PhD student in Dynamics Modelling and Control of Modular Continuum Parallel Robots (M/F)

Job Description

Thesis Subject

Dynamics Modelling and Control of Modular Continuum Parallel Robots

Your Work Environment

The project RPC-JaM: The RPC-JaM project aims at advancing fundamentally the development of parallel continuum robots, a class of deformable robots obtained by assembling slender elastic legs in parallel. Their natural compliance enables them, to a certain degree, to adapt safely to their environment in case of contact, making them particularly interesting in tasks with sensitive environments such as medical interventions and cobotics. Despite all their advantages observed and demonstrated in the litterature, developing these robots comes with challenges such as the presence of singularities and elastic stability deficiency, models with high computation time, and a complex design. To solve these issues, we propose to follow an original approach where the robot is seen as the assembly of modular limbs, each one having its own sensors, actuators and processing units. The objective of this project is first of all to tackle the previous scientific challenges, eventually simplified as they are considered as the limb level and not the entire structure, with an additional work on how the assembled legs cooperate. Our goal is also to federate french research on this strongly emerging topic, and to disseminate the results to the public. Indeed, instead of buiding demonstrators for specific applications, we propose to valorize these results through the developpement of interactive art with one or several parallel continuum robots, which will be exposed, and to observe and analyse the people's reactions.

Context: Several types of controllers have been developed for continuum robots, most of the approaches being dedicated to serial-like robots. Many techniques are based on model-based kinematics control, the motions of the robots being relatively slow. Vision-based control has also been envisaged to avoid using inaccurate models and to increase the positioning accuracy of the end-effector. Position and kinematics controllers have also been developed for Continuum Parallel Robots (CPRs), see for instance. For robots having faster displacements, dynamics-model-based control can be envisaged. Such types of controllers have been first applied for robots having small link deformation and has been later extended to continuum robots, with serial or parallel architectures. These controllers exploit the information given by different sensors and state reconstruction techniques, on which significant research work and technology maturation have been conducted. Some sensors are classical for rigid robots, such as encoders and cameras, while others are specific to continuum structures such as fiber-bragg gratings and capacitive deformation sensors. All these techniques are centralized, and they are applied to robots whose entire (static or dynamic) model is known in advance. Thus they do not offer the necessary versatility of application when dealing with modular CPRs which, by essence, have models which may evolve when adding/removing/replacing a leg, or a platform, in the robotic system. In order to overcome this problem, a promising approach is to use decentralized control strategies. The coordinated and decentralized control of multi-robot system, for us the modular legs which are assembled altogether for creating a given CPR, is an enabling technology for a variety of applications. Indeed, these systems benefit from an increased robustness against system failures due to their ability to adapt to dynamic and uncertain environments. Such type of control is also well fitted when using low cost embedded hardware or when independent modules are used in parallel to achieve a given task. Recently, such kind of approach has been applied to the control of a flying parallel robot. In this paper it has been shown that, thanks to exchanges of information, three independent drones that are rigidly linked by a multibody mechanical system may collaborate in order to position a tool in space, and to apply external wrenches on the environment with it. First attempts of decentralized controllers have been proposed for soft robots, but are limited to serial architectures with simple pseudo-rigid or small deflection models. In RPC-JaM, due to the modular nature of the designed CPRs, we are going to apply decentralized strategies for controlling the parallel assembly with Cosserat rod models. They will rely on the information given by commercially available sensors with electronic readout distributed per leg to respect modularity constraints and sense the robot state with desired amplitude, resolution, and sampling frequency.

Tasks: The goal of this thesis is to first use the sensors information and controllers on each robot leg to calculate their dynamics model in a decentralized way. We will then be able to calculate the complete dynamic model of the CPR using a Lagrangian approach with multipliers to obtain a model linear with respect to the acceleration. We will finally adapt decentralized control techniques so that the on-board controllers associated with each leg can work together to perform the desired tasks (e.g. trajectory tracking, co-manipulation) while avoiding singularities.
For this, after a deep literature review, the student will first develop a dynamics simulator of the robot. It will serve testing the algorithms designed for control purpose, and to compare the simplified models that will be defined for being able to predict the robot dynamics in real time. This simulator should offer the possibility to simulate sensors and their measurement noise to be as close as possible from the real systems.
Then, we will develop efficient leg dynamics models. Computing this model needs an iterative root-finding process, which takes time. We believe that speeding up the computation of the robot dynamics equations could be achieved by measuring or observing with the embedded sensors for each leg: (a) the wrench applied at its extremity, and (b) its deformations state (leg shape and its time derivatives). If all these quantities are known, the leg dynamics model becomes an analytical equation in the motor input efforts, much simpler to solve. However, the number of sensors, and therefore the number of measured quantities, need to be minimized to facilitate the leg mechatronic design. A compromise between design cost & complexity, and model accuracy must be found.
Finally, we are going to develop trajectory planners and controllers for our CPRs. For the planners, offline and online strategies could be developed and eventually combined. For the offline strategies, the usual approach is to define a trajectory to be achieved by the robot platform and then to use the inverse dynamics model to estimate the corresponding motor displacements. The issue is that, during motion, we may meet robot singularities. However, the location of the singularities, or the easiness to cross them by mistake, thus encountering a stable to unstable transition effect for the robot, may change depending on the robot speed and acceleration. As such, their presence becomes difficult to detect. We may adapt approaches like in, especially if the users prefer the robots to avoid stable to unstable transitions.
We are also going to take advantage of the modularity of the robot to develop decentralized control approaches. Taking advantage of our experience in the control of drone fleets, we will adapt the decentralized controller to the case of our CPR, considered then as a fleet of JaM.

Environment: The thesis will take place in the team ARMEN of LS2N, on the campus of the Ecole Centrale of Nantes, in collaboration with the team RoMoCo of FEMTO-ST in Besançon. It will be cosupervised by Sébastien Briot (expert in dynamics modelling of parallel robots, continuous robots), Isabelle Fantoni (expert in decentralized control) and Kanty Rabenorosoa (expert in continuous robot design and modelling). The Phd student will have access to LS2N experimental facilities in order to make experiments that will validate the theory.

Constraints and risks

N/A

Compensation and benefits

Compensation

2300 € gross monthly

Annual leave and RTT

44 jours

Remote Working practice and compensation

Pratique et indemnisation du TT

Transport

Prise en charge à 75% du coût et forfait mobilité durable jusqu’à 300€

About the offer

About the CNRS

The CNRS is a major player in fundamental research on a global scale. The CNRS is the only French organization active in all scientific fields. Its unique position as a multi-specialist allows it to bring together different disciplines to address the most important challenges of the contemporary world, in connection with the actors of change.

CNRS

The research professions