The Cambrian vision system was developed for cost-efficient, fast and robust binpicking, pick and place, machine feeding, feature location, assembly, supermarket, warehouse logistics and other automation and manufacturing related tasks.

The system is based on AI. To make accurate predictions the AI model uses just the image data from the stereo cameras that can be mounted on the robot or above the workplace.

The vision system doesn’t need any external light such as structured or infrared light. The system is very robust for any changes of external light conditions, even if it gets very bright or dark. The AI model calculates the exact part locations and suitable grasp points within 0.2 seconds.

The system can pick a wide range of different parts, including typical plastic and metal parts, it also works even if they are transparent, thin or very shiny.

The standard camera module can pick parts from sub millimeter range to parts fitting in a 20 cm cube. For larger parts, we provide a slightly larger camera module with a different camera lens.

Setting up is quick and easy. The user just has to define the grasp points and the AI will train without further user input.
Attachment plates to most UR ecosystem grippers are available and included.

FEATURES

• Bin pick wide range of parts robustly

• State of the art AI recognition

• Fast prediction , less than 0.3s

• Sub mm level accuracy

• Quick to setup

• Handles collisions with the box

• Doesn't use any external light

• Robust for various external light conditions

HOW DOES IT WORK?

The Cambrian system leverages the power of state of the art neural networks to recognise a wide range of different parts, such as shiny, reflective, transparent and black parts that are troublesome or impossible for traditional vision systems.

The AI model that is responsible for recognising the parts is trained with simulated data and is exposed to a wide range of conditions, hence it learns to see even the most difficult parts.

Quick description of operation:
1. Load the CAD model you want to pick into Cambrian software
2. Define your ideal grasp points on the CAD model ( ~10min or less) -> send to Cambrian.
3. Receive AI model to the local system that can robustly and accurately locate the part.
4. On UR program, insert Cambrian locate and approach nodes and start picking.

The system doesn't need to train with the real robot or the real part.