Spaces:
Running on Zero
Running on Zero
File size: 10,910 Bytes
4417ac0 | 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 | import xml.etree.ElementTree as ET
import numpy as np
import torch
from . import torch_utils
class Joint:
def __init__(self, name, dof_dim, axis):
self._name = name
self._dof_dim = dof_dim
self._axis = axis
self._dof_idx = -1 # indicate the start index of dof in the whole dof vector, -1 for root or no dof joint
def set_dof_idx(self, dof_idx):
if self._dof_dim == 0:
raise ValueError('Joint {} has no dof'.format(self._name))
self._dof_idx = dof_idx
def dof_to_rot(self, dof):
# Input dof shape: [..., dof_dim]
# Output rot shape: [..., 4]
# Function: convert 1-dim or 3-dim dof to quaternion
rot_shape = list(dof.shape[:-1]) + [4]
ret_rot = torch.zeros(rot_shape, dtype=dof.dtype, device=dof.device)
if self._dof_dim == 0:
ret_rot[..., -1] = 1.0
elif self._dof_dim == 1:
axis = self._axis # shape: [3]
axis = torch.broadcast_to(axis, ret_rot[..., 0:3].shape)
ret_rot[:] = torch_utils.axis_angle_to_quat(axis, dof.squeeze(-1))
elif self._dof_dim == 3:
ret_rot[:] = torch_utils.exp_map_to_quat(dof)
return ret_rot
def rot_to_dof(self, rot):
# Input rot shape: [..., 4]
# Output dof shape: [..., dof_dim]
# Function: convert quaternion to 1-dim or 3-dim dof
dof_shape = list(rot.shape[:-1]) + [self._dof_dim]
ret_dof = torch.zeros(dof_shape, dtype=rot.dtype, device=rot.device)
if self._dof_dim == 1:
axis = self._axis
axis, angle = torch_utils.quat_to_axis_angle(rot)
dot_axis = torch.sum(axis * self._axis, dim=-1)
angle[dot_axis < 0] *= -1
ret_dof[:] = angle.unsqueeze(-1)
elif self._dof_dim == 3:
ret_dof[:] = torch_utils.quat_to_exp_map(rot)
return ret_dof
@property
def dof_dim(self):
return self._dof_dim
@property
def name(self):
return self._name
@property
def dof_idx(self):
return self._dof_idx
class KinematicsModel:
def __init__(self, file_path, device):
self._device = device
self._file_path = file_path
self._build_kinematics_model()
self._set_dof_indices()
def _build_kinematics_model(self):
self._body_names = []
self._parent_indices = []
self._local_translation = []
self._local_rotation = []
self._joints = []
self._dof_size = []
self._dof_upper_limits = []
self._dof_lower_limits = []
if self._file_path.endswith('.xml'):
self._parse_xml()
else:
raise NotImplementedError('File type not supported')
self._parent_indices = torch.tensor(self._parent_indices, dtype=torch.long, device=self._device)
self._local_translation = torch.tensor(np.array(self._local_translation), dtype=torch.float, device=self._device)
self._local_rotation = torch.tensor(np.array(self._local_rotation), dtype=torch.float, device=self._device)
self._num_dof = sum(self._dof_size)
self._dof_lower_limits = torch.tensor(self._dof_lower_limits, dtype=torch.float, device=self._device)
self._dof_upper_limits = torch.tensor(self._dof_upper_limits, dtype=torch.float, device=self._device)
if self._rot_unit == "degree":
self._dof_lower_limits = torch.deg2rad(self._dof_lower_limits)
self._dof_upper_limits = torch.deg2rad(self._dof_upper_limits)
def _parse_xml(self):
tree = ET.parse(self._file_path)
xml_doc_root = tree.getroot()
xml_world_body = xml_doc_root.find("worldbody")
assert xml_world_body is not None, "worldbody not found"
xml_body_root = xml_world_body.find("body")
assert xml_body_root is not None, "body not found"
compiler_data = xml_doc_root.find("compiler")
self._rot_unit = compiler_data.attrib.get("angle", "degree")
assert self._rot_unit in ["degree", "radian"], f"Invalid rotation unit: {self._rot_unit}"
def _add_xml_body(xml_node, parent_index, body_index):
body_name = xml_node.attrib.get("name")
pos_data = xml_node.attrib.get("pos", "0 0 0")
pos = np.fromstring(pos_data, dtype=float, sep=" ")
rot_data = xml_node.attrib.get("quat", "1 0 0 0")
rot = np.fromstring(rot_data, dtype=float, sep=" ")
rot_w = rot[..., 0].copy()
rot[..., 0:3] = rot[..., 1:]
rot[..., 3] = rot_w
if body_index == 0:
curr_joint = Joint(name=body_name, dof_dim=0, axis=None) # root
else:
curr_joints = xml_node.findall("joint")
num_joints = len(curr_joints)
if num_joints == 0:
curr_joint = Joint(name=body_name, dof_dim=0, axis=None)
elif num_joints == 1:
_axis = np.fromstring(curr_joints[0].attrib.get("axis"), dtype=float, sep=" ")
axis = torch.from_numpy(_axis).to(self._device)
curr_joint = Joint(name=body_name, dof_dim=1, axis=axis)
_dof_limits = np.fromstring(curr_joints[0].attrib.get("range"), dtype=float, sep=" ")
self._dof_lower_limits.append(_dof_limits[0])
self._dof_upper_limits.append(_dof_limits[1])
elif num_joints == 3:
axis = None
curr_joint = Joint(name=body_name, dof_dim=3, axis=axis)
for joint in curr_joints:
_dof_limits = np.fromstring(joint.attrib.get("range"), dtype=float, sep=" ")
self._dof_lower_limits.append(_dof_limits[0])
self._dof_upper_limits.append(_dof_limits[1])
else:
raise ValueError(f"Invalid number of joints: {num_joints} of body: {body_name}")
self._body_names.append(body_name)
self._parent_indices.append(parent_index)
self._local_rotation.append(rot)
self._local_translation.append(pos)
self._joints.append(curr_joint)
self._dof_size.append(curr_joint.dof_dim)
curr_index = body_index
body_index += 1
for child in xml_node.findall("body"):
body_index = _add_xml_body(child, curr_index, body_index)
return body_index
_add_xml_body(xml_body_root, -1, 0)
def _set_dof_indices(self):
curr_dof_idx = 0
for joint in self._joints:
if joint.dof_dim > 0:
joint.set_dof_idx(curr_dof_idx)
curr_dof_idx += joint.dof_dim
def dof_to_rot(self, dof):
rot_shape = list(dof.shape[:-1]) + [self.num_joint-1, 4]
joint_rot = torch.zeros(rot_shape, dtype=dof.dtype, device=dof.device)
for j in range(1, self.num_joint):
joint = self._joints[j]
if joint.dof_idx == -1:
joint_rot[..., j-1, -1] = 1.0
else:
joint_rot[..., j-1, :] = joint.dof_to_rot(dof[..., joint.dof_idx:joint.dof_idx+joint.dof_dim])
return joint_rot
def rot_to_dof(self, rot):
dof_shape = list(rot.shape[:-2]) + [self.num_dof]
dof = torch.zeros(dof_shape, dtype=rot.dtype, device=rot.device)
for j in range(1, self.num_joint):
joint = self._joints[j]
if joint.dof_dim == 0:
continue
joint_rot = rot[..., j-1, :]
dof[..., joint.dof_idx:joint.dof_idx+joint.dof_dim] = joint.rot_to_dof(joint_rot)
dof = torch.clamp(dof, self._dof_lower_limits, self._dof_upper_limits)
return dof
def convert_local_rot_to_global(self, local_rot):
# Input local_rot shape: [..., num_joint, 4] first row is root rotation
# local rotation shape: [num_joint-1, 4]
global_rot = torch.zeros_like(local_rot)
global_rot[..., 0, :] = local_rot[..., 0, :]
for j in range(1, self.num_joint):
parent_idx = self._parent_indices[j]
parent_rot = global_rot[..., parent_idx, :]
local_rot_j = local_rot[..., j, :]
global_rot[..., j, :] = torch_utils.quat_mul(parent_rot, local_rot_j)
return global_rot
def forward_kinematics(self, root_pos, root_rot, dof_pos, fitted_shape=None):
joint_rot = self.dof_to_rot(dof_pos)
body_pos = [None] * self.num_joint
body_rot = [None] * self.num_joint
body_pos[0] = root_pos
body_rot[0] = root_rot
for j in range(1, self.num_joint):
j_rot = joint_rot[..., j-1, :]
local_trans = self._local_translation[j] if fitted_shape is None else self._local_translation[j] * fitted_shape[j]
local_rot = self._local_rotation[j]
parent_idx = self._parent_indices[j]
parent_pos = body_pos[parent_idx]
parent_rot = body_rot[parent_idx]
local_trans_broadcast = torch.broadcast_to(local_trans, parent_pos.shape)
local_rot_broadcast = torch.broadcast_to(local_rot, parent_rot.shape)
world_trans = torch_utils.quat_rotate(parent_rot, local_trans_broadcast)
curr_pos = parent_pos + world_trans
curr_rot = torch_utils.quat_mul(local_rot_broadcast, j_rot)
curr_rot = torch_utils.quat_mul(parent_rot, curr_rot)
body_pos[j] = curr_pos
body_rot[j] = curr_rot
body_pos = torch.stack(body_pos, dim=-2)
body_rot = torch.stack(body_rot, dim=-2)
return body_pos, body_rot
def get_body_idx(self, body_name):
return self._body_names.index(body_name)
@property
def body_names(self):
return self._body_names
@property
def num_dof(self):
return self._num_dof
@property
def num_joint(self):
return len(self._joints)
@property
def joint_dof_idx(self):
dof_indices = []
for joint in self._joints:
dof_indices.append(joint.dof_idx)
return dof_indices
@property
def parent_indices(self):
return self._parent_indices
def get_parent_idx(self, idx):
return self._parent_indices[idx]
def get_dof_limits(self):
return self._dof_lower_limits, self._dof_upper_limits
|