File size: 10,091 Bytes
5c2b46e | 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 | """
Optical simulator for vitai sensor simulator.
Generate tactile rgb image using sensor camera depth info.
This file implements the optical simulator using mlp rendering.
Strict image resolution is 480 x 480 for gf225 sensor.
For other tactile resolutions lower than that, downsampling is applied.
"""
from __future__ import annotations
import os
import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
from typing import TYPE_CHECKING
import cv2
import omni.ui
from vitai_core.utils.logger import get_logger
if TYPE_CHECKING:
from .optical_simulator_cfg import OpticalSimulatorCfg
from vitai_core.sensors.gf225.gf225 import GF225Sensor
class MLP(nn.Module):
"""Multi-layer perceptron for normal-to-color mapping in optical simulation."""
dropout_p = 0.05
def __init__(self, input_size = 5, output_size = 3, hidden_size = 32):
super().__init__()
self.fc1 = nn.Linear(input_size, hidden_size)
self.fc2 = nn.Linear(hidden_size, hidden_size)
self.fc3 = nn.Linear(hidden_size, hidden_size)
self.fc4 = nn.Linear(hidden_size, output_size)
self.drop = nn.Dropout(p=self.dropout_p)
def forward(self, x):
x = F.relu(self.fc1(x))
x = self.drop(x)
x = F.relu(self.fc2(x))
x = self.drop(x)
x = self.fc3(x)
x = self.drop(x)
x = self.fc4(x)
return x
class OpticalSimulator:
"""Optical simulation approach for GF225 sensor via camera depth data.
Uses MLP-based rendering to convert depth maps to tactile RGB images.
"""
cfg: OpticalSimulatorCfg
def __init__(self, cfg: OpticalSimulatorCfg, sensor: "GF225Sensor"):
"""Initialize the optical simulator.
Args:
cfg: Configuration for optical simulator
sensor: Parent GF225 sensor instance
"""
self.cfg = cfg
self.sensor = sensor
self._logger = get_logger("vitai.optical_sim", prim=str(sensor.cfg.prim_path))
def _initialize_impl(self):
"""Initialize optical simulator assets and models."""
if self.cfg.device is None:
self._device = self.sensor.device
else:
self._device = self.cfg.device
self._num_envs = self.sensor._num_envs
# Load optical sim assets
assets_dir = os.path.join(os.path.dirname(__file__), "optical_sim_utils")
self.bg_img = np.load(os.path.join(assets_dir, "vitai_bg.npy"))
self.bg_render = np.load(os.path.join(assets_dir, "init_bg_fots_vitai.npy"))
self.bg_depth = np.load(os.path.join(assets_dir, "ini_depth_sim_d0.npy"))
# Load optical MLP model
self.mlp = MLP().to(self._device)
weights_path = os.path.join(assets_dir, "mlp_n2c_vitai5.pth")
self.mlp.load_state_dict(torch.load(weights_path, map_location=self._device))
self.mlp.eval()
self._logger.debug(
f"Optical simulator initialized: device={self._device}, "
f"num_envs={self._num_envs}, "
f"tactile_res={self.cfg.tactile_img_res}"
)
# Helper method for optical simulation
def _padding(self,img):
if len(img.shape)==2:
return np.pad(img, ((1,1), (1,1)), "symmetric")
elif len(img.shape) == 3:
return np.pad(img, ((1,1), (1,1), (0,0)), "symmetric")
return img
def _generate_normals(self, height_map):
"""Generate surface normals from height map.
Args:
height_map: 2D array of height values
Returns:
Surface normals as (H, W, 3) array
"""
h, w = height_map.shape
top = height_map[0:h-2, 1:w-1]
bot = height_map[2:h, 1:w-1]
left = height_map[1:h-1, 0:w-2]
right = height_map[1:h-1, 2:w]
dzdx = (bot - top) / 2.0
dzdy = (right - left) / 2.0
direction = np.ones((h - 2, w - 2, 3))
direction[:, :, 0] = -dzdy
direction[:, :, 1] = dzdx
magnitude = np.sqrt(direction[:, :, 0] ** 2 + direction[:, :, 1] ** 2 + direction[:, :, 2] ** 2)
normal = direction / magnitude[:, :, np.newaxis]
normal = self._padding(normal)
normal = (normal + 1.0) * 0.5
return normal
def _smooth_height_map(self, height_map):
"""Smooth height map using multi-scale Gaussian blurring.
Args:
height_map: Raw depth/height map from camera
Returns:
Smoothed height map
"""
diff_depth = np.abs(height_map - self.bg_depth)
diff_depth[np.where(abs(diff_depth) < 6e-5)] = 0.0
diff_depth *= 1000.0
diff_depth /= (0.05107 * 2)
contact_mask = diff_depth > (np.max(diff_depth) * 0.4)
height_map = diff_depth
zq_back = height_map.copy()
kernel_sizes = [101, 51, 21, 11, 5]
for k in kernel_sizes:
height_map = cv2.GaussianBlur(height_map.astype(np.float32), (k, k), 0)
height_map[contact_mask] = zq_back[contact_mask]
height_map = cv2.GaussianBlur(height_map.astype(np.float32), (5, 5), 0)
return height_map
def _preproc_mlp(self, normal):
"""Preprocess normals for MLP input.
Args:
normal: Surface normal map (H, W, 3)
Returns:
Torch tensor ready for MLP input (N, 5) where N = H * W
"""
h, w = normal.shape[:2]
rows, cols = np.indices((h, w))
n_pixels = h * w
xy_coords = np.stack([rows.ravel(), cols.ravel()], axis=1)
nxyz = normal.reshape(n_pixels, 3)
# Normalize coordinates based on actual image size
norm_x = xy_coords[:, 0] / float(h)
norm_y = xy_coords[:, 1] / float(w)
input_data = np.column_stack((norm_x, norm_y, nxyz))
return torch.tensor(input_data, dtype=torch.float32, device=self._device)
# excute in main loop and output marker image
def optical_simulation(self):
"""Simulate tactile RGB image from depth data.
Returns:
Torch tensor of shape (num_envs, H, W, 3) containing RGB images
"""
if "camera_depth" not in self.sensor._data.output:
self._logger.warning("No camera depth data found, returning zero image")
return torch.zeros(
(self._num_envs, self.cfg.tactile_img_res[1], self.cfg.tactile_img_res[0], 3),
device=self._device,
dtype=torch.uint8,
)
# Process each environment
# For now, process first environment (TODO: extend to multi-env)
depth_tensor = self.sensor._data.output["camera_depth"][0]
if depth_tensor.dim() == 3:
depth_tensor = depth_tensor.squeeze(-1)
height_map = depth_tensor.cpu().numpy()
# Processing pipeline
height_map = self._smooth_height_map(height_map)
normal = self._generate_normals(height_map)
img_n = self._preproc_mlp(normal)
# MLP inference
with torch.no_grad():
sim_img_r = self.mlp(img_n)
# Post-process
sim_img_r = sim_img_r.cpu().numpy()
out_diff = (sim_img_r * 2 - 1) * 255
h, w = height_map.shape
sim_img = out_diff.reshape(h, w, 3).astype(np.float32)
sim_img = sim_img + self.bg_img
sim_img = np.clip(sim_img, 0, 255).astype(np.uint8)
return torch.tensor(sim_img, device=self._device).unsqueeze(0)
def compute_indentation_depth(self):
"""Compute indentation depth from optical simulation.
TODO: Implement indentation depth calculation
"""
return torch.zeros(self._num_envs, device=self._device)
def reset(self):
pass
def _set_debug_vis_impl(self, debug_vis: bool):
"""Setup debug visualization for optical simulation."""
if debug_vis and not hasattr(self, "_debug_windows"):
self._debug_windows = {}
self._debug_img_providers = {}
def _debug_vis_callback(self, event):
"""Debug visualization callback for showing tactile RGB output."""
if self.sensor._prim_view is None:
return
for i, prim in enumerate(self.sensor._prim_view.prims):
if "tactile_rgb" not in self.sensor.cfg.data_types:
continue
attr = prim.GetAttribute("debug_tactile_rgb")
if not attr.IsValid():
continue
show_img = attr.Get()
if show_img:
if str(i) not in self._debug_windows:
self._debug_windows[str(i)] = omni.ui.Window(
f"{self.sensor._prim_view.prim_paths[i]}/optical_sim",
width=self.cfg.tactile_img_res[0],
height=self.cfg.tactile_img_res[1],
)
self._debug_img_providers[str(i)] = omni.ui.ByteImageProvider()
# Get image data
img_data = self.sensor.data.output["tactile_rgb"][i]
if isinstance(img_data, torch.Tensor):
img_np = img_data.cpu().numpy()
else:
img_np = img_data
# Convert to RGBA
if img_np.shape[-1] == 3:
alpha = np.full((img_np.shape[0], img_np.shape[1], 1), 255, dtype=img_np.dtype)
img_np = np.concatenate([img_np, alpha], axis=-1)
# Update provider
self._debug_img_providers[str(i)].set_bytes_data(
img_np.flatten().tolist(),
[img_np.shape[1], img_np.shape[0]]
)
with self._debug_windows[str(i)].frame:
omni.ui.ImageWithProvider(self._debug_img_providers[str(i)])
elif str(i) in self._debug_windows:
self._debug_windows[str(i)].visible = False |