Open Source RoboticsRobots, software & learning

AI for Industry Challenge Toolkit

Software

ROS 2 and Gazebo toolkit for developing and evaluating industrial cable-insertion policies. [1]

Installation & first result →Official website

For
Developers studying contact-rich robot manipulation [1]
Source rights
Named licenses are documented by part. Review licenses
Unknowns
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

The AIC toolkit provides a concrete cable-insertion exercise: a ROS 2 policy receives robot observations, sends motion commands and is evaluated in a Gazebo workcell. Its released source includes the trial engine, controller interfaces and scoring infrastructure, so it is useful for studying industrial manipulation even outside the original competition. This profile covers the local simulation workflow, not access to a physical competition cell. [1] [3] [5]

Is this for you?

Known limitations
  • WaveArm only tests integration and does not insert cables.
  • The quickstart pulls a moving latest container tag.
  • Simulation trial scores do not establish physical insertion success.

Editorial assessment from cited documentation; no runtime testing.

[2] [4] · Editorial assessment

What you need

Operating system
  • Ubuntu 24.04

[2]

Languages / runtimes
  • Python
  • C++

[3] [5]

Dependencies
  • ROS 2 Kilted
  • Docker, Distrobox and Pixi 0.67.2
  • NVIDIA Container Toolkit for GPU use

[2]

Host computer

Upstream baseline: 4–8 CPU cores, 32 GB+ RAM, RTX 2070+ or equivalent with 8 GB+ VRAM; CPU-only runs are documented as slower.

[2]

Installation & first result

  • Prepare Ubuntu, Docker, Distrobox and the pinned Pixi version.
  • Install the locked workspace and start the evaluation container.
  • Start the WaveArm policy within the documented 30-second connection window.
  • Inspect trial scores before implementing an insertion policy.

Editorial assessment from cited documentation; no runtime testing.

[2] · Editorial assessment

Installation

Use the pinned source with pixi install --locked and the documented aic_eval container workflow; verify the container version separately.

Editorial assessment from cited documentation; no runtime testing.

[2] · Editorial assessment

Hardware, models & data

Inputs and outputs

Camera, joint, force/torque and task observations feed a ROS 2 policy; Cartesian or joint commands drive simulated trials and produce scores.

[1] [3] [4]

Project notes

Establish the policy connection first

Use commit 6546127c1c15d5786f59c91400f5fc5ad6894448 with the upstream quickstart. It targets Ubuntu 24.04 and ROS 2 Kilted, with Docker, Distrobox and Pixi. The guide explicitly pins Pixi 0.67.2 because newer hash behavior conflicts with its container workflow. Its baseline is 4–8 CPU cores, at least 32 GB RAM and an RTX 2070-class GPU with at least 8 GB VRAM; CPU-only operation is described as slower. [2] [6]

Run pixi install --locked, start the evaluation container, then launch the documented aic_example_policies.ros.WaveArm policy. Start the environment first because it supplies the Zenoh router, but have the policy command ready: the guide says the engine times out after 30 seconds without the model node. The quickstart pulls aic_eval:latest; a source commit alone therefore does not pin the whole environment. Record the container digest when reproducing a run. [2]

WaveArm moves the arm back and forth; it is not an insertion solution. Its purpose is to expose the three-trial orchestration and scoring path. The expected result is trial progress and output under the configured results directory, not successful connector insertion. Replace it with a policy implementing the observation and motion-command interface once that connection works. [2] [3]

Scoring separates model validity, motion quality and insertion outcome. A valid model earns only the prerequisite point; insertion correctness and proximity account for much more of the result. This distinction prevents a running container from being mistaken for a solved manipulation task. [4]

The main repository license is Apache-2.0, with a documented BSD-3 exception for aic_utils/aic_isaac. External dependencies and robot assets retain their own terms. We inspected policy and scoring source but have not built the environment, run a trial or tested a physical robot. [5] [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

  • Version-specific hardware compatibility (ecosystem tags alone do not establish it)

Updates

Upstream last updated
Oct 1, 2026 [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 / 6546127c1c15d5786f59c91400f5fc5ad6894448 · software repository

Source files and licenses

Include the distinct ROS/Gazebo manipulation evaluation environment; source includes implemented trial validation and policy interfaces, with a bounded quickstart and explicit simulation-only scope.

LayerAvailability / license
softwareavailable · Apache-2.0
Main repository Apache-2.0; aic_utils/aic_isaac BSD-3 per README; external components retain their terms. [1] [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