MAX_EPISODE_STEPS_HOPPERJUMPONBOX, max_episode_steps=MAX_EPISODE_STEPS_HOPPERJUMPONBOX,
add_mp_types=[],
) )
gym_register( register(
id='HopperThrow-v0', id='HopperThrow-v0',
entry_point='fancy_gym.envs.mujoco:HopperThrowEnv', entry_point='fancy_gym.envs.mujoco:HopperThrowEnv',
max_episode_steps=MAX_EPISODE_STEPS_HOPPERTHROW, max_episode_steps=MAX_EPISODE_STEPS_HOPPERTHROW,
add_mp_types=[],
) )
gym_register( register(
id='HopperThrowInBasket-v0', id='HopperThrowInBasket-v0',
entry_point='fancy_gym.envs.mujoco:HopperThrowInBasketEnv', entry_point='fancy_gym.envs.mujoco:HopperThrowInBasketEnv',
max_episode_steps=MAX_EPISODE_STEPS_HOPPERTHROWINBASKET, max_episode_steps=MAX_EPISODE_STEPS_HOPPERTHROWINBASKET,
add_mp_types=[],
) )
gym_register( register(
id='Walker2DJump-v0', id='Walker2DJump-v0',
entry_point='fancy_gym.envs.mujoco:Walker2dJumpEnv', entry_point='fancy_gym.envs.mujoco:Walker2dJumpEnv',
max_episode_steps=MAX_EPISODE_STEPS_WALKERJUMP, max_episode_steps=MAX_EPISODE_STEPS_WALKERJUMP,
add_mp_types=[],
) )
gym_register( register( # [MPDone
id='BeerPong-v0', id='BeerPong-v0',
entry_point='fancy_gym.envs.mujoco:BeerPongEnv', entry_point='fancy_gym.envs.mujoco:BeerPongEnv',
mp_wrapper=MPWrapper_Beerpong,
max_episode_steps=MAX_EPISODE_STEPS_BEERPONG, max_episode_steps=MAX_EPISODE_STEPS_BEERPONG,
add_mp_types=['ProMP'],
)
# Here we use the same reward as in BeerPong-v0, but now consider after the release,
# only one time step, i.e. we simulate until the end of th episode
register(
id='BeerPongStepBased-v0',
entry_point='fancy_gym.envs.mujoco:BeerPongEnvStepBasedEpisodicReward',
mp_wrapper=MPWrapper_Beerpong_FixedRelease,
max_episode_steps=FIXED_RELEASE_STEP,
add_mp_types=['ProMP'],
)
register(
id='BeerPongFixedRelease-v0',
entry_point='fancy_gym.envs.mujoco:BeerPongEnv',
mp_wrapper=MPWrapper_Beerpong_FixedRelease,
max_episode_steps=FIXED_RELEASE_STEP,
add_mp_types=['ProMP'],
) )
# Box pushing environments with different rewards # Box pushing environments with different rewards
@ -188,209 +221,66 @@ for reward_type in ["Dense", "TemporalSparse", "TemporalSpatialSparse"]:
max_episode_steps=MAX_EPISODE_STEPS_BOX_PUSHING, max_episode_steps=MAX_EPISODE_STEPS_BOX_PUSHING,
) )
# Here we use the same reward as in BeerPong-v0, but now consider after the release,
# only one time step, i.e. we simulate until the end of th episode
gym_register(
id='BeerPongStepBased-v0',
entry_point='fancy_gym.envs.mujoco:BeerPongEnvStepBasedEpisodicReward',
max_episode_steps=FIXED_RELEASE_STEP,
)
# Table Tennis environments # Table Tennis environments
for ctxt_dim in [2, 4]: for ctxt_dim in [2, 4]:
gym_register( register(
id='TableTennis{}D-v0'.format(ctxt_dim), id='TableTennis{}D-v0'.format(ctxt_dim),
entry_point='fancy_gym.envs.mujoco:TableTennisEnv', entry_point='fancy_gym.envs.mujoco:TableTennisEnv',
mp_wrapper=MPWrapper_TableTennis,
max_episode_steps=MAX_EPISODE_STEPS_TABLE_TENNIS, max_episode_steps=MAX_EPISODE_STEPS_TABLE_TENNIS,
add_mp_types=['ProMP', 'ProDMP'],
kwargs={ kwargs={
"ctxt_dim": ctxt_dim, "ctxt_dim": ctxt_dim,
'frame_skip': 4, 'frame_skip': 4,
} }
) )
gym_register( register(
id='TableTennis{}DReplan-v0'.format(ctxt_dim),
entry_point='fancy_gym.envs.mujoco:TableTennisEnv',
mp_wrapper=MPWrapper_TableTennis,
max_episode_steps=MAX_EPISODE_STEPS_TABLE_TENNIS,
add_mp_types=['ProDMP'],
kwargs={
"ctxt_dim": ctxt_dim,
'frame_skip': 4,
}
)
register(
id='TableTennisWind-v0', id='TableTennisWind-v0',
entry_point='fancy_gym.envs.mujoco:TableTennisWind', entry_point='fancy_gym.envs.mujoco:TableTennisWind',
mp_wrapper=MPWrapper_TableTennis_VelObs,
add_mp_types=['ProMP', 'ProDMP'],
max_episode_steps=MAX_EPISODE_STEPS_TABLE_TENNIS, max_episode_steps=MAX_EPISODE_STEPS_TABLE_TENNIS,
) )
gym_register( register(
id='TableTennisWindReplan-v0',
entry_point='fancy_gym.envs.mujoco:TableTennisWind',
mp_wrapper=MPWrapper_TableTennis_VelObs_Replan,
add_mp_types=['ProDMP'],
max_episode_steps=MAX_EPISODE_STEPS_TABLE_TENNIS,
)
register(
id='TableTennisGoalSwitching-v0', id='TableTennisGoalSwitching-v0',
entry_point='fancy_gym.envs.mujoco:TableTennisGoalSwitching', entry_point='fancy_gym.envs.mujoco:TableTennisGoalSwitching',
mp_wrapper=MPWrapper_TableTennis,
add_mp_types=['ProMP', 'ProDMP'],
max_episode_steps=MAX_EPISODE_STEPS_TABLE_TENNIS, max_episode_steps=MAX_EPISODE_STEPS_TABLE_TENNIS,
kwargs={ kwargs={
'goal_switching_step': 99 'goal_switching_step': 99
} }
) )
register(
# Beerpong ProMP id='TableTennisGoalSwitchingReplan-v0',
_versions = ['BeerPong-v0'] entry_point='fancy_gym.envs.mujoco:TableTennisGoalSwitching',
for _v in _versions: mp_wrapper=MPWrapper_TableTennis_Replan,
_name = _v.split("-") add_mp_types=['ProDMP'],
_env_id = f'{_name[0]}ProMP-{_name[1]}' max_episode_steps=MAX_EPISODE_STEPS_TABLE_TENNIS,
kwargs_dict_bp_promp = deepcopy(DEFAULT_BB_DICT_ProMP) kwargs={
kwargs_dict_bp_promp['wrappers'].append(mujoco.beerpong.MPWrapper) 'goal_switching_step': 99
kwargs_dict_bp_promp['phase_generator_kwargs']['learn_tau'] = True }
kwargs_dict_bp_promp['controller_kwargs']['p_gains'] = np.array([1.5, 5, 2.55, 3, 2., 2, 1.25])
kwargs_dict_bp_promp['controller_kwargs']['d_gains'] = np.array([0.02333333, 0.1, 0.0625, 0.08, 0.03, 0.03, 0.0125])
kwargs_dict_bp_promp['basis_generator_kwargs']['num_basis'] = 2
kwargs_dict_bp_promp['basis_generator_kwargs']['num_basis_zero_start'] = 2
kwargs_dict_bp_promp['name'] = _v
gym_register(
id=_env_id,
entry_point='fancy_gym.utils.make_env_helpers:make_bb_env_helper',
kwargs=kwargs_dict_bp_promp
) )
ALL_FANCY_MOVEMENT_PRIMITIVE_ENVIRONMENTS["ProMP"].append(_env_id)
# BP with Fixed release
_versions = ["BeerPongStepBased-v0", 'BeerPong-v0']
for _v in _versions:
if _v != 'BeerPong-v0':
_name = _v.split("-")
_env_id = f'{_name[0]}ProMP-{_name[1]}'
else:
_env_id = 'BeerPongFixedReleaseProMP-v0'
kwargs_dict_bp_promp = deepcopy(DEFAULT_BB_DICT_ProMP)
kwargs_dict_bp_promp['wrappers'].append(mujoco.beerpong.MPWrapper)
kwargs_dict_bp_promp['phase_generator_kwargs']['tau'] = 0.62
kwargs_dict_bp_promp['controller_kwargs']['p_gains'] = np.array([1.5, 5, 2.55, 3, 2., 2, 1.25])
kwargs_dict_bp_promp['controller_kwargs']['d_gains'] = np.array([0.02333333, 0.1, 0.0625, 0.08, 0.03, 0.03, 0.0125])
kwargs_dict_bp_promp['basis_generator_kwargs']['num_basis'] = 2
kwargs_dict_bp_promp['basis_generator_kwargs']['num_basis_zero_start'] = 2
kwargs_dict_bp_promp['name'] = _v
gym_register(
id=_env_id,
entry_point='fancy_gym.utils.make_env_helpers:make_bb_env_helper',
kwargs=kwargs_dict_bp_promp
)
ALL_FANCY_MOVEMENT_PRIMITIVE_ENVIRONMENTS["ProMP"].append(_env_id)
########################################################################################################################
# Table Tennis needs to be fixed according to Zhou's implementation
# TODO: Add later when finished
# ########################################################################################################################
#
# ## AntJump
# _versions = ['AntJump-v0']
# for _v in _versions:
# _name = _v.split("-")
# _env_id = f'{_name[0]}ProMP-{_name[1]}'
# kwargs_dict_ant_jump_promp = deepcopy(DEFAULT_BB_DICT_ProMP)
# kwargs_dict_ant_jump_promp['wrappers'].append(mujoco.ant_jump.MPWrapper)
# kwargs_dict_ant_jump_promp['name'] = _v
# gym_register(
# id=_env_id,
# entry_point='fancy_gym.utils.make_env_helpers:make_bb_env_helper',
# kwargs=kwargs_dict_ant_jump_promp
# )
# ALL_FANCY_MOVEMENT_PRIMITIVE_ENVIRONMENTS["ProMP"].append(_env_id)
#
# ########################################################################################################################
#
# ## HalfCheetahJump
# _versions = ['HalfCheetahJump-v0']
# for _v in _versions:
# _name = _v.split("-")
# _env_id = f'{_name[0]}ProMP-{_name[1]}'
# kwargs_dict_halfcheetah_jump_promp = deepcopy(DEFAULT_BB_DICT_ProMP)
# kwargs_dict_halfcheetah_jump_promp['wrappers'].append(mujoco.half_cheetah_jump.MPWrapper)
# kwargs_dict_halfcheetah_jump_promp['name'] = _v
# gym_register(
# id=_env_id,
# entry_point='fancy_gym.utils.make_env_helpers:make_bb_env_helper',
# kwargs=kwargs_dict_halfcheetah_jump_promp
# )
# ALL_FANCY_MOVEMENT_PRIMITIVE_ENVIRONMENTS["ProMP"].append(_env_id)
#
# ########################################################################################################################
# Table Tennis
_versions = ['TableTennis2D-v0', 'TableTennis4D-v0', 'TableTennisWind-v0', 'TableTennisGoalSwitching-v0']
for _v in _versions:
_name = _v.split("-")
_env_id = f'{_name[0]}ProMP-{_name[1]}'
kwargs_dict_tt_promp = deepcopy(DEFAULT_BB_DICT_ProMP)
if _v == 'TableTennisWind-v0':
kwargs_dict_tt_promp['wrappers'].append(mujoco.table_tennis.TTVelObs_MPWrapper)
else:
kwargs_dict_tt_promp['wrappers'].append(mujoco.table_tennis.TT_MPWrapper)
kwargs_dict_tt_promp['name'] = _v
kwargs_dict_tt_promp['controller_kwargs']['p_gains'] = 0.5 * np.array([1.0, 4.0, 2.0, 4.0, 1.0, 4.0, 1.0])
kwargs_dict_tt_promp['controller_kwargs']['d_gains'] = 0.5 * np.array([0.1, 0.4, 0.2, 0.4, 0.1, 0.4, 0.1])
kwargs_dict_tt_promp['phase_generator_kwargs']['learn_tau'] = False
kwargs_dict_tt_promp['phase_generator_kwargs']['learn_delay'] = False
kwargs_dict_tt_promp['phase_generator_kwargs']['tau_bound'] = [0.8, 1.5]
kwargs_dict_tt_promp['phase_generator_kwargs']['delay_bound'] = [0.05, 0.15]
kwargs_dict_tt_promp['basis_generator_kwargs']['num_basis'] = 3
kwargs_dict_tt_promp['basis_generator_kwargs']['num_basis_zero_start'] = 1
kwargs_dict_tt_promp['basis_generator_kwargs']['num_basis_zero_goal'] = 1
kwargs_dict_tt_promp['black_box_kwargs']['verbose'] = 2
gym_register(
id=_env_id,
entry_point='fancy_gym.utils.make_env_helpers:make_bb_env_helper',
kwargs=kwargs_dict_tt_promp
)
ALL_FANCY_MOVEMENT_PRIMITIVE_ENVIRONMENTS["ProMP"].append(_env_id)
for _v in _versions:
_name = _v.split("-")
_env_id = f'{_name[0]}ProDMP-{_name[1]}'
kwargs_dict_tt_prodmp = deepcopy(DEFAULT_BB_DICT_ProDMP)
if _v == 'TableTennisWind-v0':
kwargs_dict_tt_prodmp['wrappers'].append(mujoco.table_tennis.TTVelObs_MPWrapper)
else:
kwargs_dict_tt_prodmp['wrappers'].append(mujoco.table_tennis.TT_MPWrapper)
kwargs_dict_tt_prodmp['name'] = _v
kwargs_dict_tt_prodmp['controller_kwargs']['p_gains'] = 0.5 * np.array([1.0, 4.0, 2.0, 4.0, 1.0, 4.0, 1.0])
kwargs_dict_tt_prodmp['controller_kwargs']['d_gains'] = 0.5 * np.array([0.1, 0.4, 0.2, 0.4, 0.1, 0.4, 0.1])
kwargs_dict_tt_prodmp['trajectory_generator_kwargs']['weights_scale'] = 0.7
kwargs_dict_tt_prodmp['trajectory_generator_kwargs']['auto_scale_basis'] = True
kwargs_dict_tt_prodmp['trajectory_generator_kwargs']['relative_goal'] = True
kwargs_dict_tt_prodmp['trajectory_generator_kwargs']['disable_goal'] = True
kwargs_dict_tt_prodmp['phase_generator_kwargs']['tau_bound'] = [0.8, 1.5]
kwargs_dict_tt_prodmp['phase_generator_kwargs']['delay_bound'] = [0.05, 0.15]
kwargs_dict_tt_prodmp['phase_generator_kwargs']['learn_tau'] = True
kwargs_dict_tt_prodmp['phase_generator_kwargs']['learn_delay'] = True
kwargs_dict_tt_prodmp['basis_generator_kwargs']['num_basis'] = 3
kwargs_dict_tt_prodmp['basis_generator_kwargs']['alpha'] = 25.
kwargs_dict_tt_prodmp['basis_generator_kwargs']['basis_bandwidth_factor'] = 3
kwargs_dict_tt_prodmp['phase_generator_kwargs']['alpha_phase'] = 3
gym_register(
id=_env_id,
entry_point='fancy_gym.utils.make_env_helpers:make_bb_env_helper',
kwargs=kwargs_dict_tt_prodmp
)
ALL_FANCY_MOVEMENT_PRIMITIVE_ENVIRONMENTS["ProDMP"].append(_env_id)
for _v in _versions:
_name = _v.split("-")
_env_id = f'{_name[0]}ReplanProDMP-{_name[1]}'
kwargs_dict_tt_prodmp = deepcopy(DEFAULT_BB_DICT_ProDMP)
if _v == 'TableTennisWind-v0':
kwargs_dict_tt_prodmp['wrappers'].append(mujoco.table_tennis.TTVelObs_MPWrapper)
else:
kwargs_dict_tt_prodmp['wrappers'].append(mujoco.table_tennis.TT_MPWrapper)
kwargs_dict_tt_prodmp['name'] = _v
kwargs_dict_tt_prodmp['controller_kwargs']['p_gains'] = 0.5 * np.array([1.0, 4.0, 2.0, 4.0, 1.0, 4.0, 1.0])
kwargs_dict_tt_prodmp['controller_kwargs']['d_gains'] = 0.5 * np.array([0.1, 0.4, 0.2, 0.4, 0.1, 0.4, 0.1])
kwargs_dict_tt_prodmp['trajectory_generator_kwargs']['auto_scale_basis'] = False
kwargs_dict_tt_prodmp['trajectory_generator_kwargs']['goal_offset'] = 1.0
kwargs_dict_tt_prodmp['phase_generator_kwargs']['tau_bound'] = [0.8, 1.5]
kwargs_dict_tt_prodmp['phase_generator_kwargs']['delay_bound'] = [0.05, 0.15]
kwargs_dict_tt_prodmp['phase_generator_kwargs']['learn_tau'] = True
kwargs_dict_tt_prodmp['phase_generator_kwargs']['learn_delay'] = True
kwargs_dict_tt_prodmp['basis_generator_kwargs']['num_basis'] = 2
kwargs_dict_tt_prodmp['basis_generator_kwargs']['alpha'] = 25.
kwargs_dict_tt_prodmp['basis_generator_kwargs']['basis_bandwidth_factor'] = 3
kwargs_dict_tt_prodmp['phase_generator_kwargs']['alpha_phase'] = 3
kwargs_dict_tt_prodmp['black_box_kwargs']['max_planning_times'] = 3
kwargs_dict_tt_prodmp['black_box_kwargs']['replanning_schedule'] = lambda pos, vel, obs, action, t: t % 50 == 0
gym_register(
id=_env_id,
entry_point='fancy_gym.utils.make_env_helpers:make_bb_env_helper',
kwargs=kwargs_dict_tt_prodmp
)
ALL_FANCY_MOVEMENT_PRIMITIVE_ENVIRONMENTS["ProDMP"].append(_env_id)