Robobrain.vision from the Munich-based company Robominds took first place in the robotics category. It is an AI-based vision platform that gives robots a sense of sight. The solution consists of a hardware package with an industrial PC and camera and a software architecture for executing the most complex image processing algorithms. As the central element of the solution, the Robobrain industrial PC provides the necessary computing power for industrial robot applications. The robust and secure platform architecture can be used flexibly thanks to its modular design. Examples of this are the robobrain.control modules for controlling robots and other automation components or robobrain.fleet for managing the entire robot fleet. This enables modular expansion of the system right down to the process control level.

The camera system combines 2D and 3D data for optimum part recognition results. Thanks to its compact design, the camera can be mounted directly on the robot flange in addition to conventional static mounting. Its low weight reduces the possible robot payload only minimally. Attachment to the robot flange increases the flexibility of the application and enables even deeper process integration. The flexible detection algorithm for optimum gripping point detection makes pick-and-place applications possible without teach-in. A wide variety of object types are already supported, from sheet metal parts to packaging and consumer goods. Gripping point detection is independent of the object, which makes it possible to grip new components without teach-in time or CAD models. This allows the system to react independently to small changes in the products or the environment. For particularly complex cases, it is also possible to increase the performance of the system by re-learning process data.

The manufacturer has designed the user interface and industrial components of this robot AI to be extremely user-friendly. The system can be commissioned without specific knowledge of computer vision using just a few parameters, which reduces maintenance and commissioning costs. The settings can be flexibly parameterized directly in the process via a web interface and adapted to the specific application. The modular design of the software means that different grippers can be used. With the help of so-called "AI skills", various functionalities can be integrated and executed for the respective process step. In addition to vacuum grippers, parallel grippers with integrated collision avoidance are also supported during gripping. The user can also integrate special grippers - for handling boxes, for example.

Multi-compatible interfaces, such as OPC UA, enable the robot to grip the object directly - both via the manufacturer's integrated solution and via the Robominds robotic controller robobrain.control. By combining AI with state-of-the-art image processing algorithms, the gripping point is calculated in less than 500 milliseconds. These algorithms are based on a coordinate-independent mode of operation so that superimposed and randomized objects can be gripped reliably. pb