File size: 17,186 Bytes
f96995c |
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 |
import numpy as np
from pathlib import Path
from copy import deepcopy
from functools import partial
import torch
import torch.nn as nn
import warp as wp
from dgl.geometry import farthest_point_sampler
import random
import kornia
import open3d as o3d
from tqdm import tqdm, trange
from pgnd.sim import Friction, CacheDiffSimWithFrictionBatch, MPMStaticsBatch, MPMCollidersBatch
from pgnd.utils import get_root, mkdir
from pgnd.ffmpeg import make_video
from train_eval import transform_gripper_points
from gs.convert import read_splat
root: Path = get_root(__file__)
def fps(x, n, device, random_start=False):
start_idx = random.randint(0, x.shape[0] - 1) if random_start else 0
fps_idx = farthest_point_sampler(x[None], n, start_idx=start_idx)[0]
fps_idx = fps_idx.to(x.device)
return fps_idx
class DynamicsModule:
def __init__(self, cfg, exp_root, ckpt_path, batch_size, num_steps_total):
wp.init()
wp.ScopedTimer.enabled = False
wp.set_module_options({'fast_math': False})
wp.config.verify_autograd_array_access = True
self.exp_root = exp_root
self.batch_size = batch_size
self.num_steps_total = num_steps_total
gpus = [int(gpu) for gpu in cfg.gpus]
self.gpus = gpus
wp_devices = [wp.get_device(f'cuda:{gpu}') for gpu in gpus]
torch_devices = [torch.device(f'cuda:{gpu}') for gpu in gpus]
device_count = len(torch_devices)
assert device_count == 1
wp_device = wp_devices[0]
torch_device = torch_devices[0]
self.device = torch_device
self.cfg = cfg
n_history = cfg.sim.n_history
material: nn.Module = getattr(pgnd.material, cfg.env.blob.material.material.cls)(cfg.env.blob.material.material, n_history)
material.set_params(num_grids=cfg.sim.num_grids)
material.to(torch_device)
material.requires_grad_(False)
material.train(False)
friction: nn.Module = Friction(np.array([cfg.sim.friction])[None].repeat(batch_size, axis=0)) # type: ignore
friction.to(torch_device)
assert len(list(friction.parameters())) > 0
friction.requires_grad_(False)
friction.train(False)
ckpt = torch.load(ckpt_path, map_location=torch_device)
material.load_state_dict(ckpt['material'])
self.material = material
self.friction = friction
self.material.eval()
self.friction.eval()
self.preprocess_metadata = {}
self.downsample_indices = None
self.target_state = None
if cfg.sim.gripper_points: # do some manual transformation here
pts, colors, scales, quats, opacities = read_splat('experiments/log/gs/ckpts/gripper_new.splat')
n_gripper_particles = 500
R = np.array(
[[1, 0, 0],
[0, 0, -1],
[0, 1, 0]]
)
eef_global_T = np.array([cfg.env.blob.eef_t[0], cfg.env.blob.eef_t[1], cfg.env.blob.eef_t[2] - 0.01]) # 1018_sloth: 0.185, 1018_rope_short: 0.013)
pts = pts + eef_global_T
pts = pts @ R.T
scale = cfg.sim.preprocess_scale
pts = pts * scale
axis = np.array([0, 1, 0])
angle = -27 # degree
R = o3d.geometry.get_rotation_matrix_from_axis_angle(axis * np.pi / 180 * angle)
translation = np.array([-0.015, 0.06, 0])
pts = pts @ R.T
pts = pts + translation
R = np.array(
[[0, 0, 1],
[-1, 0, 0],
[0, -1, 0]]
)
pts = pts @ R.T
gripper_pts = torch.from_numpy(pts).to(torch.float32).to(self.device)
downsample_indices = fps(gripper_pts, n_gripper_particles, self.device, random_start=True)
gripper_pts = gripper_pts[downsample_indices]
self.gripper_pts = gripper_pts
def set_target_state(self, target_state):
self.target_state = target_state
def reset_model(self, x=None):
return
def reset_preprocess_meta(self, pts):
cfg = self.cfg
dx = cfg.sim.num_grids[-1]
p_x = torch.tensor(pts).to(torch.float32).to(self.device)
R = torch.tensor(
[[1, 0, 0],
[0, 0, -1],
[0, 1, 0]]
).to(p_x.device).to(p_x.dtype)
p_x_rotated = p_x @ R.T
scale = cfg.sim.preprocess_scale
p_x_rotated_scaled = p_x_rotated * scale
cfg = self.cfg
if cfg.sim.preprocess_with_table:
global_translation = torch.tensor([
0.5 - (p_x_rotated_scaled[:, 0].max() + p_x_rotated_scaled[:, 0].min()) / 2,
dx * (cfg.env.blob.clip_bound + 0.5) + 1e-5 - p_x_rotated_scaled[:, 1].min(),
0.5 - (p_x_rotated_scaled[:, 2].max() + p_x_rotated_scaled[:, 2].min()) / 2,
], dtype=p_x_rotated_scaled.dtype, device=p_x_rotated_scaled.device)
else:
global_translation = torch.tensor([
0.5 - (p_x_rotated_scaled[:, 0].max() + p_x_rotated_scaled[:, 0].min()) / 2,
0.5 - (p_x_rotated_scaled[:, 1].max() + p_x_rotated_scaled[:, 1].min()) / 2,
0.5 - (p_x_rotated_scaled[:, 2].max() + p_x_rotated_scaled[:, 2].min()) / 2,
], dtype=p_x_rotated_scaled.dtype, device=p_x_rotated_scaled.device)
self.preprocess_metadata = {
'R': R,
'scale': scale,
'global_translation': global_translation,
}
def reset_downsample_indices(self, pts, uniform=True):
cfg = self.cfg
if uniform:
downsample_indices = fps(pts, cfg.sim.n_particles, self.device, random_start=True)
else:
downsample_indices = torch.randperm(pts.shape[0])[:cfg.sim.n_particles]
self.downsample_indices = downsample_indices
def rollout(self, pts, eef_xyz, eef_rot, eef_gripper, pts_his=None):
cfg = self.cfg
# preprocess eef
# eef_xyz: (batch_size, n_look_forward + 1, n_grippers, 3)
# eef_rot: (batch_size, n_look_forward + 1, n_grippers, 3, 3)
# eef_gripper: (batch_size, n_look_forward + 1, n_grippers, 1) 0: close, 1: open
batch_size = eef_xyz.shape[0]
assert eef_xyz.shape[1] == eef_rot.shape[1] == eef_gripper.shape[1]
R = self.preprocess_metadata['R']
scale = self.preprocess_metadata['scale']
global_translation = self.preprocess_metadata['global_translation']
eef_xyz = eef_xyz @ R.T
eef_xyz = eef_xyz * scale
eef_xyz += global_translation
eef_rot = eef_rot @ R.T
eef_quat = kornia.geometry.conversions.rotation_matrix_to_quaternion(eef_rot)
n_frames = eef_xyz.shape[1] - 1
eef_vel = torch.zeros_like(eef_xyz[:, 1:]) # (batch_size, n_frames, n_grippers, 3)
eef_vel = (eef_xyz[:, 1:] - eef_xyz[:, :-1]) / cfg.sim.dt
eef_rot_this = kornia.geometry.conversions.quaternion_to_rotation_matrix(eef_quat[:, :-1].reshape(-1, 4)) # (batch_size * n_frames * n_grippers, 3, 3)
eef_rot_next = kornia.geometry.conversions.quaternion_to_rotation_matrix(eef_quat[:, 1:].reshape(-1, 4)) # (batch_size * n_frames * n_grippers, 3, 3)
eef_rot_delta = eef_rot_this.bmm(eef_rot_next.inverse())
eef_aa = kornia.geometry.conversions.rotation_matrix_to_axis_angle(eef_rot_delta) # (batch_size * n_frames * n_grippers, 3)
eef_quat_vel = torch.zeros((batch_size, n_frames, cfg.sim.num_grippers, 3)).to(self.device).to(torch.float32)
eef_quat_vel = eef_aa.reshape(batch_size, n_frames, -1, 3) / cfg.sim.dt # (batch_size, n_frames, n_grippers, 3), radian per second
grippers = torch.zeros((batch_size, n_frames, cfg.sim.num_grippers, 15)).to(self.device).to(torch.float32)
grippers[:, :, :, :3] = eef_xyz[:, :-1] # not using the last position although we gonna arrive there
grippers[:, :, :, 3:6] = eef_vel
grippers[:, :, :, 6:10] = eef_quat[:, :-1]
grippers[:, :, :, 10:13] = eef_quat_vel
grippers[:, :, :, 13] = cfg.env.blob.gripper_radius
grippers[:, :, :, 14] = eef_gripper[:, :-1].squeeze(-1)
# preprocess pts
x = pts[self.downsample_indices]
R = self.preprocess_metadata['R']
scale = self.preprocess_metadata['scale']
global_translation = self.preprocess_metadata['global_translation']
# data frame to model frame
x = x @ R.T
x = x * scale
x = x + global_translation
x = x[None].repeat(batch_size, 1, 1)
x_pred, v_pred = self.rollout_preprocessed(x, grippers=grippers) # assumes static
# inverse preprocess
x_pred = x_pred - global_translation
x_pred = x_pred / scale
x_pred = x_pred @ R
v_pred = v_pred / scale
v_pred = v_pred @ R
return x_pred, v_pred
@torch.no_grad()
def rollout_preprocessed(self, x, v=None, x_his=None, v_his=None, grippers=None):
cfg = self.cfg
# reset model
wp_devices = [wp.get_device(f'cuda:{gpu}') for gpu in self.gpus]
torch_devices = [torch.device(f'cuda:{gpu}') for gpu in self.gpus]
device_count = len(torch_devices)
assert device_count == 1
wp_device = wp_devices[0]
torch_device = torch_devices[0]
batch_size = x.shape[0]
num_particles = x.shape[1]
assert num_particles == cfg.sim.n_particles
assert batch_size == self.batch_size
clip_bound = torch.tensor(cfg.env.blob.clip_bound)
if cfg.sim.gripper_points:
gripper_points = self.gripper_pts.clone()[None].repeat(batch_size, 1, 1) # (bsz, num_grippers, 3)
gripper_x, gripper_v, gripper_mask = transform_gripper_points(cfg, gripper_points, grippers) # (bsz, num_steps, num_grippers, 3)
num_gripper_particles = gripper_x.shape[2]
num_particles_orig = num_particles
num_particles = num_particles + num_gripper_particles
num_grippers = 0
else:
gripper_x = None
gripper_v = None
gripper_mask = None
num_particles_orig = num_particles
num_gripper_particles = 0
num_grippers = cfg.sim.num_grippers
sim = CacheDiffSimWithFrictionBatch(cfg, num_steps_total, batch_size, self.wp_device, requires_grad=True)
statics = StaticsBatch()
statics.init(shape=(batch_size, num_particles), device=self.wp_device)
statics.update_clip_bound(clip_bound)
colliders = CollidersBatch()
colliders.init(shape=(batch_size, num_grippers), device=wp_device)
if v is None:
v = torch.zeros_like(x)
if cfg.sim.n_history > 0:
if x_his is None:
x_his = x.clone().repeat(1, 1, cfg.sim.n_history)
if v_his is None:
v_his = v.clone().repeat(1, 1, cfg.sim.n_history)
colliders.initialize_grippers(grippers[:, 0])
assert cfg.sim.skip_frame == cfg.sim.interval # required for the following line
xs = []
vs = []
for step in trange(self.num_steps_total):
colliders.update_grippers(grippers[:, step]) # ignore gripper radius
if cfg.sim.gripper_forcing:
x_in = x.clone()
else:
x_in = None
if cfg.sim.gripper_points:
assert gripper_x is not None and gripper_v is not None and gripper_mask is not None
x = torch.cat([x, gripper_x[:, step]], dim=1) # gripper_x: (bsz, num_steps, num_particles, 3)
v = torch.cat([v, gripper_v[:, step]], dim=1)
x_his = torch.cat([x_his, torch.zeros((gripper_x.shape[0], gripper_x.shape[2], cfg.sim.n_history * 3), device=x_his.device, dtype=x_his.dtype)], dim=1) # type: ignore
v_his = torch.cat([v_his, torch.zeros((gripper_x.shape[0], gripper_x.shape[2], cfg.sim.n_history * 3), device=v_his.device, dtype=v_his.dtype)], dim=1) # type: ignore
if C.shape[1] < num_particles:
C = torch.cat([C, torch.zeros((gripper_x.shape[0], gripper_x.shape[2], 3, 3), device=C.device, dtype=C.dtype)], dim=1)
if F.shape[1] < num_particles:
F = torch.cat([F, torch.eye(3, device=F.device).unsqueeze(0).unsqueeze(0).repeat(gripper_x.shape[0], gripper_x.shape[2], 1, 1)], dim=1)
if enabled.shape[1] < num_particles:
enabled = torch.cat([enabled, gripper_mask[:, step]], dim=1)
statics.update_enabled(enabled.cpu())
pred = self.material(x, v, x_his, v_his, C, F)
if pred.isnan().any():
print('pred isnan', pred.min().item(), pred.max().item())
break
if pred.isinf().any():
print('pred isinf', pred.min().item(), pred.max().item())
break
x, v = sim(statics, colliders, step, x, v, self.friction.mu, pred)
if cfg.sim.gripper_forcing:
assert not cfg.sim.gripper_points
assert grippers is not None and x_in is not None
gripper_xyz = grippers[:, step, :, :3] # (bsz, num_grippers, 3)
gripper_v = grippers[:, step, :, 3:6] # (bsz, num_grippers, 3)
x_from_gripper = x_in[:, None] - gripper_xyz[:, :, None] # (bsz, num_grippers, num_particles, 3)
x_gripper_distance = torch.norm(x_from_gripper, dim=-1) # (bsz, num_grippers, num_particles)
x_gripper_distance_mask = x_gripper_distance < cfg.env.blob.gripper_radius
x_gripper_distance_mask = x_gripper_distance_mask.unsqueeze(-1).repeat(1, 1, 1, 3) # (bsz, num_particles, num_grippers, 3)
gripper_v_expand = gripper_v[:, :, None].repeat(1, 1, num_particles, 1) # (bsz, num_grippers, num_particles, 3)
gripper_closed = grippers[:, step, :, -1] < 0.5 # (bsz, num_grippers) # 1: open, 0: close
x_gripper_distance_mask = torch.logical_and(x_gripper_distance_mask, gripper_closed[:, :, None, None].repeat(1, 1, num_particles, 3))
gripper_quat_vel = grippers[:, step, :, 10:13] # (bsz, num_grippers, 3)
gripper_angular_vel = torch.norm(gripper_quat_vel, dim=-1, keepdim=True) # (bsz, num_grippers, 1)
gripper_quat_axis = gripper_quat_vel / (gripper_angular_vel + 1e-10) # (bsz, num_grippers, 3)
grid_from_gripper_axis = x_from_gripper - \
(gripper_quat_axis[:, :, None] * x_from_gripper).sum(dim=-1, keepdim=True) * gripper_quat_axis[:, :, None] # (bsz, num_grippers, num_particles, 3)
gripper_v_expand = torch.cross(gripper_quat_vel[:, :, None], grid_from_gripper_axis, dim=-1) + gripper_v_expand
for i in range(gripper_xyz.shape[1]):
x_gripper_distance_mask_single = x_gripper_distance_mask[:, i]
x[x_gripper_distance_mask_single] = x_in[x_gripper_distance_mask_single] + cfg.sim.dt * gripper_v_expand[:, i][x_gripper_distance_mask_single]
v[x_gripper_distance_mask_single] = gripper_v_expand[:, i][x_gripper_distance_mask_single]
if cfg.sim.n_history > 0:
assert x_his is not None and v_his is not None
if cfg.sim.gripper_points:
x_his_particles = torch.cat([x_his[:, :num_particles_orig].reshape(batch_size, num_particles_orig, -1, 3)[:, :, 1:], x[:, :num_particles_orig, None].detach()], dim=2)
v_his_particles = torch.cat([v_his[:, :num_particles_orig].reshape(batch_size, num_particles_orig, -1, 3)[:, :, 1:], v[:, :num_particles_orig, None].detach()], dim=2)
x_his = x_his_particles.reshape(batch_size, num_particles_orig, -1)
v_his = v_his_particles.reshape(batch_size, num_particles_orig, -1)
else:
x_his = torch.cat([x_his.reshape(batch_size, num_particles, -1, 3)[:, :, 1:], x[:, :, None].detach()], dim=2)
v_his = torch.cat([v_his.reshape(batch_size, num_particles, -1, 3)[:, :, 1:], v[:, :, None].detach()], dim=2)
x_his = x_his.reshape(batch_size, num_particles, -1)
v_his = v_his.reshape(batch_size, num_particles, -1)
if cfg.sim.gripper_points:
x = x[:, :num_particles_orig]
v = v[:, :num_particles_orig]
enabled = enabled[:, :num_particles_orig]
colliders_save = colliders.export()
colliders_save = {key: torch.from_numpy(colliders_save[key])[0].to(x.device).to(x.dtype) for key in colliders_save}
xs.append(x.detach().clone())
vs.append(v.detach().clone())
xs = torch.stack(xs, dim=1) # (batch_size, num_steps_total, num_particles, 3)
vs = torch.stack(vs, dim=1) # (batch_size, num_steps_total, num_particles, 3)
return xs, vs
|