Festo and DarwinAI team up in a research project for federated learning in pick-and-place applications

Manufacturing

Festo

Festo is a part of the FLAIROP project (Federated Learning for Robot Picking), which employs state-of-the-art neural networks to enable and automate robotic tasks while preserving data security and privacy

 

 

Image source: Festo SE & Co. KG

About This Project

As part of the FLAIROP project, Festo will leverage federated learning and use our Explainable AI (XAI) for a number of potential industrial applications. The innovative two-year initiative combines deep learning, robotics, and data security expertise from Festo SE & Co. KG and the Karlsruhe Institute of Technology (KIT) with XAI and optimization specialization from DarwinAI and the University of Waterloo (UW).

This project is supported in part by advisory services and research and development funding support from the National Research Council of Canada Industrial Research Assistance Program (NRC IRAP) in Canada and the Federal Ministry of Economic Affairs and Energy (BMWi) in Germany.

FLAIROP is expected to enable new manufacturing possibilities and automation efficiencies within production lines that use robotics solutions. Through the project, Festo, DarwinAI, and other partners are creating meaningful data and performance analysis to show the significant benefits, practical scalability, and compounding network effects of an innovative new approach. Importantly, the federated learning technique ensures compliance with Europe’s General Data Protection Regulations (GDPR) data processing and storage requirements.

Difference Makers

  • Automation and manufacturing expertise turns vision into reality
  • Transparency into neural network decisions builds trust with operators
  • Versatility allows AI platform to become foundation of world-first federated learning application
  • Low compute needs enables real-time performance in edge robotics

About Festo

With an annual revenue of $3.8B USD, offices in 60+ countries, and a commitment to invest ~8% of revenue into research and development, Festo SE & Co. KG is one of the world’s largest and most innovative industrial and process automation companies.

Founded
1925
Employees
20,000+
Industries served
40+

Embedding Intelligence Increases Efficiency

The project advances a number of robot capabilities, including:

  • object detection
  • object grasp-point detection
  • automated learning data generation
  • predictive maintenance

By using our technology, Festo might be able to shrink the size and computational requirements of a deep neural network, making it possible to embed advanced capabilities into the relatively smaller microprocessing power available on edge devices like robotic arms. This intelligence enables Festo to recognize objects in real time, without needing to send image data for processing in a cloud—thereby increasing efficiency within manufacturing processes and avoiding security and privacy legislation challenges.

Human-Machine Collaboration

As robots gain greater dexterity through improved vision processing, their ability to assist humans with complex tasks—safely and efficiently—increases dramatically.

image description

Applying AI to Industrial Robotics

The FLAIROP project applies federated learning—augmented by our Explainable AI—across industrial robotics use cases and enabling components. In doing so, it advances the current state of the art in robotics and automation in a number of ways.

  1. Building and operationalizing state-of-the-art neural networks for object detection and grasp-point detection in edge robotics
  2. Developing modular and ensemble deep neural networks to improve functional accuracy while maintaining efficiency
  3. Creating a first-of-its-kind industrial robot federated learning infrastructure that performs multi-model optimization with multimodal input data
  4. Automating learning data generation for pick-and-place tasks

Image sources: Festo SE & Co. KG

Accelerating Innovation with Federated Learning

Federated learning is a machine learning technique that trains an algorithm across decentralized edge devices holding local data samples, without exchanging the samples themselves. Originally developed in the healthcare space, this approach allows neural networks on one system to benefit from knowledge acquired by other systems. For example, in a production line with multiple robot arms performing grasping or pick-and-place tasks, each benefits from the continuous learning of all the others. When combined with significantly smaller and highly efficient deep neural networks, federated learning promises to create strong efficiency gains in industrial automation use cases by embedding much higher intelligence and capabilities within edge devices than is possible today.

Crucially, federated learning shares knowledge, rather than data—that is, insights are shared within the network of connected devices, but the raw data (e.g., images) is not. This characteristic has important implications for security and privacy—concerns about which have slowed innovation in automation—because confidential information does not need to be transferred across a network or processed in a third-party cloud environment.

Interested in leveraging our technology for high-tech manufacturing?