NOTE! This site uses cookies and similar technologies.

If you not change browser settings, you agree to it. Learn more

I understand

Learn more about cookies at :

December 7, 2021 | Robust blockchains for the financial markets

illustration Florence Pillet corrigéeClearmatics Technologies develops blockchain-based solutions for the financial markets. The company turned to CEA-List for support developing solutions that meet the unique needs of financial transactions.

A blockchain is a distributed data ledger to which all participants have access. Different protocols can be used to make sure that the ledger is coherent. Consensus protocols are currently the most effective, and the only ones capable of immediately recording transactions without the usual latency of blockchain-based systems. However, consensus protocols are not reliable in instances where more than one-third of network participants are malicious or inactive.

Clearmatics Technologies asked CEA-List to develop algorithms to limit the risk of exceeding this critical threshold for its blockchain for the financial markets. An algorithm inspired by CEA-List distributed systems research was developed to provide proof of breach of protocol and identify the parties responsible. It ensures early detection of abnormal behavior and—crucially—proof, which is a good deterrent against bad actors. The algorithm compares the exchanges between network participants, such as the kind of reply to a message, with what would be expected. A second algorithm incentivizes participation by rewarding active members based on the number of messages exchanged.

The specifications for both software applications were transferred to Clearmatics Technologies for further development and implementation in the company's systems. These next steps will give CEA-List engineers a test case that is very similar to a real-world use case so that their approach can be validated, and further development work completed.

Read article at



November 25, 2021 | Sim2Real Challenge: CEA-List launches an Object Detection and Pose-Estimation Challenge

Metrics 1At the heart of the European H2020 METRICS project coordinated by LNE and sixteen other partners, the competition, led by CEA-List, targets agile production. Its aim is to detect and recognise different components of a system and estimate their position and orientation.

Object detection and pose estimation are fundamental steps for industrial robotic systems. Training algorithms with CAD models is therefore essential for the agility of industrial systems. The challenge proposed by Sim2Real Challenge is to evaluate the performance of detection, classification and pose estimation algorithms while having limited access to real data during the training phase.

Participants will have to rely on CAD models of the parts to develop and train their algorithms, while the test phase will be carried out using real images.

This challenge is a step in the ADAPT (ADvanced Agile ProducTion) competition, which aims to implement dexterous manipulation of industrial components, while facilitating the operator's task via intuitive and multimodal interfaces. The challenge is to validate the performance and maturity of machine learning and robotics techniques, in order to integrate them into the assembly processes of industrial systems.


Join the Object Detection Challenge

Join the Installation Estimation Challenge



Training and Validation Phase  28/10/2021 – 25/01/2022
Test Phase  26/01/2022 – 08/02/2022
Announcing of Results  10-20/02/2022


More information:

The H2020 METRICS project (2020-2024), dedicated to metrological evaluation and testing campaigns of robotic and artificial intelligence systems (in the field and in cascade), is structured in 4 thematic challenges: Health (HEART-MET), Inspection and Maintenance (RAMI), Agriculture and Agri-Food (ACRE), Agile Production (ADAPT).



November 22, 2021 | Bringing artificial intelligence in machine vision to optimise bridge maintenance

Visual inspection is the most common technique used for the maintenance of civil engineering structures like bridges. While it does offer the advantage of being non-destructive, visual inspection also depends on inspectors who interpret photographs and write up inspection reports—which then become the main source of information used by the operators of these structures to optimise their maintenance plans. SOCOTEC leads the project called SOFIA in partnership with CEA-List to develop a groundbreaking inspection tool that will bring the power of artificial intelligence to machine vision. Together in a consortium with CEA-List, SOFIA won a French call launched by CEREMA on “connected bridges”.

ponts connectés Sofia socotec en

Bridges are critical components of road networks that play a strategic role in our economy and society1. France has around 200,000 bridges2. According to a 2019 report by the French Senate’s Infrastructure Development and Sustainability Committee, the nation’s bridges are getting old3. To keep people safe and roads open, the maintenance of these aging bridges must be optimised.

In December 2020, the French government issued a call for projects to develop solutions for connected bridges1. Managed by CEREMA, a government agency that advises on infrastructure development and sustainability issues, the call for projects is financed by France’s economic recovery plan. The objective is to come up with new tools and methods for safer, more effective, and cheaper bridge maintenance.

The SOFIA project, led by SOCOTEC, an independent trusted third party that provides testing, inspection, and certification services to the construction and civil engineering industries, was selected by CEREMA. The project will integrate SOCOTEC’s inspection tools with artificial intelligence. Thanks to its strong expertise in machine learning and artificial intelligence for vision, CEA-List was the natural partner for SOCOTEC.

The main innovation in this project lies in the algorithms. No additional instrumentation will be needed, and SOCOTEC can continue to use its common inspection techniques. A powerful vision system based on advanced detection and recognition algorithms will automatically detect and characterize defects photographed by inspectors on site. More than 100,000 images from SOCOTEC's database will be used to train the system. The technology developed will guide inspectors as they enter data into their inspection reports and calculate a rating for the condition of the structure being inspected.

SOCOTEC Infrastructure has a database with more than 100,000 photographs from 2,500 inspection reports characterizing 250 types of defects. This database will be used as training data for the machine vision models. After this initial training, the AI-based models will continue to learn from subsequent inspections.

Data from past inspections will be used to produce more standardized inspection reports. The experience gained with each inspection will make future assessments of the condition of bridges more reliable—the ultimate goal. Inspectors will be able to do their jobs faster and more easily while producing more reliable inspection reports. And they will not need to change the way they carry out their visual inspections.

At the end of the project, the new features will be integrated into SOCOTEC’s inspection software. A decision assistance module will also be included to analyse and summarise the information entered and to generate the inspection report. As AI and computer vision will make it easier to flag defects that require further analysis, SOCOTEC’s inspections will gain in accuracy.

Looking ahead, SOCOTEC will reap additional benefits from this new solution by using AI to mine the massive amounts of inspection data collected for predictive maintenance insights.

Contact: This email address is being protected from spambots. You need JavaScript enabled to view it. and This email address is being protected from spambots. You need JavaScript enabled to view it.

1CEREMA is a government agency that advises on infrastructure development and sustainability issues (

2Civil engineering structures in France: French Society of Engineers and Scientists (IESF) Construction and Civil Engineering Committee Report on Monitoring and Maintenance, December 2018 (

3French Senate Development and Sustainability Committee Report, June 2019 (



19 novembre 2021 | Sport Quantum invente la cible électronique

sport quantum 250Après les écrans tactiles, voici les écrans « impactiles » ! Conçus par la start-up Sport Quantum pour le tir sportif, ils localisent le point d’impact d’un projectile à 100 microns près, soit l’épaisseur d’un cheveu.

L’assemblage comprend une plaque transparente en plastique rigide sous laquelle sont placés quatre capteurs piézoélectriques qui mesurent l’onde de choc générée par l’impact. Derrière, un écran d’ordinateur affiche la cible, des méthodes d’entraînement ou des jeux. « C’est une rupture technologique », annonce Jean-Marc Alexandre, co-fondateur de la start-up avec Robert Boden, puis Jean Sreng. Plus besoin de cibles en carton qu’il faut changer entre les tirs et ramener pour être lues. Les résultats sont visibles en temps réel sur une tablette, grâce à une connexion sans fil et à des algorithmes de traitement du signal. « Je pressentais le fort potentiel des capteurs piézoélectriques sur lesquels nous travaillions au CEA-List, se souvient Jean-Marc Alexandre. Mais c’est Robert Boden, adepte du tir sportif, qui a lancé en 2016 l’idée décisive de la cible électronique ».

Un franc succès auprès des athlètes

Depuis, plus de 500 dispositifs ont été installés en France et 300 à l’international. « Plusieurs athlètes internationaux ont déjà adopté notre modèle SQ10 pour leur entraînement. Celui-ci est aujourd’hui homologué par la fédération anglaise de tir et en cours d’homologation par la fédération française (FFT) et nous allons continuer cette démarche pour aller vers l’homologation par la fédération internationale de tir », ajoute-t-il. En parallèle, la start-up développe SynQro, un logiciel de supervision qui permet de gérer plus de 100 cibles à la fois. Objectif : le marché de l’organisation des compétitions.


  • Capteurs piézoélectriques placés sous une plaque en polycarbonate. Largement utilisés dans la vie courante (pare-chocs de voiture, vibreurs de téléphone portable, etc.), ils émettent un signal électrique en correspondance avec une sollicitation mécanique reçue ou, à l’inverse, vibrent en fonction d’un signal électrique reçu. Ici, ils mesurent le point d’impact du projectile.
  • Affichage des cibles sur un écran situé derrière la plaque.
  • Logiciel associé, proposant de nombreuses fonctionnalités : analyse des scores (moyennes, écarts-types, suivi des performances).


  • Clubs de tirs
  • Particuliers
  • Organisateurs de compétitions.

Dates clés

  • 2017 : Création de Sport Quantum
  • 2018 : Premières ventes de cibles
  • 2019 : Levée de fonds de 600 000 €
  • 2020 : Homologation en cours de la cible SQ10 par la fédération française de tir sportif ; Cibles SQ10 choisies par la région Île-de-France pour ses compétitions (110 cibles)
  • 2021 : Première médaille de la Team Sportive Sport Quantum aux JO 2020 de Tokyo (Lucas Kozniesky - US).


Novembre 09, 2021 | Real-time 3D imaging for better non-destructive testing

imagerie 3D 250
Florence Pillet

​CEA-List has developed new algorithms that bring the power of real-time 3D image reconstruction to non-destructive testing.

​Non-destructive testing (NDT) is used in the nuclear, automotive, petroleum, and other industries to inspect a wide variety of parts for defects in their geometry or materials. In conventional NDT, the signals acquired are used to generate 2D images of the parts being inspected in real time. The jump from 2D to much more complex 3D imaging, however, remains a challenge. The main difficulty is that current NDT equipment simply cannot accommodate the large number of signals and processing power required. For the first time ever, this technological hurdle has been addressed.

CEA-List scientists used a Fourier approach to 3D reconstruction to develop new algorithms that effectively bring the number of computational operations required to reconstruct the image to a bare minimum. They then integrated the algorithms into a prototype 4D (real-time 3D) imager.

And it worked: In tests, the prototype successfully detected cracks, porosities, and shrinkage intentionally introduced into steel blocks produced by additive manufacturing at the Additive Factory Hub (AFH). Even better, the prototype outperformed the state of the art, "seeing" porosities measuring just 0.6 mm in diameter and precision-locating them deep inside the material to within a tenth of a millimeter.

The researchers are now trying to make this new imaging technique faster and more powerful without compromising on image quality—the prerequisite to successful scale up and transfer of the technology.

Read article at