Demonstrators
Experience interactive prototypes and real-world demonstrations that bring our innovations to life
AnalogVision: Real-Time Handwritten Digit Recognition with Analog AI
Our analog AI demo uses the REDAC (Reconfigurable Discrete Analog Computer), a hybrid analog-digital computing platform provided by Anabrid, to recognize handwritten digits (0–9). We transform conventional artificial neurons into neural ordinary differential equations (neural ODEs) and implement the resulting neural network on the REDAC hardware. During the live demonstration, a user draws a digit on a trackpad or tablet, and the input is processed by the REDAC in real time to identify the digit being drawn. The demo showcases the potential of hybrid analog-digital computing for AI inference using neural ODEs.
Closed-Loop Robot Control via Neural Decoding
This demonstrator consists of a closed-loop system that links a robot to a biological or spiking neural network. The robot is equipped with a camera that captures its surroundings and encodes the track in neural activity. These signals stimulate a biological (or spiking) neural network grown on an array of electrodes, where activity is recorded, decoded, and classified into a driving command that controls the robot. As the neural network learns, the robot makes fewer and fewer errors over time, showcasing live learning using biological (or spiking) neural networks.
Privacy-Preserving Edge AI for Bowel Grasping
This demonstrator explores how advanced AI can support surgical robotic applications directly at the point of care. The main motivation is to keep sensitive patient data local, preserving privacy while avoiding dependence on costly cloud infrastructure. By bringing powerful AI to compact, affordable edge devices, the approach aims to make intelligent surgical systems more accessible and economically sustainable for the German healthcare system. The key contribution is a way to make complex AI models significantly smaller and more efficient without sacrificing the reliability required in demanding applications, finding the right balance between performance, cost, and resource consumption. The demonstrator illustrates the feasibility of this approach through a bowel grasping application, showing how efficient, privacy-preserving AI can be deployed in a realistic surgical scenario.