Scientific direction Development of key enabling technologies
Transfer of knowledge to industry

PhD : selection by topics

Inverse reinforcement learning of a task performed by a human





Learning from demonstration involves an agent (e.g., a robot) learning a task by watching another agent (e.g., a human) performing the same task. It often uses reinforcement-learning methods to improve the robot's ability to perform a task in new situations (i.e., generalization). These methods involve providing a positive reinforcement (i.e., a reward) when the outputs of the algorithms help achieving the task, but require a human designed reward function. The more the task is complex the more difficult is the reward function to design, but it can be learned from a series of examples with methods called inverse reinforcement learning. The use, jointly or not, of these techniques has shown encouraging results, but which are limited to toy examples and cannot be adapted as such to tasks more representative of the industrial environment. During the thesis, the PhD student will analyze and test state-of-the-art previous works. S/He will then propose a method, combining inverse reinforcement learning to other algorithms (e.g., generative adversarial networks, GAN), so that the robot will understand the task performed by the operator (with as little explanation from the operator as possible), and will generalize enough to make the robot robust to dynamic environments (obstacles, moving objects?). This method should be suited for a "pick and place" task in an industrial environment and ensure a reasonable enough learning period (information a priori, feedback from the operator) for tasks of medium complexity.

3D Objects discovery in 3D scene





Object detection and localization in images is a problem studied since many years. The latest technological developments now allow the real-time acquisition of depth data coupled to color data (RGBD). At the same time, modern machine computing capabilities and intelligent image processing methods have led to significant advances in the detection / localization of 2D objects with many different approaches (bounding boxes, contours, from CAD models ...). An important step is being taken in recent years with the research conducted to directly extract the volume of detected objects and their position in 3D. These works are still in their infancy, but the first results are encouraging, both from 2D images (eg DeepManta) and from 3D images (eg Deep Sliding Shapes). However, there remain several identifiable scientific / technological barriers before allowing the democratization of this type of approach for the automatic extraction of objects in potentially unknown scenes. The objective of this work is to identify the current approaches of detection / localization of 3D objects, to target their weaknesses and work on new processing technologies to mitigate them. Moreover, the object discovery in unknown environments and the inference of the operator's intention by observation / location of his attention are two areas of interest that this work aims at addressing. Beyond their applications for demonstration learning, the software bricks resulting from this project can also be reused for other applications such as augmented reality ("smart" scanning, etc.), surveillance or mobile mobility for example.

Person re-identification and cross-domain adaptability

Département Intelligence Ambiante et Systèmes Interactifs (LIST)

Vision & Ingénierie des Contenus (SAC)



Automatically re-identifying people viewed by cameras is a key functionality for videoprotection applications. It consists in retrieving occurrences of a person from a set of images. Despite the many studies on this topic in the past few years, modeling human appearance remains a challenge. Indeed, re-identification models have to discriminate distinct people (in spite of their possible similarity) while being robust against the high variability of their visual appearance (caused by their posture, lighting conditions, camera viewpoint, sensitivity and resolution, ?). Besides, partial occlusion and alignment errors on the detected people have to be coped with. Even if deep supervised learning methods have been greatly improving re-identification performances on some academic datasets, difficulties remain for real implementations in operational environments. Indeed, a model trained on a specific dataset usually does not perform well if applied on other datasets as it is. Furthermore, manual data annotation in the target domain is a tedious thus costly task. In this thesis, we will study the appearance model adaptability to target domains in which only data without annotation is available. Unsupervised transfer learning methods can be used. The proposed approaches will cope with scalability issues in order to address large datasets.

modeling biomass torrefaction at pilot scale with data measured in laboratory at small scale

Département Thermique Biomasse et Hydrogène (LITEN)

Laboratoire de Préparation de la Bioressource



Torrefaction is a thermal pretreatment applied to biomass, carried out under neutral atmosphere for several tens of minutes, at temperatures between 200 and 300°C. Once treated, the solid exhibits properties closer to those of coal (fossil), making it suitable to the same industrial facilities as this latter. The biomass platform of CEA Grenoble has been equipped with a pilot-scale torrefaction oven (capacity: 150kg/h of wood). The results obtained in this pilot oven are always out of sync with the torrefaction data measured in the laboratory. Therefore, the validity of the change of scale for this process is questionable. The aim of this thesis is to improve the extrapolation at pilot scale of data measured with small analytical equipment. Three successive phd prepared in the laboratory, have led to a model representing the different chemical transformations of biomass during torrefaction. This model will be used in the proposed phd. This work will require to perform a lot of experimental investigations, in the laboratory (small scale) as well as to participate to torrefaction campaigns with the pilot.

Integrated vector network analyzer for biomedical sensing applications

Département Architectures Conception et Logiciels Embarqués (LIST-LETI)

Laboratoire Architectures Intégrées Radiofréquences



This thesis addresses the topic of highly integrated vector network analyzer (VNA) in the context of biomedical sensing applications. The thesis study will cover the architecture, the design and the measurement of such a VNA. This PhD research will give the opportunity to work in cross-scientific disciplinary from the microelectronic design to the understanding of biological material characteristics. To achieve this objective several milestone will have to be successfully completed. The awaited innovation will encompass several aspect: high precision local oscillator, high sensitivity, low cost CMOS process. The thesis will take place in the CEA Leti institute under the supervision of Dr Martineau and Dr Gonzalez (HDR). The publication in journals and international conferences will be encouraged and facilitated.

Study and design of an integrated system for the automatic calibration of dispersions within a transducers array and application to a PMUT array

Département Architectures Conception et Logiciels Embarqués (LIST-LETI)



The purpose of this thesis is to study and design an integrated electronic system dedicated to the automatic and continuous compensation of dispersions within a MEMS (Microelectromechanical Systems) array. With the dissemination and the continual expansion of Internet of Things (IoT) and Cyber-Physical Systems (CPS), man-machine and machine-machine interfaces require increasingly efficient and sophisticated sensors. In addition to advantages in cost, reliability, size and power consumption, MEMS based transducers enable sensors to integrate more and more intelligence in their front-end electronics. They also allow innovative topological configurations giving access to measurement ranges that are not addressable by their discrete counterparts. Arrays of MEMS based transducers enable the spatial discretization of the transduction surfaces and improve the measurements yields and accuracies (gas detector, mass spectrometry, pressure distribution, etc.). They also enable the resolution improvement of electromagnetic and acoustic beams (location, navigation, communication, etc.). Despite the considerable technological advancements that MEMS are continually enjoying, some application requirements are beyond the transducers intrinsic performances. It is then necessary to implement calibration systems to correct the transducers biases introduced during manufacture or evolving with the operating conditions. The evaluation and compensation of these errors requires costly calibration process in a dedicated test laboratory, that are not compatible with massive production. The aim of this thesis is to achieve an integrated electronic diagnostic alternative, an electromechanical BIST (Built-In Self-Test) specific to transducers arrays, combined with an automatic correction system, which will operate in coexistence with the main functions of the sensor interface. The proposed use-case is that of PMUT (Piezoelectric Micromachined Ultrasonic Transducer) arrays. These devices offer alternatives and complementary solutions to electromagnetic sensors for detection and localization [1], gesture recognition [2] or wake-up signals detection [3]. For most applications, these resonant transducers operate in transmit / receive modes (TX / RX) and need to be actuate at their resonance frequency to optimize the transmission power. The emitted and received beam is focused and steered by phase control. Errors and dispersion in the PMUT characteristics generates biases in their resonant frequency, gain and quality factor, leading to losses and distortions in the emitted and received beams. For example, a few percent of dispersions on the mechanical stiffness of the transducers can lead to several tens of percent loss on the acoustic power transmitted to a target. As a first step, the doctoral student will get familiar with the quantities and physical phenomena characterizing PMUT arrays. Based on an analytical model developed within the host laboratory, he will be able to understand the sensitivities to dispersions and their impact on the beam power and directivity. He will then define the electronic methods and architectures that will allow the system to converge towards the optimal operating conditions, for example by identifying the average resonance frequency of the array the required phase and gain correction coefficients to allocate to each transducer. The architecture and implementation choices must allow the system to adapt itself according to dispersions and drifts in a continuous and autonomous way, without disrupting the main measurement functions. The chosen solution will be implemented and validated in a mixed design environment in order to result in a functional demonstrator. [1] Przybyla, R. J., Tang, H. -., Guedes, A., Shelton, S. E., Horsley, D. A., & Boser, B. E. (2015). 3D ultrasonic rangefinder on a chip. IEEE Journal of Solid-State Circuits, 50(1), 320-334. [2] Ling, K., Dai, H., Liu, Y., & Liu, A. X. (2018). Ultragesture: Fine-grained gesture sensing and recognition. Paper presented at the 2018 15th Annual IEEE International Conference on Sensing, Communication, and Networking, SECON 2018, 1-9. [3] Yadav, K., Kymissis, I., & Kinget, P. R. (2013). A 4.4-µ W wake-up receiver using ultrasound data. IEEE Journal of Solid-State Circuits, 48(3), 649-660.

73 Results found (Page 2 of 13)
first   previous  1 - 2 - 3 - 4 - 5  next   last

Voir toutes nos offres