Implementation:Haosulab ManiSkill GeometryUtils

Knowledge Sources	Haosulab_ManiSkill
Domains	Robotics, Simulation, 3D Geometry
Last Updated	2026-02-15 08:00 GMT

Overview

Concrete tool for 3D geometry operations including sampling, rotation, bounding box computation, and coordinate transformations for physics objects.

Description

The geometry.py module provides a collection of geometry utility functions used across ManiSkill for spatial reasoning, bounding volume computation, and coordinate transformations.

Sampling:

sample_on_unit_sphere(rng) -- Uniformly samples a point on the unit sphere using Gaussian sampling.
sample_on_unit_circle(rng) -- Uniformly samples a point on the unit circle.

Rotation/Angle:

rotation_between_vec(a, b) -- Computes the scipy Rotation from vector a to vector b.
angle_between_vec(a, b) -- Computes the angle in radians between two vectors.
angle_distance(q0, q1) -- Angular distance between two sapien.Pose quaternions (normalized to [0, 1]).
rotate_2d_vec_by_angle(vec, theta) -- Rotates a 2D vector by angle theta.
wxyz_to_xyzw() / xyzw_to_wxyz() -- Quaternion convention converters.

Bounding volumes:

get_axis_aligned_bbox_for_articulation(art) -- AABB for an entire articulation.
get_axis_aligned_bbox_for_actor(actor) -- AABB for a single actor entity.
get_local_axis_aligned_bbox_for_link(link) -- AABB in link-local frame.
get_local_aabc_for_actor(actor) -- Axis-aligned bounding cylinder in local frame.
get_oriented_bounding_box_for_2d_points(points_2d) -- OBB via PCA for 2D point sets.

Transformations:

transform_points(H, pts) -- Batched transformation of 3D points by 4x4 matrices.
invert_transform(H) -- Invert a batch of 4x4 transformation matrices.
rotate_vector(v, q) -- Rotate a vector by a quaternion (pulled from SAPIEN).

Usage

Import specific functions for spatial reasoning in task reward computation, object manipulation planning, and scene understanding.

Code Reference

Source Location

Repository: Haosulab_ManiSkill
File: mani_skill/utils/geometry/geometry.py

Signature

def sample_on_unit_sphere(rng) -> np.ndarray: ...
def rotation_between_vec(a, b) -> Rotation: ...
def angle_between_vec(a, b) -> float: ...
def get_axis_aligned_bbox_for_articulation(art: PhysxArticulation) -> tuple[np.ndarray, np.ndarray]: ...
def get_axis_aligned_bbox_for_actor(actor: sapien.Entity) -> tuple[np.ndarray, np.ndarray]: ...
def transform_points(H: torch.Tensor, pts: torch.Tensor) -> torch.Tensor: ...
def invert_transform(H: np.ndarray) -> np.ndarray: ...
def get_oriented_bounding_box_for_2d_points(points_2d: np.ndarray, resolution=0.0) -> dict: ...

Import

from mani_skill.utils.geometry.geometry import (
    sample_on_unit_sphere,
    get_axis_aligned_bbox_for_articulation,
    transform_points,
)

I/O Contract

Inputs

Name	Type	Required	Description
rng	np.random.RandomState	Yes	Random number generator (for sampling functions)
art	PhysxArticulation	Yes	Articulation for AABB computation
H	torch.Tensor (N, 4, 4)	Yes	Batch of transformation matrices
pts	torch.Tensor (N, 3)	Yes	Points to transform

Outputs

Name	Type	Description
mins, maxs	np.ndarray (3,)	AABB minimum and maximum corners
transformed	torch.Tensor (N, 3)	Transformed points

Usage Examples

Basic Usage

import numpy as np
from mani_skill.utils.geometry.geometry import (
    sample_on_unit_sphere,
    get_axis_aligned_bbox_for_actor,
    angle_between_vec,
)

# Sample a random direction
rng = np.random.RandomState(42)
direction = sample_on_unit_sphere(rng)

# Compute angle between two vectors
angle = angle_between_vec(np.array([1, 0, 0]), np.array([0, 1, 0]))
# angle ~= pi/2

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment