File size: 11,908 Bytes
b6a9d87 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 | # SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import cv2
import gymnasium as gym
from gymnasium.envs.registration import register
import numpy as np
import simpler_env
from simpler_env.utils.env.observation_utils import get_image_from_maniskill2_obs_dict
from transforms3d import euler as te, quaternions as tq
class GoogleFractalEnv(gym.Env):
def __init__(self, env_name: str, image_size: tuple[int, int]):
env = simpler_env.make(env_name)
env._max_episode_steps = 10000
self.env = env
agent_space = env.observation_space["agent"]
print("[SimplerEnv] agent space keys:", list(agent_space.spaces.keys()))
# assert False
obs_low = env.observation_space["agent"]["eef_pos"].low
obs_high = env.observation_space["agent"]["eef_pos"].high
self.observation_space = gym.spaces.Dict(
{
"video.image": gym.spaces.Box(
low=0,
high=255,
shape=(image_size[0], image_size[1], 3),
dtype=np.uint8,
),
"state.x": gym.spaces.Box(low=obs_low[0], high=obs_high[0], shape=(1,)),
"state.y": gym.spaces.Box(low=obs_low[1], high=obs_high[1], shape=(1,)),
"state.z": gym.spaces.Box(low=obs_low[2], high=obs_high[2], shape=(1,)),
"state.rx": gym.spaces.Box(low=obs_low[3], high=obs_high[3], shape=(1,)),
"state.ry": gym.spaces.Box(low=obs_low[4], high=obs_high[4], shape=(1,)),
"state.rz": gym.spaces.Box(low=obs_low[5], high=obs_high[5], shape=(1,)),
"state.rw": gym.spaces.Box(low=obs_low[6], high=obs_high[6], shape=(1,)),
"state.gripper": gym.spaces.Box(low=obs_low[7], high=obs_high[7], shape=(1,)),
"annotation.human.action.task_description": gym.spaces.Text(max_length=512),
}
)
action_low = env.action_space.low
action_high = env.action_space.high
self.action_space = gym.spaces.Dict(
{
"action.x": gym.spaces.Box(low=action_low[0], high=action_high[0], shape=(1,)),
"action.y": gym.spaces.Box(low=action_low[1], high=action_high[1], shape=(1,)),
"action.z": gym.spaces.Box(low=action_low[2], high=action_high[2], shape=(1,)),
"action.roll": gym.spaces.Box(low=action_low[3], high=action_high[3], shape=(1,)),
"action.pitch": gym.spaces.Box(low=action_low[4], high=action_high[4], shape=(1,)),
"action.yaw": gym.spaces.Box(low=action_low[5], high=action_high[5], shape=(1,)),
"action.gripper": gym.spaces.Box(
low=action_low[6], high=action_high[6], shape=(1,)
),
}
)
self.image_size = image_size
self.previous_gripper_action = None
self.sticky_action_is_on = False
self.sticky_gripper_action = 0.0
self.gripper_action_repeat = 0
self.sticky_gripper_num_repeat = 15
def reset(self, seed=None, options=None):
self.previous_gripper_action = None
self.sticky_action_is_on = False
self.sticky_gripper_action = 0.0
self.gripper_action_repeat = 0
observation, info = self.env.reset(seed=int(seed) if seed is not None else None)
observation = self._process_observation(observation)
info["success"] = False
return observation, info
def step(self, action):
action_vector = np.concatenate(
[
action["action.x"],
action["action.y"],
action["action.z"],
action["action.roll"],
action["action.pitch"],
action["action.yaw"],
self._postprocess_gripper(action["action.gripper"]),
],
axis=0,
)
observation, reward, done, truncated, info = self.env.step(action_vector)
observation = self._process_observation(observation)
info["success"] = bool(info.get("success", False))
return observation, reward, done, truncated, info
def _process_observation(self, obs):
img = get_image_from_maniskill2_obs_dict(self.env, obs)
proprio = obs["agent"]["eef_pos"]
qunat_xyzw = np.roll(proprio[3:7], -1)
gripper_closedness = 1 - proprio[7]
return {
"video.image": cv2.resize(img, (self.image_size[1], self.image_size[0])),
"state.x": [proprio[0]],
"state.y": [proprio[1]],
"state.z": [proprio[2]],
"state.rx": [qunat_xyzw[0]],
"state.ry": [qunat_xyzw[1]],
"state.rz": [qunat_xyzw[2]],
"state.rw": [qunat_xyzw[3]],
"state.gripper": [gripper_closedness],
"annotation.human.action.task_description": self.env.unwrapped.get_language_instruction(),
}
def _postprocess_gripper(self, current_gripper_action: float) -> float:
current_gripper_action = (current_gripper_action * 2) - 1 # [0,1] -> [-1,1]
relative_gripper_action = -current_gripper_action
if np.abs(relative_gripper_action) > 0.5 and self.sticky_action_is_on is False:
self.sticky_action_is_on = True
self.sticky_gripper_action = relative_gripper_action
if self.sticky_action_is_on:
self.gripper_action_repeat += 1
relative_gripper_action = self.sticky_gripper_action
if self.gripper_action_repeat == self.sticky_gripper_num_repeat:
self.sticky_action_is_on = False
self.gripper_action_repeat = 0
self.sticky_gripper_action = 0.0
return relative_gripper_action
class WidowXBridgeEnv(gym.Env):
def __init__(self, env_name: str, image_size: tuple[int, int]):
env = simpler_env.make(env_name)
env._max_episode_steps = 10000
self.env = env
obs_low = env.observation_space["agent"]["eef_pos"].low
obs_high = env.observation_space["agent"]["eef_pos"].high
self.observation_space = gym.spaces.Dict(
{
"video.image_0": gym.spaces.Box(
low=0,
high=255,
shape=(image_size[0], image_size[1], 3),
dtype=np.uint8,
),
"state.x": gym.spaces.Box(low=obs_low[0], high=obs_high[0], shape=(1,)),
"state.y": gym.spaces.Box(low=obs_low[1], high=obs_high[1], shape=(1,)),
"state.z": gym.spaces.Box(low=obs_low[2], high=obs_high[2], shape=(1,)),
"state.roll": gym.spaces.Box(low=obs_low[3], high=obs_high[3], shape=(1,)),
"state.pitch": gym.spaces.Box(low=obs_low[4], high=obs_high[4], shape=(1,)),
"state.yaw": gym.spaces.Box(low=obs_low[5], high=obs_high[5], shape=(1,)),
"state.pad": gym.spaces.Box(low=obs_low[6], high=obs_high[6], shape=(1,)),
"state.gripper": gym.spaces.Box(low=obs_low[7], high=obs_high[7], shape=(1,)),
"annotation.human.action.task_description": gym.spaces.Text(max_length=512),
}
)
action_low = env.action_space.low
action_high = env.action_space.high
self.action_space = gym.spaces.Dict(
{
"action.x": gym.spaces.Box(low=action_low[0], high=action_high[0], shape=(1,)),
"action.y": gym.spaces.Box(low=action_low[1], high=action_high[1], shape=(1,)),
"action.z": gym.spaces.Box(low=action_low[2], high=action_high[2], shape=(1,)),
"action.roll": gym.spaces.Box(low=action_low[3], high=action_high[3], shape=(1,)),
"action.pitch": gym.spaces.Box(low=action_low[4], high=action_high[4], shape=(1,)),
"action.yaw": gym.spaces.Box(low=action_low[5], high=action_high[5], shape=(1,)),
"action.gripper": gym.spaces.Box(
low=action_low[6], high=action_high[6], shape=(1,)
),
}
)
self.image_size = image_size
# Bridge orientation adjustment
self.default_rot = np.array([[0, 0, 1.0], [0, 1.0, 0], [-1.0, 0, 0]])
def reset(self, seed=None, options=None):
observation, info = self.env.reset(seed=int(seed) if seed is not None else None)
observation = self._process_observation(observation)
info["success"] = False
return observation, info
def step(self, action):
action_vector = np.concatenate(
[
action["action.x"],
action["action.y"],
action["action.z"],
action["action.roll"],
action["action.pitch"],
action["action.yaw"],
self._postprocess_gripper(action["action.gripper"]),
],
axis=0,
)
observation, reward, done, truncated, info = self.env.step(action_vector)
observation = self._process_observation(observation)
info["success"] = bool(info.get("success", False))
return observation, reward, done, truncated, info
def _process_observation(self, obs):
img = get_image_from_maniskill2_obs_dict(self.env, obs)
proprio = obs["agent"]["eef_pos"]
rm_bridge = tq.quat2mat(proprio[3:7])
rpy_bridge_converted = te.mat2euler(rm_bridge @ self.default_rot.T)
return {
"video.image_0": cv2.resize(img, (self.image_size[1], self.image_size[0])),
"state.x": [proprio[0]],
"state.y": [proprio[1]],
"state.z": [proprio[2]],
"state.roll": [rpy_bridge_converted[0]],
"state.pitch": [rpy_bridge_converted[1]],
"state.yaw": [rpy_bridge_converted[2]],
"state.pad": [0],
"state.gripper": [proprio[7]],
"annotation.human.action.task_description": self.env.unwrapped.get_language_instruction(),
}
def _postprocess_gripper(self, action):
# trained with [0, 1], 0 close, 1 open -> convert to SimplerEnv [-1, 1]
return 2.0 * (action > 0.5) - 1.0
def register_simpler_envs():
# Google/Fractal
for env_name in [
"google_robot_pick_coke_can",
"google_robot_pick_object",
"google_robot_move_near",
"google_robot_open_drawer",
"google_robot_close_drawer",
"google_robot_place_in_closed_drawer",
]:
register(
id=f"simpler_env_google/{env_name}",
entry_point="gr00t.eval.sim.SimplerEnv.simpler_env:GoogleFractalEnv",
kwargs={"env_name": env_name, "image_size": (256, 320)},
)
# WidowX/Bridge
for env_name in [
"widowx_spoon_on_towel",
"widowx_carrot_on_plate",
"widowx_stack_cube",
"widowx_put_eggplant_in_basket",
"widowx_put_eggplant_in_sink",
"widowx_open_drawer",
"widowx_close_drawer",
]:
register(
id=f"simpler_env_widowx/{env_name}",
entry_point="gr00t.eval.sim.SimplerEnv.simpler_env:WidowXBridgeEnv",
kwargs={"env_name": env_name, "image_size": (256, 256)},
)
|