File size: 10,442 Bytes
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 | """
Step-by-step planning collection tool.
All step-by-step collection interfaces uniformly return a 5-tuple:
(obs_batch, reward_batch, terminated_batch, truncated_batch, info_batch)
- obs_batch/info_batch: dict[str, list], keys from single step dictionary.
- reward_batch: torch.float32, shape [N].
- terminated_batch/truncated_batch: torch.bool, shape [N].
"""
import numpy as np
import torch
def _collapse_singleton_lists(value, key=None):
"""Recursively unwrap singleton lists while preserving non-singleton lists."""
if key == "task_goal":
return value
while isinstance(value, list) and len(value) == 1:
value = value[0]
return value
def _to_scalar(value):
"""Convert scalar or tensor value to Python scalar."""
if isinstance(value, torch.Tensor):
if value.numel() == 0:
return 0
return value.reshape(-1)[0].item()
return value
def _snapshot_value(value):
"""Snapshot values that might reuse underlying memory to avoid overwriting previous frames in subsequent steps."""
if isinstance(value, torch.Tensor):
return value.detach().clone()
if isinstance(value, np.ndarray):
return value.copy()
if isinstance(value, dict):
return {k: _snapshot_value(v) for k, v in value.items()}
if isinstance(value, list):
return [_snapshot_value(v) for v in value]
if isinstance(value, tuple):
return tuple(_snapshot_value(v) for v in value)
return value
def _snapshot_step(out):
"""Deep copy snapshot of single step output."""
if not (isinstance(out, tuple) and len(out) == 5):
return out
obs, reward, terminated, truncated, info = out
return (
_snapshot_value(obs),
_snapshot_value(reward),
_snapshot_value(terminated),
_snapshot_value(truncated),
_snapshot_value(info),
)
def _is_columnar_dict(batch_dict, n):
if not isinstance(batch_dict, dict):
return False
for value in batch_dict.values():
if not isinstance(value, list):
return False
if len(value) != n:
return False
return True
def _output_to_steps(out):
"""
Normalize step output to "list of raw step tuples".
Supports both single step tuple and unified batch tuple input formats.
"""
if isinstance(out, tuple) and len(out) == 5:
obs_part, reward_part, terminated_part, truncated_part, info_part = out
if (
isinstance(reward_part, torch.Tensor)
and isinstance(terminated_part, torch.Tensor)
and isinstance(truncated_part, torch.Tensor)
and reward_part.ndim == 1
and terminated_part.ndim == 1
and truncated_part.ndim == 1
):
n = int(reward_part.numel())
if (
terminated_part.numel() == n
and truncated_part.numel() == n
and _is_columnar_dict(obs_part, n)
and _is_columnar_dict(info_part, n)
):
steps = []
obs_keys = list(obs_part.keys())
info_keys = list(info_part.keys())
for idx in range(n):
obs = {k: _snapshot_value(obs_part[k][idx]) for k in obs_keys}
info = {k: _snapshot_value(info_part[k][idx]) for k in info_keys}
steps.append(
(
obs,
_snapshot_value(reward_part[idx]),
_snapshot_value(terminated_part[idx]),
_snapshot_value(truncated_part[idx]),
info,
)
)
return steps
return [_snapshot_step(out)]
def _dicts_to_columnar_dict(dict_steps):
"""
Convert step dictionaries to dict[str, list], filling missing keys with None.
"""
n = len(dict_steps)
out = {}
for idx, item in enumerate(dict_steps):
current = item if isinstance(item, dict) else {}
for key in current:
if key not in out:
out[key] = [None] * idx
for key in out:
out[key].append(_collapse_singleton_lists(current.get(key, None), key=key))
for key in out:
if len(out[key]) < n:
out[key].extend([None] * (n - len(out[key])))
return out
def empty_step_batch():
"""Return an empty batch following the unified contract."""
return (
{},
torch.empty(0, dtype=torch.float32),
torch.empty(0, dtype=torch.bool),
torch.empty(0, dtype=torch.bool),
{},
)
def to_step_batch(collected_steps):
"""
Convert collected step tuples to unified batch output.
collected_steps: [(obs, reward, terminated, truncated, info), ...]
"""
if not collected_steps:
return empty_step_batch()
obs_steps = [x[0] for x in collected_steps]
reward_steps = [_to_scalar(x[1]) for x in collected_steps]
terminated_steps = [bool(_to_scalar(x[2])) for x in collected_steps]
truncated_steps = [bool(_to_scalar(x[3])) for x in collected_steps]
info_steps = [x[4] for x in collected_steps]
obs_batch = _dicts_to_columnar_dict(obs_steps)
info_batch = _dicts_to_columnar_dict(info_steps)
reward_batch = torch.tensor(reward_steps, dtype=torch.float32)
terminated_batch = torch.tensor(terminated_steps, dtype=torch.bool)
truncated_batch = torch.tensor(truncated_steps, dtype=torch.bool)
return (obs_batch, reward_batch, terminated_batch, truncated_batch, info_batch)
def concat_step_batches(batches):
"""
Concatenate multiple unified batches into one unified batch.
"""
valid = []
for batch in batches:
if batch is None:
continue
obs_batch, reward_batch, terminated_batch, truncated_batch, info_batch = batch
if reward_batch.numel() == 0:
continue
valid.append((obs_batch, reward_batch, terminated_batch, truncated_batch, info_batch))
if not valid:
return empty_step_batch()
obs_out = {}
info_out = {}
reward_out = []
terminated_out = []
truncated_out = []
n_total = 0
for obs_batch, reward_batch, terminated_batch, truncated_batch, info_batch in valid:
n = int(reward_batch.numel())
for key in obs_batch:
if key not in obs_out:
obs_out[key] = [None] * n_total
for key in obs_out:
values = obs_batch.get(key, None)
if values is None:
obs_out[key].extend([None] * n)
else:
obs_out[key].extend(_collapse_singleton_lists(v, key=key) for v in values)
for key in info_batch:
if key not in info_out:
info_out[key] = [None] * n_total
for key in info_out:
values = info_batch.get(key, None)
if values is None:
info_out[key].extend([None] * n)
else:
info_out[key].extend(_collapse_singleton_lists(v, key=key) for v in values)
reward_out.append(reward_batch.reshape(-1).to(torch.float32))
terminated_out.append(terminated_batch.reshape(-1).to(torch.bool))
truncated_out.append(truncated_batch.reshape(-1).to(torch.bool))
n_total += n
return (
obs_out,
torch.cat(reward_out, dim=0) if reward_out else torch.empty(0, dtype=torch.float32),
torch.cat(terminated_out, dim=0) if terminated_out else torch.empty(0, dtype=torch.bool),
torch.cat(truncated_out, dim=0) if truncated_out else torch.empty(0, dtype=torch.bool),
info_out,
)
def _collect_dense_steps(planner, fn):
"""
Intercept planner.env.step when running fn(), collecting raw step tuples.
If fn returns -1, return -1; otherwise return collected result list.
"""
collected = []
original_step = planner.env.step
def _step(action):
out = original_step(action)
collected.extend(_output_to_steps(out))
return out
planner.env.step = _step
try:
result = fn()
if result == -1:
return -1
return collected
finally:
planner.env.step = original_step
def _run_with_dense_collection(planner, fn):
"""
Run fn() and return unified batch; return -1 if fn returns -1.
"""
collected = _collect_dense_steps(planner, fn)
if collected == -1:
return -1
return to_step_batch(collected)
def move_to_pose_with_RRTStar(planner, pose):
"""
Call planner.move_to_pose_with_RRTStar(pose) and return unified batch.
Return -1 on planning failure.
"""
return _run_with_dense_collection(
planner, lambda: planner.move_to_pose_with_RRTStar(pose)
)
def move_to_pose_with_screw(planner, pose):
"""
Call planner.move_to_pose_with_screw(pose) and return unified batch.
Return -1 on planning failure.
"""
return _run_with_dense_collection(
planner, lambda: planner.move_to_pose_with_screw(pose)
)
def close_gripper(planner):
"""
Call planner.close_gripper() and return unified batch.
Return -1 on failure.
"""
return _run_with_dense_collection(planner, lambda: planner.close_gripper())
def open_gripper(planner):
"""
Call planner.open_gripper() and return unified batch.
Return -1 on failure.
"""
return _run_with_dense_collection(planner, lambda: planner.open_gripper())
# ---- Call Relationships ----
#
# _collect_dense_steps:
# - DemonstrationWrapper.get_demonstration_trajectory()
# Wrap entire solve_callable, monkey-patch planner.env.step to collect all underlying steps
#
# _run_with_dense_collection:
# - OraclePlannerDemonstrationWrapper
# Wrap solve() in solve_options, collect all underlying steps and directly return unified batch
#
# move_to_pose_with_RRTStar:
# - Execute single step move in MultiStepDemonstrationWrapper
# (MultiStepDemonstrationWrapper.py line 106)
#
# move_to_pose_with_screw:
# - Currently no external call, reserved as symmetric API to move_to_pose_with_RRTStar
#
# close_gripper:
# - Execute gripper close in MultiStepDemonstrationWrapper
# (MultiStepDemonstrationWrapper.py line 112)
#
# open_gripper:
# - Execute gripper open in MultiStepDemonstrationWrapper
# (MultiStepDemonstrationWrapper.py line 121)
|