adithyamurali's picture
Add visualization scripts
44e46d6
# Copyright (c) 2025, NVIDIA CORPORATION. All rights reserved.
#
# NVIDIA CORPORATION and its licensors retain all intellectual property
# and proprietary rights in and to this software, related documentation
# and any modifications thereto. Any use, reproduction, disclosure or
# distribution of this software and related documentation without an express
# license agreement from NVIDIA CORPORATION is strictly prohibited.
#
'''
Utility functions for visualization using meshcat.
Installation:
pip install trimesh==4.5.3 objaverse==0.1.7 meshcat==0.0.12 webdataset==0.2.111
NOTE: Start meshcat server (in a different terminal) before running this script:
meshcat-server
'''
import numpy as np
import meshcat
import meshcat.geometry as g
import meshcat.transformations as mtf
import trimesh
import trimesh.transformations as tra
from typing import List, Optional, Tuple, Union, Any
control_points_franka = np.array([
[ 0.05268743, -0.00005996, 0.05900000],
[-0.05268743, 0.00005996, 0.05900000],
[ 0.05268743, -0.00005996, 0.10527314],
[-0.05268743, 0.00005996, 0.10527314]
])
control_points_robotiq2f140 = np.array([
[ 0.06801729, -0, 0.0975],
[-0.06801729, 0, 0.0975],
[ 0.06801729, -0, 0.1950],
[-0.06801729, 0, 0.1950]
])
control_points_suction = np.array([
[ 0, 0, -0.10],
[ 0, 0, -0.05],
[ 0, 0, 0],
])
control_points_data = {
"franka": control_points_franka,
"robotiq2f140": control_points_robotiq2f140,
"suction": control_points_suction,
}
def get_gripper_control_points(gripper_name: str = 'franka') -> np.ndarray:
"""
Get the control points for a specific gripper.
Args:
gripper_name (str): Name of the gripper ("franka", "robotiq2f140", "suction")
Returns:
np.ndarray: Array of control points for the specified gripper
Raises:
NotImplementedError: If the specified gripper is not implemented
"""
if gripper_name in control_points_data:
return control_points_data[gripper_name]
else:
raise NotImplementedError(f"Gripper {gripper_name} is not implemented.")
return control_points
def get_gripper_depth(gripper_name: str) -> float:
"""
Get the depth parameter for a specific gripper type.
Args:
gripper_name (str): Name of the gripper ("franka", "robotiq2f140", "suction")
Returns:
float: Depth parameter for the specified gripper
Raises:
NotImplementedError: If the specified gripper is not implemented
"""
# TODO: Use register module. Don't have this if-else name lookup
pts, d = None, None
if gripper_name in ["franka", "robotiq2f140"]:
pts = get_gripper_control_points(gripper_name)
elif gripper_name == "suction":
return 0.069
else:
raise NotImplementedError(f"Control points for gripper {gripper_name} not implemented!")
d = pts[-1][-1] if pts is not None else d
return d
def get_gripper_offset(gripper_name: str) -> np.ndarray:
"""
Get the offset transform for a specific gripper type.
Args:
gripper_name (str): Name of the gripper
Returns:
np.ndarray: 4x4 homogeneous transformation matrix representing the gripper offset
"""
return np.eye(4)
def load_visualize_control_points_suction() -> np.ndarray:
"""
Load visualization control points specific to the suction gripper.
Returns:
np.ndarray: Array of control points for suction gripper visualization
"""
h = 0
pts = [
[0.0, 0],
]
pts = [generate_circle_points(c, radius=0.005) for c in pts]
pts = np.stack(pts)
ptsz = h * np.ones([pts.shape[0], pts.shape[1], 1])
pts = np.concatenate([pts, ptsz], axis=2)
return pts
def generate_circle_points(center: List[float], radius: float = 0.007, N: int = 30) -> np.ndarray:
"""
Generate points forming a circle in 2D space.
Args:
center (List[float]): Center coordinates [x, y] of the circle
radius (float): Radius of the circle
N (int): Number of points to generate around the circle
Returns:
np.ndarray: Array of shape (N, 2) containing the circle points
"""
angles = np.linspace(0, 2 * np.pi, N, endpoint=False)
x_points = center[0] + radius * np.cos(angles)
y_points = center[1] + radius * np.sin(angles)
points = np.stack((x_points, y_points), axis=1)
return points
def get_gripper_visualization_control_points(gripper_name: str = 'franka') -> List[np.ndarray]:
"""
Get control points for visualizing a specific gripper type.
Args:
gripper_name (str): Name of the gripper ("franka", "robotiq2f140", "suction")
Returns:
List[np.ndarray]: List of control point arrays for gripper visualization
"""
if gripper_name == "suction":
control_points = load_visualize_control_points_suction()
offset = get_gripper_offset('suction')
ctrl_pts = [tra.transform_points(cpt, offset) for cpt in control_points]
d = get_gripper_depth(gripper_name)
line_pts = np.array([[0,0,0], [0,0,d]])
line_pts = np.expand_dims(line_pts, 0)
line_pts = [tra.transform_points(cpt, offset) for cpt in line_pts]
line_pts = line_pts[0]
ctrl_pts.append(line_pts)
return ctrl_pts
else:
control_points = get_gripper_control_points(gripper_name)
mid_point = (control_points[0] + control_points[1]) / 2
control_points = [
control_points[-2], control_points[0], mid_point,
[0, 0, 0], mid_point, control_points[1], control_points[-1]
]
return [control_points, ]
def get_color_from_score(labels: Union[float, np.ndarray], use_255_scale: bool = False) -> np.ndarray:
"""
Convert score labels to RGB colors for visualization.
Args:
labels (Union[float, np.ndarray]): Score values between 0 and 1
use_255_scale (bool): If True, output colors in [0-255] range, else [0-1]
Returns:
np.ndarray: RGB colors corresponding to the input scores
"""
scale = 255.0 if use_255_scale else 1.0
if type(labels) in [np.float32, float]:
return scale * np.array([1 - labels, labels, 0])
else:
scale = 255.0 if use_255_scale else 1.0
score = scale * np.stack(
[np.ones(labels.shape[0]) - labels, labels, np.zeros(labels.shape[0])],
axis=1,
)
return score.astype(np.int)
def trimesh_to_meshcat_geometry(mesh: trimesh.Trimesh) -> g.TriangularMeshGeometry:
"""
Convert a trimesh mesh to meshcat geometry format.
Args:
mesh (trimesh.Trimesh): Input mesh in trimesh format
Returns:
g.TriangularMeshGeometry: Mesh in meshcat geometry format
"""
return meshcat.geometry.TriangularMeshGeometry(mesh.vertices, mesh.faces)
def visualize_mesh(
vis: meshcat.Visualizer,
name: str,
mesh: trimesh.Trimesh,
color: Optional[List[int]] = None,
transform: Optional[np.ndarray] = None
) -> None:
"""
Visualize a mesh in meshcat with optional color and transform.
Args:
vis (meshcat.Visualizer): Meshcat visualizer instance
name (str): Name/path for the mesh in the visualizer scene
mesh (trimesh.Trimesh): Mesh to visualize
color (Optional[List[int]]): RGB color values [0-255]. Random if None
transform (Optional[np.ndarray]): 4x4 homogeneous transform matrix
"""
if vis is None:
return
if color is None:
color = np.random.randint(low=0, high=256, size=3)
mesh_vis = trimesh_to_meshcat_geometry(mesh)
color_hex = rgb2hex(tuple(color))
material = meshcat.geometry.MeshPhongMaterial(color=color_hex)
vis[name].set_object(mesh_vis, material)
if transform is not None:
vis[name].set_transform(transform)
def rgb2hex(rgb: Tuple[int, int, int]) -> str:
"""
Convert RGB color values to hexadecimal string.
Args:
rgb (Tuple[int, int, int]): RGB color values (0-255)
Returns:
str: Hexadecimal color string (format: "0xRRGGBB")
"""
return "0x%02x%02x%02x" % (rgb)
def create_visualizer(clear: bool = True) -> meshcat.Visualizer:
"""
Create a meshcat visualizer instance.
Args:
clear (bool): If True, clear the visualizer scene upon creation first
Returns:
meshcat.Visualizer: Initialized meshcat visualizer
"""
print(
"Waiting for meshcat server... have you started a server? Run `meshcat-server` to start a server"
)
vis = meshcat.Visualizer(zmq_url="tcp://127.0.0.1:6000")
if clear:
vis.delete()
return vis
def visualize_pointcloud(
vis: meshcat.Visualizer,
name: str,
pc: np.ndarray,
color: Optional[Union[List[int], np.ndarray]] = None,
transform: Optional[np.ndarray] = None,
**kwargs: Any
) -> None:
"""
Args:
vis: meshcat visualizer object
name: str
pc: Nx3 or HxWx3
color: (optional) same shape as pc[0 - 255] scale or just rgb tuple
transform: (optional) 4x4 homogeneous transform
"""
if vis is None:
return
if pc.ndim == 3:
pc = pc.reshape(-1, pc.shape[-1])
if color is not None:
if isinstance(color, list):
color = np.array(color)
color = np.array(color)
# Resize the color np array if needed.
if color.ndim == 3:
color = color.reshape(-1, color.shape[-1])
if color.ndim == 1:
color = np.ones_like(pc) * np.array(color)
# Divide it by 255 to make sure the range is between 0 and 1,
color = color.astype(np.float32) / 255
else:
color = np.ones_like(pc)
vis[name].set_object(
meshcat.geometry.PointCloud(position=pc.T, color=color.T, **kwargs)
)
if transform is not None:
vis[name].set_transform(transform)
def load_visualization_gripper_points(gripper_name: str = "franka") -> List[np.ndarray]:
"""
Load control points for gripper visualization.
Args:
gripper_name (str): Name of the gripper to visualize
Returns:
List[np.ndarray]: List of control point arrays, each of shape [4, N]
where N is the number of points for that segment
"""
ctrl_points = []
for ctrl_pts in get_gripper_visualization_control_points(gripper_name):
ctrl_pts = np.array(ctrl_pts, dtype=np.float32)
ctrl_pts = np.hstack([ctrl_pts, np.ones([len(ctrl_pts),1])])
ctrl_pts = ctrl_pts.T
ctrl_points.append(ctrl_pts)
return ctrl_points
def visualize_grasp(
vis: meshcat.Visualizer,
name: str,
transform: np.ndarray,
color: List[int] = [255, 0, 0],
gripper_name: str = "franka",
**kwargs: Any
) -> None:
"""
Visualize a gripper grasp pose in meshcat.
Args:
vis (meshcat.Visualizer): Meshcat visualizer instance
name (str): Name/path for the grasp in the visualizer scene
transform (np.ndarray): 4x4 homogeneous transform matrix for the grasp pose
color (List[int]): RGB color values [0-255] for the grasp visualization
gripper_name (str): Name of the gripper to visualize
**kwargs: Additional arguments passed to MeshBasicMaterial
"""
if vis is None:
return
grasp_vertices = load_visualization_gripper_points(gripper_name)
for i, grasp_vertex in enumerate(grasp_vertices):
vis[name + f"/{i}"].set_object(
g.Line(
g.PointsGeometry(grasp_vertex),
g.MeshBasicMaterial(color=rgb2hex(tuple(color)), **kwargs),
)
)
vis[name].set_transform(transform.astype(np.float64))