Federated Learning for Robot Picking
The research project FLAIROP (Federated Learning for Robot Picking) develops a distributed learning approach for pick-and-place robots to robustly recognize and grasp known as well as unknown objects.
The goal is to provide current AI solutions with more data while respecting privacy regulations. There should be no exchange of training data (e.g. images, grasping points, etc.). In this context, the international research project investigates how training data from multiple plants or even companies can be used to increase recognition performance compared to single robot training.
The project focuses on automated generation of learning data required for the grasp detection and federated learning algorithms. A cloud structure will be used to provide this data and algorithms for distributed learning. The AI models are designed to be as efficient as possible both to run locally in each side and globally on the central cloud server. This represents the next stage of development for the simple handling of autonomously acting systems in the context of Industry 4.0. In particular, AI-based data processing methods are combined with the field of industrial robotics.
So far, federated learning has been used predominantly in the medical sector for image analysis (protection of patient data). Transferring the technology to the increasingly interconnected Industry 4.0 / Logistics 4.0 offers strong potential for the robust use of artificial intelligence and development of new, more powerful algorithms - while maintaining data protection guidelines.
• Festo SE & Co. KG (Project leader, Germany)
• Karlsruhe Institute für Technology (KIT) (Germany)
• University of Waterloo (Canada)
• Darwin AI (Canada)
Festo SE & Co. KG
Project duration: Feb. 2021 - Jul. 2023
Total costs: 1.4 million euros
Total funding: 1.0 million euros
- Recommend this page:
- Print view