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 21, 2018 | Neural networks will soon be capable of incremental learning

neurones 250
© whitehoune –

A method based on a model of human memory has made it possible for neural networks to learn incrementally. This advance will open the door to new possibilities in the field of autonomous systems.

Neural networks are very powerful for deep learning applications. However, they are not particularly well-suited to incremental learning. Currently, when a neural network learns a new piece of information, old information is overwritten. A solution to this "catastrophic forgetting" would make neural networks more operational in autonomous systems running in constantly-changing environments.

Researchers at Leti, a CEA Tech institute, worked with cognitive neuropsychology lab LPNC, which has been developing a human memory model since the 90s, and fellow CEA Tech institute List*, which is developing an artificial neural network simulation tool called N2D2, to come up with a model that could be a game changer. The model can re-learn all information (old and new) together using two neural networks, eliminating the need to save old information to an external memory, which would drastically increase memory requirements.

Here's how it works: The first network is presented with alternating true examples that correspond to the new information being learned and "pseudo examples" that are generated by the second network. These "pseudo examples" represent what has already been learned and are used to "refresh" the first network's memory, so to speak. The primary advantages of the method are that it does not limit network plasticity and does not require additional memory.

*List earned the prestigious Institut Carnot seal in 2006 (Institut Carnot TN@UPSaclay).



November 29, 2018 | Euclid space telescope’s structural welds inspected using ultrasonic testing

List Euclid 250
© Valerio Pardi - Fotolia

A non-destructive testing (NDT) method for inspecting sintered welds was developed to ensure the structural integrity of the European Space Agency's Euclid space telescope.

The Euclid satellite will be embarking on a mission to understand the accelerating expansion of our Universe. So, it had better be sturdy! The satellite's telescope is made from four pieces of silicon carbide welded together using a sintering process patented by Leti, a CEA Tech institute.

Before the satellite can be launched, the telescope's welds must first be inspected on the ground. To perform the inspection, Mersen Boostec, an expert in sintered silicon carbide, turned to another CEA Tech institute, List, which had already successfully inspected the ESA's Gaia satellite.

Silicon carbide offers some exceptional thermal properties that make it suitable for space applications. However, the material is also very stiff, which makes it difficult to probe using ultrasonic NDT techniques. Ultrasonic waves are propagated at a high speed and are reflected off of the walls of parts with complex geometries, creating parasite echoes that can make interpreting the results a challenge. To inspect the entire surface of the sintered welds through to their centers, engineers at Airbus Defence & Space worked with researchers at List to make adjustments to the geometries of the areas to be inspected. They then used CIVA simulation software to calculate and tweak the focal parameters (like the angle of the ultrasonic beam) to detect all defects measuring more than 4 sq. mm.

They tested the results of the simulations on several reference defects in the lab before successfully using them on the Euclid telescope.

*List earned the prestigious Institut Carnot seal in 2006 (Institut Carnot TN@UPSaclay).




November 22, 2018 | EVOC multi-mode nuclear physics training platform

evoc List 250
© franz12 – Adobe Stock

The INSTN (France's national institute for nuclear science and technology) EVOC training platform has been up and running since October. The platform's innovative virtual-reality training environment makes it like no other facility in the world.

The EVOC (Advanced Virtual Open Core) nuclear physics training platform combines top-notch teaching staff, in-depth physics and nuclear science knowledge, and excellence in digital technology. List, a CEA Tech institute, drew on resources from across the CEA to help create training programs that are unique worldwide. The EVOC platform leverages virtual reality and reactor simulators. It was unveiled in late June 2018 at the WNE (World Nuclear Exposition).

The multi-mode platform positions INSTN as a provider of immersive 1:1 scale learning experiences for nuclear-industry professionals. The new kinetic reactor simulator integrated into a control panel is paired with List's multi-physics virtual reality simulator, XDE, which has already been used in a variety of other industries. These digital technologies enhance the virtual experience through 3D models of tangible elements from the real world, including sounds, haptic effects, and objects in the scene, to provide a realistic simulation of physical phenomena.

All of users' interactions with each other and with their environment are calculated, taking into account gravity, touch, light, and the movement of air or water. The software is capable of reproducing these physical phenomena in real time via a headset and on the various virtual reactor control panel interfaces, despite the multitude and complexity of the objects and of the scene.

Very few research organizations in the world are capable of such advanced technology. In addition, the platform requires very few sensors and switches on users and in the environment. The new platform could potentially be adapted for use in the aeronautics, medical, and other industries. The hardware and software architectures are easy to modify and are well-suited to other environments.

*List earned the prestigious Institut Carnot seal in 2006 (Institut Carnot TN@UPSaclay).




November 15, 2018 | From industrial robots to cobots

robotique cobotique cea list 250
© Echord++ / Hauke Seyfarth

Industrial robots were developed to replace human operators on repetitive or dangerous tasks. Cobots—or collaborative robots—are different. Human operators continue to use their unique know-how, while the robot works with them to make their work easier.

List*, a CEA Tech institute, originally started developing robots to replace human operators in hazardous environments. New developments mainly consisted of remotely-operated robots for the nuclear industry. These robots were made up of two arms that were first connected mechanically and, later, by computer. This technology is known as remotely-operated force feedback robotics.

List’s later research applied this force-feedback robotics technology to industrial settings, and collaborative robotics was born. Cobots do not replace human operators. They interact with their human counterparts, guiding their movements and helping them lift loads, for example.

It takes knowledge in fields as diverse as mechanical engineering, electronics, automation, algorithms, and IT do develop a cobot. Researchers at List patented an innovative very-low-friction cable and ball-screw mechanism for its cobots. This advance enables cobots that are safe, accurate, and very sensitive in terms of force feedback, making them ideal for a wide range of tasks in the nuclear, offshore, mining, manufacturing, and healthcare industries.

Isybot, a List spinoff, developed the institute’s cobotics technologies to totally reinvent industrial workstations like sanding, polishing, and finishing. The company designs intuitive, accurate, lightweight cobots that are completely safe for the operator they are assisting. Plus, the operator can program the cobot simply by performing the task to be completed. The robot memorizes the movement and amount of force to apply. No computer programming is required! The operator can also make changes or corrections at any time, retaining full responsibility for quality completion of the task.

*List earned the prestigious Institut Carnot seal in 2006 (Institut Carnot TN@UPSaclay).




November 8, 2018 | Electricity grids get even smarter

reseaux electriques 250
© Mimi Potter - Fotolia

Research conducted under two EU projects has given rise to a new distributed energy management platform capable of coordinating the energy consumption of households at the scale of an entire neighborhood.

The objective of the EU FUSE-IT and MAS2TERING projects is to develop technologies to support the transition from centralized to distributed energy management. List, a CEA Tech institute, contributed to these projects with “agents” (autonomous software entities) that are installed in each household. However, the major innovation is not the agents themselves, but rather getting them to communicate and coordinate for more efficient energy usage both by individual households and for the neighborhood as a whole.

Here is how it works: The agents negotiate which source of energy should be used and when, factoring in occupant comfort, flexibility, and other user needs. Distributed decision-making also offers the advantage of protecting personal data, as two agents can arrive at an agreement without exchanging any sensitive information. This is particularly important in light of the new European General Data Protection Regulation (GDPR), which has already come into force in France and other countries in Europe.

This distributed energy management system was tested on around a hundred households. It offers the potential for expansion to other types of energy utilities (heat). Systems in which each piece of equipment would be controlled by its own agent are also a possibility—one that is currently being explored in research conducted under four additional EU projects.

*List earned the prestigious Institut Carnot seal in 2006 (Institut Carnot TN@UPSaclay).