Artificial Intelligence for Digitizing Industry (AI4DI)
(AI4DI Grant Agreement No. 826060)
Total Budget: ~30M Eiro
Project Coordinator: Reiner John, Infineon Technologies AG
Countries (12): Germany, Austria, Czech Republic, Italy, Norway, Latvia, Taiwan, Belgium, Lithuania, France, Greece, Finland
AI4DI aims to strengthen and expand AI usage in European industry digitization process. Enabling of performance, industry and humanity by AI for digitising industry is the key to push the AI revolution in Europe and step into the digital age. Potential users of AI are not sufficiently supported to facilitate the integration of AI into their applications. Existing services providing state of the art machine learning (ML) and artificial intelligence solutions are currently available in the cloud. AI4DI project aim is to transfer machine learning and AI from the cloud to the edge in manufacturing, mobility and robotics. To achieve these targets AI4DI will connect factories, processes, and devices within digitised industry by utilizing ML and AI for human machine collaboration, change detection, and detection of abnormalities.
In this project, EDI leads the Sub Supply chain 3.2 “Smart Robot” that consists of different cognitive sensor modules that are the basis for three use-cases which can supply different industries by using AI-based solutions. EDI develops cognitive sensing to perceive and understand dynamically changing environment for randomly dropped object detection, pose estimation and pick-up by an industrial robot – AI is used to analyze the data of a stereo vision system, which incorporates processing on the edge (FPGA based SoC). The training process is accelerated in collaboration with VIF by introducing semi-automatic labelling approaches and generation of synthetic life-like labelled data, furthermore standardized extensive automated testing and validation of the reliable localization and classification of objects is done by VIF.
Within SC3.2 TUD in collaboration with EDI enables robots of any size to “feel” – a reflectometric sensor is proposed featuring miniaturized electronics and single-channel measurement also for large scale areas. The sensor is based on the electric time domain reflectometry method that enables the spatial resolved measurement of impedance distribution of a two-core electrical transmission line (ETL) based on reflection of an incident electrical impulse. The transmission line for the robot sensitive skin is supposed to be composed of copper conductors and a soft honeycomb as core structure. Neural networks are used to map the sensor signal to the apparent deformation which is related to the touch input or collision event.
IMEC in collaboration with EDI develops technologies to detect the position between the robot and surrounding objects, and to allow unobtrusive/contactless interaction between the operator and the robot. This includes detection and classification of hand gestures based on mm-wave radar data and machine learning during training and inference. AI process includes raw data analysis, segmentation, feature extraction (micro-doppler) and classification.
The developments are planned to be demonstrated using EDI industrial robot infrastructure