Open Source RoboticsRobots, software & learning

Collision Avoidance Metric

Software

Point-cloud evaluation tool measuring false-positive and missed collisions for configured gripper paths. [1]

Installation & first result →Official website

For
Developers comparing 3D sensing for robotic manipulation [1]
Source rights
Named licenses are documented by part. Review licenses
Unknowns
operating system, compute requirements not fully established. Review gaps

Source researched · Not tested by us
Sources checked · Research details

Licenses & related hardware

Requirements · Research details and dates

On this page

Overview

This tool compares a camera's point cloud with ground truth by asking where a configured gripper path would appear blocked or clear. It reports false-positive and false-negative collision fractions, making it useful for evaluating 3D sensing for manipulation. It is an offline evaluation metric, not a robot motion planner or a safety certification. [1] [4]

Is this for you?

Known limitations
  • Requires suitable ground-truth point clouds.
  • Results depend on configured path geometry, directions and tolerances.
  • The default scalar tolerance is not normalized into an iterable; use the explicit list in example.py.

Editorial assessment from cited documentation; no runtime testing.

[3] [4] · Editorial assessment

What you need

Languages / runtimes
  • Python

[3]

Dependencies
  • Python 3.10 reference environment
  • NumPy 1.25.1
  • Open3D 0.18.0
  • OpenCV 4.10.0.82

[1] [2]

Installation & first result

  • Prepare query and ground-truth point clouds in a shared coordinate frame and units.
  • Install requirements.txt.
  • Run python example.py QUERY.ply GROUND_TRUTH.ply.
  • Inspect FP and FN for the configured tolerances and directions.

Editorial assessment from cited documentation; no runtime testing.

[1] [3] [4] · Editorial assessment

Installation

Install requirements.txt in an isolated Python 3.10 environment from the pinned checkout.

Editorial assessment from cited documentation; no runtime testing.

[1] [2] · Editorial assessment

Hardware, models & data

Inputs and outputs

Two matching input types (PLY paths, NumPy arrays or Open3D point clouds) produce FP/FN fractions keyed by gripper tolerance.

[4]

Project notes

Supply comparable point clouds

The reviewed commit is 0ce92a51ab3eda61c663c873153dc1a67dbefb00. Upstream documents Python 3.10; its requirements file pins NumPy 1.25.1, Open3D 0.18.0 and OpenCV 4.10.0.82. Install that file in an isolated environment and run python example.py QUERY.ply GROUND_TRUTH.ply with your own measured and reference clouds. The example also demonstrates NumPy-array and Open3D inputs. [1] [2] [3] [6]

Both inputs must have the same Python type. The evaluator rotates the clouds and builds collision maps; it does not establish their registration for you. Prepare them in a shared frame and consistent units. Its gripper dimensions and tolerances are documented in millimeters. The example supplies a single direction, a 10 mm tolerance and an outlier threshold of five points, so its result describes that configuration rather than every possible approach path. [3] [4]

Use the example's explicit gripper_z_tolerances=[10.0]. The constructor defaults to the scalar 10.0, but only converts integers to a list; the compute method later iterates over the value. Source inspection therefore identifies a default-argument mismatch that the supplied example avoids. We have not executed it to reproduce the exception. [4]

The output is a dictionary containing FP and FN for each tolerance. Those values depend on the reference cloud and chosen path geometry; comparing cameras requires keeping that setup consistent. The implementation carries Apache-2.0. No numerical validation, camera capture or robot experiment was performed for this profile. [4] [7]

What we checked

Source researched only; no installation, training, benchmark or physical hardware test performed. No affiliate links or reused media.

Not yet established

  • Requirements: operating system, host computer
  • Version-specific hardware compatibility (ecosystem tags alone do not establish it)

Updates

Upstream last updated
Jun 21, 2024 [6]
Latest release
Not established
Profile updated
Oct 18, 2026
Sources last checked
Oct 18, 2026
Next review
Nov 18, 2026
Reviewed revision
commit snapshot; no stable GitHub release / 0ce92a51ab3eda61c663c873153dc1a67dbefb00 · software repository

Source files and licenses

Include an inspectable, distinct robotics camera-evaluation implementation; the example and metric define usable inputs, outputs and limitations independent of popularity.

LayerAvailability / license
softwareavailable · Apache-2.0
Repository materials only; external components retain their own terms. [7]
documentationavailable · Apache-2.0
Repository materials only; external components retain their own terms. [7]

No learned weights required for the scoped software workflow.

No training dataset required for the scoped software workflow.

References

Each source links to the exact revision we read. “Used for” shows where this page relies on it.

Correction guidance · Discussion is not enabled · Profile format 0.7.0-draft.5

Change history