> For the complete documentation index, see [llms.txt](https://captic-2.gitbook.io/docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://captic-2.gitbook.io/docs/depalletizing.md).

# Bag Depalletizer

AI vision that finds, classifies and localises bags on real-world pallets, so a robot can unload them.

The Captic Bag Depalletizer gives a robot the eyes to unload bags from pallets. A 3D camera looks at the pallet, the AI finds and classifies every bag, and the system sends the pose of the next bag to pick to your robot or PLC.

## Why bags are hard

Bags are one of the hardest things to depalletize automatically. They are deformable, so no two look the same. Pallets are messy: bags shift, sag and overlap, stacking patterns change, and lighting varies across the day. Rule-based vision systems and DIY vision toolkits struggle with this variability, which slows down projects and drives up costs.

## What it does

* **Finds every bag, on any pallet.** The AI adapts to deformable bags, changing light and pallet variability. You don't need to configure stacking patterns, layer heights or layouts in advance.
* **Estimates the pick pose.** For each bag, the system calculates where and at what angle the robot should pick it.
* **Classifies bags.** Built-in classification tells bag types apart and can recognise damaged or deformed bags, for example to route them for manual inspection.
* **Speaks your robot's language.** Poses are sent in the coordinate convention of your robot brand, with support for 35 brands. See [Coordinate Conventions](/docs/overview/hands-on/integrating-your-robot/integration-flow/robots/coordinate-conventions.md).
* **Keeps operators in control.** In Human Mode, an operator validates each result before it's sent, and the Captic team can review and override results remotely.

## At a glance

|               |                                                                                                                                          |
| ------------- | ---------------------------------------------------------------------------------------------------------------------------------------- |
| Cycle time    | Under 1.5 s per bag                                                                                                                      |
| Bag weights   | 5 to 50 kg                                                                                                                               |
| Bag materials | Paper, plastic, woven polypropylene and multi-layer bags                                                                                 |
| Robots        | 35 brands, natively or through a PLC or OPC UA. See [Robots](/docs/overview/hands-on/integrating-your-robot/integration-flow/robots.md). |
| Output        | Pick pose (X, Y, Z and rotation) and bag class                                                                                           |

## How it works

1. The 3D camera captures the pallet.
2. The AI detects the bags, classifies them and estimates each pick pose.
3. Detections outside the [Volume of Interest](/docs/depalletizing/getting-started.md#step-5-set-the-volume-of-interest), the area the robot can reach, are ignored.
4. The pose of the next bag is sent to the robot or PLC through the [Integration Flow](/docs/overview/hands-on/integrating-your-robot/integration-flow.md).
5. The robot picks the bag, and the cycle repeats.

## Setting up

Follow [Getting Started](/docs/depalletizing/getting-started.md) to go from delivery to the first pick.

## Key settings

* **Confidence score:** how confident the AI must be before a bag is picked.
* **Model selection:** the AI model for your bags and pallets.
* **Volume of Interest:** the 3D area in which bags are considered for picking.
* **Coordinate convention:** matches the output to your robot brand.

{% hint style="info" %}
Need something tailored to your line? Contact <support@captic.com>.
{% endhint %}