Séminaire Robotique et Vivant
🤖 🌱
Bio-inspired Robotics Seminar

Terrestrial vertebrates are thought to have achieved high locomotor ability and ecological success through their musculoskeletal systems. Efficient movement is achieved by selecting one solution from the redundant combinations of joint angles and muscle activities available for a given action. However, a method for generating smooth and efficient motion in animal musculoskeletal systems with such highly redundant degrees of freedom, or in robots that reproduce them, has not yet been established. One key to this problem may be the existence of passive mechanisms in vertebrate musculoskeletal systems specialized for locomotion. For example, in the horse hindlimb, passive coupling within the muscular system coordinates joints distal to the knee in response to hip flexion and extension, thereby automatically generating swing-phase motion. Similarly, in the avian hindlimb, passive coupling within the muscular system coordinates joints distal to the ankle during knee flexion and extension, contributing to swing-phase motion. In the presence of ground reaction force, these systems lock the joints and efficiently support body weight. Moreover, we have shown that crocodilian hindlimbs also possess a mechanism for supporting body weight with the ground reaction force. These functionally distinct passive mechanisms are adaptively realized through interaction with the environment while relying on largely overlapping specific muscle groups. We refer to the musculoskeletal mechanism that adaptively realizes inter-joint coordination through passive muscular coupling, based on interaction with the environment, as an adaptive passive coordination mechanism. The significance of this mechanism lies in its ability to greatly reduce the redundant degrees of freedom of the musculoskeletal system for a specific function through passive muscular coupling adaptively selected based on interactions with the environment, while requiring little active control from the nervous system. Moreover, applying this principle to robots produces highly natural walking, suggesting that adaptive passive coordination mechanisms, which generate motion from the mechanical properties governing the body and environment, are a fundamental source of smooth and efficient locomotion. Crocodilians improved their locomotor performance by evolving from sprawled to semi-erect locomotion, whereas birds possess legged locomotor capabilities equal to or exceeding those of mammals. Both groups belong to Archosauria. This suggests that adaptive passive coordination mechanisms may be widely present in archosaurs as functionally homologous mechanisms and may also be applicable to extinct members of the lineage, such as dinosaurs. Since adaptive passive coordination mechanisms are realized by a limited set of specific muscle groups, their effectiveness can be tested through dissection that preserves those muscles or through robotic implementation. However, muscular motion is constrained in complex ways, making it difficult to isolate the mechanism solely by dissection, except in exceptional cases such as horses. It is therefore necessary to combine anatomical approaches with an understanding-by-building approach using robots. The aim of this study is therefore to identify, from anatomical and functional-morphological viewpoints, the adaptive passive coordination mechanism for locomotion that is embedded in the archosaur musculoskeletal system and expressed through environmental interaction, and to validate it by implementing it in robots based on skeletal models. By focusing on functional homology across the diverse species within Archosauria, we aim to clarify whether adaptive passive coordination mechanisms for locomotion are universally present in this clade.

Modular robots can adapt their shape and function to diverse tasks, but this flexibility introduces challenges in both control and reliability. In this talk, I present a framework for user-guided control of shape-changing robot collectives using optimization-based safety constraints and morphology-aware interfaces. I also introduce a resource-sharing approach that improves system resilience by enabling modules to support each other in power, communication, and sensing. Together, these contributions show how modular robot systems can become both easier to control and more robust as they scale.

La résilience est la capacité de résister à des difficultés inattendues ou de s'en remettre rapidement. Appliquée à la robotique mobile, elle peut être définie comme la capacité des robots à absorber les chocs pendant le mouvement, à entrer en collision sans subir de dommages structurels irréversibles ou à subir l'impact d'objets volants sans s'écraser. La survie est aujourd'hui une donnée fondamentale pour les robots lorsqu'ils évoluent dans un environnement non structuré, et elle reste un défi. La multimodalité en robotique s’applique de plus en plus mais reste pour l’instant peu robuste aux chocs notamment. La fourmi du désert un exemple emblématique de résilience : navigation dans un environnement extrême et hostile nécessitant des capacités hors du commun de grand intérêt pour des robots mobiles pouvant naviguer dans des environnements dépourvus de système de navigation (GNSS-denied). Les mécanismes déployables, les structures souples et les origamis ont permis aux roboticiens de concevoir des structures originales, capables de changer de forme dynamiquement tout en résistant aux chocs pour certains. A l’instar de leurs modèles animaux, le zoobot de la robotique résiliente et multimodale est riche en diversité. J’aborderai ici quelques exemples emblématiques de cette robotique mobile de demain.

Learning to change shape is a fundamental strategy of adaptation and evolution of living organisms, from bacteria and cells to tissues and animals. Human-made materials can also exhibit advanced shape morphing capabilities, but lack the ability to learn. Here, we build metamaterials that can learn complex shape-changing responses using a contrastive learning scheme. By being shown examples of the target shape changes, our metamaterials are able to learn those shape changes by progressively updating internal learning degrees of freedom—the local stiffnesses. Unlike traditional materials that are designed once and for all, our metamaterials have the ability to forget and learn new shape changes in sequence, to learn multiple shape changes that break reciprocity, and to learn multistable shape changes, which in turn allows them to perform reflex gripping actions and locomotion. Our findings establish metamaterials as an exciting platform for physical learning, which in turn opens avenues for the use of physical learning to design adaptive materials and robots.

Dans ma thèse récemment soutenue, nous avons construit un modèle géométriquement exact de matériaux micro-structurés en généralisant la notion de variétés de Riemann--Cartan, utilisée dans les modèles micromorphes. Dans cet exposé, je commencerais par motiver physiquement la nécessité d'un tel modèle. Je prendrais un soin particulier à détailler l'intuition du problème dans un cadre simplifié, HPP heuristique. Nous verrons comment ce modèle repose sur l'interaction de deux échelles : une échelle macroscopique et une échelle microscopique. Leur couplage engendre des phénomènes émergents tels que les dislocations (torsion) et les disclinaisons (courbure). Dans un second temps, nous reprendrons du début le problème en, cette fois-ci, modélisant rigoureusement en grande déformation. Le placement est décrit par un morphisme F, analogue à un placement au second ordre du matériau dans l'espace. Afin de permettre l'apparition de disclinaisons (courbure), F n'est en général pas un gradient. Au cœur du modèle se trouve la notion nouvelle de projection d'interprétation et la pseudo-métrique qui lui est associée. Une notion d'objectivité (isotropie de l'espace) est formalisée et les invariants associés sont calculés. Via le calcul variationnel, les équations d'Euler-Lagrange sont obtenues. En découlent, dans le cas linéarisé, des EDPs de type Helmholtz reliant dislocation et disclinaisons, illustrées par des simulations numériques. Enfin, si le temps le permet, nous verrons comment le modèle peut être appliqué aux poutres, donnant un framework continûment paramétrisée et incluant notamment Timoshenko et Euler-Bernoulli (pour le cas HPP).

The formulation of robotics tasks as optimization problems is ubiquitous, spanning from traditional sense-act architectures to contemporary policy learning paradigms. A significant challenge, however, is the non-convex nature of these problems, which often leads to suboptimal solutions when using standard local solvers without adequate initialization. Consequently, current robotics pipelines often employ sophisticated initialization schemes, large-scale computational searches, or mitigation strategies to address suboptimal solutions. This talk will discuss a more principled alternative for achieving global optimality, rooted in the field of polynomial optimization—specifically, sums-of-squares and moment hierarchy methods. While these techniques have generated significant interest in robotics since their inception a few decades ago, broad adoption has been slowed down by practical challenges. I will present our recent contributions aimed at lowering the barrier to entry of these methods by automating the formulation and tightening of their relaxations. Furthermore, I will discuss our advances in enhancing scalability, moving beyond poorly conditioned monomial feature functions, and developing structure-exploiting solvers for the resulting semidefinite programs. The talk will conclude with an outline of persistent challenges and promising directions for future research.

Active filaments are a workhorse for propulsion and actuation across biology, soft robotics and mechanical metamaterials. However, artificial active rods suffer from limited robustness and adaptivity because they rely on external control, or are tethered to a substrate. Here we bypass these constraints by demonstrating that non-reciprocal interactions lead to large-scale unidirectional dynamics in free-standing slender structures. By coupling the bending modes of a buckled beam anti-symmetrically, we transform the multistable dynamics of elastic snap-through into persistent cycles of shape change. In contrast to the critical point underpinning beam buckling, this transition to self-snapping is mediated by a critical exceptional point, at which bending modes simultaneously become unstable and degenerate. Upon environmental perturbation, our active filaments exploit self-snapping for a range of functionality including crawling, digging and walking. Our work advances critical exceptional physics as a guiding principle for programming instabilities into functional active materials.

Cilia and flagella exhibit a characteristic periodic movement, enabling them to swim as efficiently as possible in a viscous fluid. This movement is generated by a periodic internal activation structure present along the entire length of the filament, called the axoneme. In the axoneme, forces are generated by molecular motors which, together, contribute to the curvature of the filament. After studying the molecular motor model, several versions of coupled systems of partial differential equations will be proposed, modeling activation in a periodic cell. The theoretical results will also be illustrated by numerical simulations.

This project presents the integrated development of a soft, cable-driven exosuit for upper limb rehabilitation, to be controlled via real-time surface electromyography (sEMG) and inertial measurement unit (IMU) data. The work bridges two core components: (i) a wearable exosuit designed to assist elbow and wrist movements using optimized cable actuation, and (ii) a machine learning-based system to recognize movement intent from EMG signals using artificial neural networks (ANNs). EMG data is processed through robust time- and frequency-domain feature extraction, and classified into movement states such as Rest, Flexion, and Extension with classification accuracies up to 92%. The exosuit, actuated via Maxon motors and controlled by a National Instruments sbRIO unit, was optimized for force and moment efficiency to enhance wearer comfort. Real-time integration is being in trial using LabVIEW, enabling live classification and joint angle feedback using IMUs. The system is being validated through preliminary trials on healthy individuals, should be demonstrating feasibility for assisting in activities of daily living and rehabilitation tasks. This multidisciplinary effort represents a promising step toward intent-driven assistive exosuits for neuro-rehabilitation and functional support. In future we will have multiple trials and validation on healthy subjects with their feedback and validation to go before clinical trials for patients.

In this talk I will provide an introduction to generative modeling and the current state-of-the-art method of the field: diffusion models. I will give an historical overview of the method and present some of the current competitive open source methods. I will then move on to the use of generative models in the field of robotics. In particular, I will discuss diffusion policies which are a new way to generate robot behavior by representing a robot's visuomotor policy as a conditional denoising diffusion process. I will conclude with some of the open questions and challenges lying at the intersection of generative modeling and robotics.

The study of animal movement and related morphological adaptations provides a better understanding of evolutionary processes and can also be particularly interesting for the design of bio-inspired robots. Indeed, bio-inspiration has led to the emergence of technical solutions reproducing principles or structure initially described in living organisms and has provided insights to make efficient autonomous machines. In addition, these robots can be used in return as scientific tools to investigate animal locomotion. Subterranean animals, such as moles, display a very specific morphology resulting from their lifestyle. As they are all strictly fossorial, Talpa species are of particular interest when focusing on burrowing performance and fossorial adaptations. Hence, our project aims at understanding the relation between morpho-functional patterns and burrowing performance in two Talpa sister-species co-occuring in metropolitan France (Talpa europaea and the newly described species Talpa aquitania). To go deeper in the understanding, a bio-inspired « mole-robot » will be designed. Anatomical dissections were performed and the process to obtain robotic data from these biological data will be described. These results allowed us to more precisely define muscles insertion and joints areas in moles forelimb, thus allowing us to design the kinematic joints on the kinematic chain constituting the mole-robot.

Vine growing robots are soft, inflatable, bio-inspired robots which grow at the tip to deploy. Unless traditional endoscopes or inspection tools which deploy by rigid body motion, the apical extension of vine growing robots enables deployment without applying shear of friction forces onto their environment, which is one of their major benefits. To have such characteristics, vine robots are composed by a thin tube everted in itself. When pressurized, the material stored inside the tube translates towards the tip of the tube where it everts. The material everted at the tip then forms the vine robot body, which remains stationary with respect to the environment. These robots have been advantageously proposed for medical applications such as the deployment in the vasculature, in the mammary duct, in the intestine, and for industrial and larger scale applications, such as growth in granular environments, inspection of archaeological sites, and for search and rescue operations. However, these robots lack important features in order to be useful for most applications, or suffer from limitations. For instance, most applications require a passageway for tools, i.e. a working channel, in order to provide direct access to the robot tip from the base. This enables tools to be inserted and swapped, in order to perform some tasks. Tasks can include the use of a camera and light source for site visualization, laser, grippers and cutting tools in surgical applications, or the transmission of water or goods for search and rescue applications. Current approaches for including working channels strongly limit the deployable lengths and speed of these robots. In addition, retracting these robots is challenging, as they tend to buckle. Finally, steering these robots through curved paths is important in order to limit normal forces applied to their environments and to enable navigation capabilities. In this presentation, I will present some of my latest work on the inclusion of working channels, retraction and steering systems inside vine growing robots, the functionalize them and unlock some of their potential in the context of medical applications.

In this talk, I will focus on the way insects, either because of the architecture of their antenna or because they move them, enhance the capture of odors. In particular, sex pheromones are emmitted in tiny amounts by females, but males can find their way in a turbulent environment over hundreds of meters, a feat which is still unsolved as today. This fundamental research has also important applications, from agroecology to explosive detection, and I will briefly describe the new bioinspired project which started last autumn.