phd_offer


PhD offer: Developmental robotics of birdsong

The Neurocybernetic team of ETIS Lab (CNRS, CY Cergy-Paris University, ENSEA) is seeking applicants for a fully funded PhD place providing an exciting opportunity to pursue a postgraduate research in the fields of bio/neuro-inspired robotics, ethology, neuroscience.
Webpage: https://www.etis-lab.fr/jobs/fully-funded-phd-position-in-the-field-of-neuro-inspired-robotics/

This PhD is funded by the French ANR, under the 4 years’ project “Nirvana” on sensorimotor integration of variability during birdsong learning. Partners are University Paris Nanterre, Paris-Saclay University, CY Cergy-Paris University and LS2N.

Motor variability, by allowing the exploration of the motor space, is an essential component of how sensorimotor circuits change across the learning course and may adapt to different conditions. Yet, knowledge remains sparse on how the variability of the input model contributes to the efficiency of sensorimotor integration during both speech acquisition and birdsong learning. This proposal will shed light on the impact of behavioral variability on learning skills and on the neurobiological processes at play to eventually feed neurocomputational models of human speech learning.

This PhD will propose new conceptual approaches to design an interactive bird robot that will be used both to teach and to learn dynamically from social interactions with a live bird. An artificial neural model, developmental and brain-inspired, will learn the sound structure in real time and without explicit supervision. Until now, AI models for developmental learning of vocalizations have been solely validated by comparison against a human-annotated corpus and not yet via direct sensorimotor interactions with living animals.

The PhD lasts for 3 years and includes a small teaching component.

To apply, send us an email first.
Please include:
1. A statement of research interests
2. A detailed CV
3. A transcript of your diplomas
4. Your Master/Diploma thesis, and any draft or published papers

Contact:
Prof. Alex Pitti, alexandre.pitti at cyu.fr
Assistant Prof. Sofiane Boucenna, sofiane.boucenna at cyu.fr
Dr. Vincent Lostanlen, vincent.lostanlen at cnrs.fr

Illustration by Louisane Araguas. From Alice Araguas’s PhD dissertation, a forerunner to the NIRVANA project.

Offre de thèse: Tenségrité et Intelligence Incarnée pour robot souple et agile

La plupart des robots existants actuellement sont rigides, lourds, et dangereux. Les enjeux écologiques et d’interactions avec l’humain insistent à rechercher plus de légèreté (économie de matière et d’énergie) et de compliance pour favoriser une interaction plus collaborative avec les humains.

Parmi les approches pour atteindre cet objectif, la tenségrité est une voie intéressante. Une structure tensègre est un ensemble d’éléments rigides soumis à de la compression mis en équilibre via des éléments élastiques en tension. La plupart des éléments biologiques sont des structures tensègres [1].

Les structures robotiques tensègres bio-inspirés sont souvent actionnées via des câbles (qui modélisent les muscles) dont la longueur et les forces exercées vont varier. Ceci sera la base d’un nouvel équilibre. C’est souvent ainsi que l’on transforme une structure tensègre en un mécanisme ou robot tensègre. L’actionnement par câbles impose des conditions d’unilatéralité sur les forces que les câbles peuvent exercer.

Un empilement de vertèbres peut être utilisé pour constituer un robot manipulateur tensègre avec une mobilité intéressante. Différents exemples dans la littérature existent, inspirés du cou des oiseaux [2,3] ou de colonnes vertébrales humaines [4], de serpents [5] ou de poissons [6].

L’intelligence est souvent associée au cerveau, à la résolution de problèmes complexes et à l’apprentissage. Il existe une autre forme d’intelligence : celle du corps en interaction avec l’environnement, que l’on appelle intelligence incarnée [7]. Elle est largement exploitée dans la nature et est importante en robotique pour exploiter les propriétés des corps des robots.

Dans le cadre de cette thèse, différents éléments peuvent contribuer à l’intelligence incarnée, comme par exemple l’équilibre postural des pattes d’oiseaux. Cet équilibre peut être vu comme un système de tenségrité passif qui leur permet de dormir debout, voire en équilibre sur une patte et de résister à des bourrasques de vents en équilibre sur une branche ou un fil [8].

L’empilement de structures de tenségrité peut permettre de filtrer et amortir les vibrations pour éviter de transmettre le long du cou d’un oiseau des vibrations dues à la marche jusqu’à la tête des oiseaux [3].

L’actionnement par câbles est généralement antagoniste, pour dépasser la limite due à l’unilatéralité des forces et tirer inspiration de l’actionnement antagoniste des muscles. Il est connu que la co-activation de muscles antagonistes permet dans les systèmes biologiques d’accroitre la raideur musculaire. Cependant, cette propriété intéressante ne peut se retrouver dans les systèmes mécaniques que pour un choix adapté des formes d’articulation et du choix de l’attache des muscles [9].

Les systèmes construits par empilement de modules de tenségrité commandés par un nombre limité d’actionneurs peuvent devenir instables si on construit leur commande sur des méthodes classiques d’inversion de modèle, et il est plus approprié d’utiliser la stabilisation naturelle du système pour obtenir des commandes stables [10]. Des modèles d’apprentissage et de contrôle prédictif [11, 12], exploitant la physique du corps [4,5], peuvent permettre de faire face à la complexité de tels systèmes afin d’atteindre une compliance dans l’action et une souplesse dans l’interaction [11].

L'objectif général de la thèse est de mettre en avant les principes de l'intelligence incarnée et de combiner différentes techniques de tenségrité pour la construction d’un robot agile et pour la réalisation de tâches en collaboration sûre avec l'humain dans le cadre de l'industrie 4.0. Les types de tâches considérées seront la manipulation, la saisie par enroulement et des actions mécaniques sur l'environnement (efforts extérieurs). La bio-inspiration guidera les choix d'actionnement: l'actionnement sera adapté en fonction des tâches et en cours de mouvement, à la manière des muscles qui sont activés par groupes différents selon les cas.

Les objectifs visés dans cette thèse porteront donc sur les aspects suivants :

- Conception d’un robot en s’intéressant en particulier au choix de l’actionnement ;

- Commande d’un tel système ; avec apprentissage prédictif et modèle génératif

- Interaction avec l’environnement.

Le planning de la thèse est le suivant

1ere année : prise en main et conception d’un système en étudiant en particulier l’effet du

choix de l’actionnement (en base ou sur les câbles)

2eme année : réalisation et commande du système

3eme année : tâches d'interaction physique

Le financement de cete thèse est assuré par le PEPR exploratoire 02R Organic Robots qui vise

à revisiter la conception et la commande de robots au plus près de l’humain. La thèse sera

menée en collaboration avec le LS2N à Nantes et le laboratoire ETIS à Cergy. Le doctorant sera

principalement au LS2N et sera encadré par C.Chevallereau (LS2N), A. Pitti (ETIS) et P. Wenger(LS2N).

[1]Robert E. Skelton and Mauricio de Oliveira. Tensegrity Systems. United States:Springer, 2009. doi: 10.1007/978-0-387-74242-7.

[2]Benjamin Fasquelle et al. « Identification and Control of a 3-X Cable-Driven Manipulator Inspired From the Bird’s Neck ». In: Journal of Mechanisms and Robotics 14.1 (2021), p. 011005. doi: 10.1115/1.4051521.

[3] Sun, Feng Wang, Jian Xu, « A novel dynamic stabilization and vibration isolation structure inspired by the role of avian neck », International Journal of Mechanical Sciences, Volume 193, 2021, 106166.

[4] Artem Melnyk & Alexandre Pitti (2018) Synergistic control of a multi-segments vertebral column robot based on tensegrity for postural balance, Advanced Robotics, 32:15, 850-864, DOI: 10.1080/01691864.2018.1483209

[5] X. Li, J. He and A. Pitti, "Travelling wave locomotion of a tensegrity robotic snake based on self-excitation controllers," 2022 9th IEEE RAS/EMBS International Conference for Biomedical Robotics and Biomechatronics (BioRob), Seoul, Korea, Republic of, 2022, pp. 01-06, doi: 10.1109/BioRob52689.2022.9925514.

[6] Bingxing Chen and Hongzhou Jiang. « Body Stiffness Variation of a Tensegrity Robotic Fish Using Antagonistic Stiffness in a Kinematically Singular Configuration». In: IEEE Transactions on Robotics 37.5 (2021), pp. 1712–1727. doi:10.1109/TRO.2021.3049430.

[7] Rolf Pfeifer, Alexandre Pitti, La révolution de l’intelligence du corps, 2012, Manuella Ed.

[8]A Abourachid, C Chevallereau, I Pelletan, P Wenger, An upright life, the postural stability of birds: a tensegrity system,

Journal of the Royal Society Interface 20 (208), 20230433, 2023

[9] V Muralidharan, N Testard, C Chevallereau, A Abourachid, P Wenger, Variable stiffness and antagonist actuation for cable-driven manipulators inspired by the bird neck, Journal of Mechanisms and Robotics 15 (3), 035002, 2023

[10] NJS Testard, C Chevallereau, P Wenger, Dynamics and computed torque control stability of an under-actuated tendon-driven manipulator, IFToMM World Congress Mechanism and Machine Science, 332-341, 2023

[11] Annabi, L. Pitti, A. Quoy, M. (2021) Bidirectional interaction between visual and motor generative models using Predictive Coding and Active Inference, Neural Networks, 143, 638-656.

[12] Chen, X Pitti, A (2022) Visuo-Motor Remapping for 3D, 6D Reach and Tool -Use Reach using Gain-Field Networks, IEEE ICDL Epirob.


PhD Offer: Tensegrity and Embodied Intelligence for a flexible, agile robot

Most existing robots are rigid, heavy and dangerous. Ecological issues and the need to interact with humans are increasingly driving the need for robots to be lighter (to save materials and energy) and more compliant so that they can interact with humans in a more collaborative way.

Tensegrity is an interesting approach to achieving this objective. A tensegrity structure is a set of rigid elements subjected to compression and brought into equilibrium via elastic elements in tension. Most biological elements are tensegrity structures [1].

Bio-inspired tensegrity robotic structures are often actuated via cables (which model muscles) whose length and the forces being exerted will vary. This will form the basis of a new equilibrium. This is often how a tensegrity structure is transformed into a tensegrity mechanism or robot. Actuation by cables imposes conditions of unilaterality on the forces that the cables can exert.

A stack of vertebrae can be used to form a tensegrity manipulator robot with interesting mobility. Various examples exist in the literature, inspired by the necks of birds [2,3] or human vertebral columns [4], snakes [5] or fish [6].

Intelligence is often associated with the brain, solving complex problems and learning. There is another form of intelligence: that of the body interacting with the environment, known as embodied intelligence [7]. It is widely exploited in nature and is important in robotics for exploiting the properties of robots' bodies.

In the context of this thesis, various elements can contribute to embodied intelligence, such as the postural balance of bird legs. This balance can be seen as a passive tensegrity system that allows them to sleep upright, even on one single leg, and resist gusts of wind while balanced on a branch or a wire [8].

Stacked tensegrity structures can filter and dampen vibrations to prevent walking vibrations from being transmited along a bird's neck to its head [3].

Cable actuation is generally antagonistic, to overcome the limit due to the unilaterality of the forces and draw inspiration from the antagonistic actuation of the muscles. It is known that the co-activation of antagonistic muscles in biological systems increases muscle stiffness.

However, this interesting property can only be found in mechanical systems if there is a suitable choice of joint shape and muscle atachment [9].

Systems built by stacking tensegrity systems controlled by a limited number of actuators can become unstable if their control is built on classical model inversion methods, and it is more appropriate to use the natural stabilization of the system to obtain stable control [10].

Learning and predictive control models [11, 12], exploiting body physics [4,5], can cope with the complexity of such systems in order to achieve compliance in action and flexibility in interaction [11].

The general aim of the thesis is to put forward the principles of embodied intelligence and to combine different tensegrity techniques for building an agile robot and for carrying out tasks in safe collaboration with humans in the context of Industry 4.0. The types of tasks considered will be manipulation, wind-up grasping and mechanical actions on the environment (external forces). Bio-inspiration will guide actuation choices: actuation will be adapted according to the task and during movement, in the same way as muscles are activated in different groups depending on the case.

The objectives of this thesis will therefore focus on the following aspects:

- Robot design, with particular emphasis on the choice of actuation;

- Control of such a system; with predictive learning and generative model

- Interaction with the environment.

The schedule for the thesis is as follows

1st year: familiarization with and design of a system, studying in particular the effect of the choice of actuator (on base or on cables)

2nd year: implementation and control of the system

3rd year: physical interaction tasks

Funding for this thesis is provided by the exploratory PEPR 02R Organic Robots programme, which aims to revisit the design and control of robots as close as possible to human beings.

The thesis will be carried out in collaboration with LS2N in Nantes and the ETIS laboratory in Cergy. The PhD student will work mainly at LS2N and will be supervised by C. Chevallereau (LS2N), A. Pitti (ETIS) and P. Wenger(LS2N).

[1]Robert E. Skelton and Mauricio de Oliveira. Tensegrity Systems. United States:Springer, 2009. doi: 10.1007/978-0-387-74242-7.

[2]Benjamin Fasquelle et al. « Identification and Control of a 3-X Cable-Driven Manipulator Inspired From the Bird’s Neck ». In: Journal of Mechanisms and Robotics 14.1 (2021), p. 011005. doi: 10.1115/1.4051521.

[3] Sun, Feng Wang, Jian Xu, « A novel dynamic stabilization and vibration isolation structure inspired by the role of avian neck », International Journal of Mechanical Sciences, Volume 193, 2021, 106166.

[4] Artem Melnyk & Alexandre Pitti (2018) Synergistic control of a multi-segments vertebral column robot based on tensegrity for postural balance, Advanced Robotics, 32:15, 850-864, DOI: 10.1080/01691864.2018.1483209

[5] X. Li, J. He and A. Pitti, "Travelling wave locomotion of a tensegrity robotic snake based on self-excitation controllers," 2022 9th IEEE RAS/EMBS International Conference for Biomedical Robotics and Biomechatronics (BioRob), Seoul, Korea, Republic of, 2022, pp. 01-06, doi: 10.1109/BioRob52689.2022.9925514.

[6] Bingxing Chen and Hongzhou Jiang. « Body Stiffness Variation of a Tensegrity Robotic Fish Using Antagonistic Stiffness in a Kinematically Singular Configuration». In: IEEE Transactions on Robotics 37.5 (2021), pp. 1712–1727. doi:10.1109/TRO.2021.3049430.

[7] Rolf Pfeifer, Alexandre Pitti, La révolution de l’intelligence du corps, 2012, Manuella Ed.

[8] A Abourachid, C Chevallereau, I Pelletan, P Wenger, An upright life, the postural stability of birds: a tensegrity system, Journal of the Royal Society Interface 20 (208), 20230433, 2023

[9] V Muralidharan, N Testard, C Chevallereau, A Abourachid, P Wenger, Variable stiffness and antagonist actuation for cable-driven manipulators inspired by the bird neck, Journal of Mechanisms and Robotics 15 (3), 035002, 2023

[10] NJS Testard, C Chevallereau, P Wenger, Dynamics and computed torque control stability of an under-actuated tendon-driven manipulator, IFToMM World Congress Mechanism and Machine Science, 332-341, 2023

[11] Annabi, L. Pitti, A. Quoy, M. (2021) Bidirectional interaction between visual and motor generative models using Predictive Coding and Active Inference, Neural Networks, 143, 638-656.

[12] Chen, X Pitti, A (2022) Visuo-Motor Remapping for 3D, 6D Reach and Tool -Use Reach using Gain-Field Networks, IEEE ICDL Epirob.