Website: https://soe.lau.edu.lb/about/people/charbel-dalely-tawk.php

Résumé :

Biographie :

Mots-clés : 3D Printed Soft Grippers with Integrated Sensing Capabilities; 3D-Printed Soft Actuators and Sensors; Soft Human-Machine Interfaces; Additive Manufacturing

Visual-RRT*: Generating Asymptotically Optimal Trajectories with Vision-based Controllers

In this talk, we will present a planning method for robotic systems that, by construction, guarantees the planned trajectories to be executable in closed-loop with a visual servoing controller, in addition to assuring their asymptotically optimality in the state space. The design of this method requires one to obtain a locally optimal steering method, locally stable closed-loop controllers used to track the optimal trajectories, and a mapping between the motion of the robot in the state space and the corresponding motion of the visual features in the image space. As a result, the proposed approach allows the system to navigate in a closed-loop manner among non-convex obstacles, following asymptotically optimal trajectories. The method is validated using two systems: a nonholonomic robot and a robotic arm. Real-world experiments are carried out for the first system.

Mots-clés : Motion Planning; Visual servoing

Seminar by Vianney Monnier presented 16/09/2024:

The presentation will sum up previous studies on fluid mechanics and gaseous detonation, introduce salient results from Vianney's work and will propose directions for research in mechanics and mechanical design to discuss at ROB.

We have analyzed the cellular patterns of a steady gaseous detonation using frontal and longitudinal and soot plates in tubes whose cross sections have the same area of 16 cm2 but different shapes, i.e., round and square. The reactive mixture is stoechiometric hydrogen-oxygen, diluted or not by a 40% volume of argon. The results show that, for large enough initial pressures p0, all front-view patterns are irregular and independent of the cross-section shapes. Therefore, the notion of regularity depends not only on the mixture but also on both p0 and the cross-sectional shape. They also show that the longitudinal records for the considered mixture show regular patterns independently of p0 and the cross-sectional shape. That raises questions about the ability of longitudinal recordings alone to characterize cell structure. They certainly include how the transverse waves interact with each other as well as the walls. We have developed a model to predict the mean cell width for a cell quantity on the detonation front, or equivalently, p0, sufficiently large. The ingredients are graph theory, geometric probabilities and the Zeldovich-Von Neuman-Döring model. The predictions agree well with measurements, at least within the experimental uncertainties, depending on the mixture and the scheme of chemical kinetics considered. We have also studied the influence of the shape of the tube cross-section on the transient phenomena resulting from the diffraction of a steady detonation through a single-hole orifice plate, depending on the blockage ratio, initial pressure and argon dilution of the mixture. The shape of the tube cross-section induces marked qualitative and quantitative differences, all other things being equal.

Abstract : The Reinforcement Learning (RL) field has been successfully applied to a wide range of problems in the past years, demonstrating both its versatility and efficiency. The robotic domain in particular benefited from the tremendous progresses made in RL. Despite numerous improvement regarding its sample-efficiency, learning a policy from scratch still requires millions (if not dozens of millions) of interactions with the environment to converge to a high-reward policy. This is usually because the agent does not have any prior information about the task or its own physical embodiment. In general, the canonical way to address and mitigate data-hungriness is to use Transfer Learning (TL), a method leveraging prior experience or knowledge to speed up learning of a distinct but related task. This is currently extensively used in deep learning based Computer Vision or Natural Language Processing. However, finding a practical and general approach for pre-training and transfer learning in the context of RL policies, is still an open research problem. In this work, we explore TL in the context of RL with the specific purpose of transferring policies from one agent to another, even in the presence of morphology discrepancies or different state-action spaces. In other words, we wish to reuse knowledge acquired by an agent (source) on a task to speed up (or even avoid) the learning process of a dissimilar agent (target) on the same task. We highlight the particular challenges raised by the application of TL in the context of cross-robot skill transfer and, inspired by prior works, propose an efficient method to address these matters.