File size: 16,738 Bytes
06c11b0 467d2ce 06c11b0 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 | from typing import Any, Dict, Union
import numpy as np
import sapien
import torch
import mani_skill.envs.utils.randomization as randomization
from mani_skill.agents.robots import SO100, Fetch, Panda
from mani_skill.envs.sapien_env import BaseEnv
from mani_skill.envs.tasks.tabletop.pick_cube_cfgs import PICK_CUBE_CONFIGS
from mani_skill.sensors.camera import CameraConfig
from mani_skill.utils import sapien_utils
from mani_skill.utils.building import actors
from mani_skill.utils.registration import register_env
from mani_skill.utils.scene_builder.table import TableSceneBuilder
from mani_skill.utils.structs.pose import Pose
#Robomme
import matplotlib.pyplot as plt
import random
from mani_skill.utils.geometry.rotation_conversions import (
euler_angles_to_matrix,
matrix_to_quaternion,
)
from .utils import *
from .utils.subgoal_evaluate_func import static_check
from .utils.object_generation import spawn_fixed_cube, build_board_with_hole
from .utils import reset_panda
from .utils.difficulty import normalize_robomme_difficulty
from ..logging_utils import logger
PICK_CUBE_DOC_STRING = """**Task Description:**
A simple task where the objective is to grasp a red cube with the {robot_id} robot and move it to a target goal position. This is also the *baseline* task to test whether a robot with manipulation
capabilities can be simulated and trained properly. Hence there is extra code for some robots to set them up properly in this environment as well as the table scene builder.
**Randomizations:**
- the cube's xy position is randomized on top of a table in the region [0.1, 0.1] x [-0.1, -0.1]. It is placed flat on the table
- the cube's z-axis rotation is randomized to a random angle
- the target goal position (marked by a green sphere) of the cube has its xy position randomized in the region [0.1, 0.1] x [-0.1, -0.1] and z randomized in [0, 0.3]
**Success Conditions:**
- the cube position is within `goal_thresh` (default 0.025m) euclidean distance of the goal position
- the robot is static (q velocity < 0.2)
"""
@register_env("ButtonUnmask")
class ButtonUnmask(BaseEnv):
_sample_video_link = "https://github.com/haosulab/ManiSkill/raw/main/figures/environment_demos/PickCube-v1_rt.mp4"
SUPPORTED_ROBOTS = [
"panda",
"fetch",
"xarm6_robotiq",
"so100",
"widowxai",
]
agent: Union[Panda]
goal_thresh = 0.025
cube_spawn_half_size = 0.05
cube_spawn_center = (0, 0)
config_hard = {
'bin':15,
"pick":2,
}
config_easy = {
'bin':3,
"pick":1,
}
config_medium = {
'bin':5,
"pick":1,
}
# Combine into a dictionary
configs = {
'hard': config_hard,
'easy': config_easy,
'medium': config_medium
}
def __init__(self, *args, robot_uids="panda_wristcam", robot_init_qpos_noise=0,seed=0,Robomme_video_episode=None,Robomme_video_path=None,
**kwargs):
self.use_demonstrationwrapper=False
self.demonstration_record_traj=False
self.robot_init_qpos_noise = robot_init_qpos_noise
if robot_uids in PICK_CUBE_CONFIGS:
cfg = PICK_CUBE_CONFIGS[robot_uids]
else:
cfg = PICK_CUBE_CONFIGS["panda"]
self.cube_half_size = cfg["cube_half_size"]
self.goal_thresh = cfg["goal_thresh"]
self.cube_spawn_half_size = cfg["cube_spawn_half_size"]
self.cube_spawn_center = cfg["cube_spawn_center"]
self.max_goal_height = cfg["max_goal_height"]
self.sensor_cam_eye_pos = cfg["sensor_cam_eye_pos"]
self.sensor_cam_target_pos = cfg["sensor_cam_target_pos"]
self.human_cam_eye_pos = cfg["human_cam_eye_pos"]
self.human_cam_target_pos = cfg["human_cam_target_pos"]
self.seed = seed
normalized_robomme_difficulty = normalize_robomme_difficulty(
kwargs.pop("difficulty", None)
)
self.robomme_failure_recovery = bool(
kwargs.pop("robomme_failure_recovery", False)
)
self.robomme_failure_recovery_mode = kwargs.pop(
"robomme_failure_recovery_mode", None
)
if isinstance(self.robomme_failure_recovery_mode, str):
self.robomme_failure_recovery_mode = (
self.robomme_failure_recovery_mode.lower()
)
if normalized_robomme_difficulty is not None:
self.difficulty = normalized_robomme_difficulty
else:
seed_mod = seed % 3
if seed_mod == 0:
self.difficulty = "easy"
elif seed_mod == 1:
self.difficulty = "medium"
else: # seed_mod == 2
self.difficulty = "hard"
#self.difficulty = "hard"
# Use seed to randomly determine number of repetitions (1-5) arbitrarily
generator = torch.Generator()
generator.manual_seed(seed)
self.num_repeats = torch.randint(1, 6, (1,), generator=generator).item()
logger.debug(f"Task will repeat {self.num_repeats} times (pickup-drop cycles)")
self.generator = generator
super().__init__(*args, robot_uids=robot_uids, **kwargs)
@property
def _default_sensor_configs(self):
pose = sapien_utils.look_at(
eye=self.sensor_cam_eye_pos, target=self.sensor_cam_target_pos
)
camera_eye=[0.3,0,0.4]
camera_target =[0,0,-0.2]
pose = sapien_utils.look_at(
eye=camera_eye, target=camera_target
)
return [CameraConfig("base_camera", pose, 256, 256, np.pi / 2, 0.01, 100)]
@property
def _default_human_render_camera_configs(self):
pose = sapien_utils.look_at(
eye=self.human_cam_eye_pos, target=self.human_cam_target_pos
)
return CameraConfig("render_camera", pose, 512, 512, 1, 0.01, 100)
def _load_agent(self, options: dict):
super()._load_agent(options, sapien.Pose(p=[-0.615, 0, 0]))
def _load_scene(self, options: dict):
generator = torch.Generator()
generator.manual_seed(self.seed)
self.table_scene = TableSceneBuilder(
self, robot_init_qpos_noise=self.robot_init_qpos_noise
)
self.table_scene.build()
button_obb_1 = build_button(
self,
center_xy=(-0.2, 0),
scale=1.5,
generator=generator,
name="button",
randomize=True,
randomize_range=(0.1, 0.1)
)
# Store first button before building second one
self.button_left = self.button
self.button_joint_1 = self.button_joint
avoid = [button_obb_1]
# Generate 3 bins
self.spawned_bins = []
for i in range(self.configs[self.difficulty]['bin']):
try:
bin_actor = spawn_random_bin(
self,
avoid=avoid, # Use current avoidance list, containing all spawned objects
region_center=[0, 0],
region_half_size=0.2,
min_gap=self.cube_half_size*2, # bins need larger gap, increased to 6x to avoid collision
name_prefix=f"bin_{i}",
max_trials=256,
generator=generator
)
except RuntimeError as e:
break
self.spawned_bins.append(bin_actor)
# Assign bin to self.bin_0, self.bin_1 etc. attributes
setattr(self, f"bin_{i}", bin_actor)
# Add newly generated bin to avoidance list
avoid.append(bin_actor)
# Generate 3 dynamic cubes under each bin (using fixed position, colors red, green, blue)
spawned_dynamic_cubes = []
cube_colors = [(1, 0, 0, 1), (0, 1, 0, 1), (0, 0, 1, 1)] # Red, Green, Blue
color_names = ["red", "green", "blue"]
# Use seed to randomly shuffle color order
shuffle_indices = torch.randperm(len(cube_colors), generator=generator).tolist()
cube_colors = [cube_colors[i] for i in shuffle_indices]
color_names = [color_names[i] for i in shuffle_indices]
# Store color_names for RecordWrapper access
self.color_names = color_names
# Generate cubes only for first 3 bins
for i, bin_actor in enumerate(self.spawned_bins[:3]):
# Get bin position
bin_pos = bin_actor.pose.p
if isinstance(bin_pos, torch.Tensor):
bin_pos = bin_pos[0].detach().cpu().numpy()
cube_position = [bin_pos[0], bin_pos[1]]
# Generate cube using fixed position, colors red, green, blue
cube_actor = spawn_fixed_cube(
self,
position=cube_position,
half_size=self.cube_half_size/1.2,
color=cube_colors[i], # Use red, green, blue in order
name_prefix=f"target_cube_{color_names[i]}",
yaw=0.0, # No rotation
dynamic=True
)
spawned_dynamic_cubes.append(cube_actor)
# Assign cube to self.target_cube_red, self.target_cube_green, self.target_cube_blue etc. attributes
setattr(self, f"target_cube_{color_names[i]}", cube_actor)
# Also store using numeric index for easy access
setattr(self, f"target_cube_{i}", cube_actor)
# Add newly generated cube to avoidance list
avoid.append(cube_actor)
tasks = [
{
"func": lambda: is_button_pressed(self, obj=self.button_left),
"name": "press the button",
"subgoal_segment":"press the button at <>",
"choice_label": "press the button",
"demonstration": False,
"failure_func":None,
"solve": lambda env, planner: solve_button(env, planner, obj=self.button_left),
"segment":self.cap_link,
},]
tasks.append(
{
"func": (lambda: is_bin_pickup(self, obj=self.bin_0)),
"name": f"pick up the container that hides the {self.color_names[0]} cube",
"subgoal_segment":f"pick up the container at <> that hides the {self.color_names[0]} cube",
"choice_label": "pick up the container",
"demonstration": False,
"failure_func": lambda: [
is_any_bin_pickup(self,[bin for bin in self.spawned_bins if bin != self.bin_0]), ],
"solve": lambda env, planner: [solve_pickup_bin(env, planner, obj=self.bin_0)],
"segment":self.bin_0,
})
if self.configs[self.difficulty]['pick']>1:
tasks.append({
"func": (lambda: is_bin_putdown(self, obj=self.bin_0)),
"name": "put down the container",
"subgoal_segment":"put down the container",
"choice_label": "put down the container",
"demonstration": False,
"failure_func": lambda:is_any_bin_pickup(self,[bin for bin in self.spawned_bins if bin != self.bin_0]),
"solve": lambda env, planner: solve_putdown_whenhold(env, planner),
})
tasks.append(
{
"func": (lambda: is_bin_pickup(self, obj=self.bin_1)),
"name": f"pick up the container that hides the {self.color_names[1]} cube",
"subgoal_segment":f"pick up the container at <> that hides the {self.color_names[1]} cube",
"choice_label": "pick up the container",
"demonstration": False,
"failure_func": lambda: is_any_bin_pickup(self,[bin for bin in self.spawned_bins if bin != self.bin_1]),
"solve": lambda env, planner: solve_pickup_bin(env, planner, obj=self.bin_1),
"segment":self.bin_1,
})
self.task_list = tasks
# Set recovery related attributes
# Record pickup related task indices and items for recovery
self.recovery_pickup_indices, self.recovery_pickup_tasks = task4recovery(self.task_list)
if self.robomme_failure_recovery:
# Only inject an intentional failed grasp when recovery mode is enabled
self.fail_grasp_task_index = inject_fail_grasp(
self.task_list,
generator=self.generator,
mode=self.robomme_failure_recovery_mode,
)
else:
self.fail_grasp_task_index = None
def _initialize_episode(self, env_idx: torch.Tensor, options: dict):
with torch.device(self.device):
b = len(env_idx)
self.table_scene.initialize(env_idx)
qpos=reset_panda.get_reset_panda_param("qpos")
self.agent.reset(qpos)
def _get_obs_extra(self, info: Dict):
return dict()
def evaluate(self,solve_complete_eval=False):
self.successflag=torch.tensor([False])
self.failureflag = torch.tensor([False])
# Use encapsulated sequence task check function
if(self.use_demonstrationwrapper==False):# change subgoal after planner ends during recording
if solve_complete_eval==True:
allow_subgoal_change_this_timestep=True
else:
allow_subgoal_change_this_timestep=False
else:# during demonstration, video needs to call evaluate(solve_complete_eval), video ends and flag changes in demonstrationwrapper
if solve_complete_eval==True or self.demonstration_record_traj==False:
allow_subgoal_change_this_timestep=True
else:
allow_subgoal_change_this_timestep=False
all_tasks_completed, current_task_name, task_failed ,self.current_task_specialflag= sequential_task_check(self, self.task_list,allow_subgoal_change_this_timestep=allow_subgoal_change_this_timestep)
#print(f"Current Task: {current_task_name}")
# If task failed, mark as failed immediately
if task_failed:
self.failureflag = torch.tensor([True])
logger.debug(f"Task failed: {current_task_name}")
# If static_check succeeds or all tasks completed, set success flag
if all_tasks_completed and not task_failed:
self.successflag = torch.tensor([True])
return {
"success": self.successflag,
"fail": self.failureflag,
}
def compute_dense_reward(self, obs: Any, action: torch.Tensor, info: Dict):
tcp_to_obj_dist = torch.linalg.norm(
self.agent.tcp_pose.p - self.agent.tcp_pose.p, axis=1
)
reaching_reward = 1 - torch.tanh(5 * tcp_to_obj_dist)
reward = reaching_reward*0
return reward
def compute_normalized_dense_reward(
self, obs: Any, action: torch.Tensor, info: Dict
):
return self.compute_dense_reward(obs=obs, action=action, info=info) / 5
def _get_other_bins_for_pair(self, idx_a: int, idx_b: int):
"""Return bins that are not part of the provided pair indices."""
if not hasattr(self, "spawned_bins"):
return []
total_bins = len(self.spawned_bins)
if idx_a >= total_bins or idx_b >= total_bins:
return []
# Prefer precomputed lists when available
if hasattr(self, "otherbins") and idx_a < len(self.otherbins):
other_candidates = [
bin_actor
for bin_actor in self.otherbins[idx_a]
if bin_actor is not self.spawned_bins[idx_b]
]
return other_candidates
return [
bin_actor
for i, bin_actor in enumerate(self.spawned_bins)
if i not in (idx_a, idx_b)
]
#Robomme
def step(self, action: Union[None, np.ndarray, torch.Tensor, Dict]):
timestep = self.elapsed_steps
#Lift and drop bins (bin_0 to bin_4 if they exist)
for i in range(15):
bin_attr = f"bin_{i}"
if hasattr(self, bin_attr):
lift_and_drop_objects_back_to_original(
self,
obj=getattr(self, bin_attr),
start_step=0,
end_step=32*2,
cur_step=timestep,
)
obs, reward, terminated, truncated, info = super().step(action)
return obs, reward, terminated, truncated, info
|