CE10 - Usine du futur : Homme, organisation, technologies

Robot collaboratif – CoBot

CollaBorative roBot (CoBot)

The goal of the CoBot (Collaborative roBot) project is to enable humanoid robots to interact and collaborate safely with humans during handling tasks. <br />The project is in the axe « Interaction, Robotics - Artificial intelligence » of the challenge «Society of information and communication». The project matches gesture, movement, collaborative interactions, increased human, humanoid, adaptable systems, planning, cognitive architecture, autonomous decision, human-robot collaboration, mobility...

About the project CoBot

To date, the necessary safety conditions for collaboration between machine and man have not been met. Humans have a flexible architecture and their behaviour has been largely optimised in the course of evolution. Despite the technological progress of recent years, no robot can yet collaborate with humans in real time under safe conditions. The objective of the CoBot project is to study the collaboration strategies of two individuals during a load-carrying task and then to transfer the results to the interaction between a humanoid robot and a human. An intermediate simulation stage between the experimental part and the transfer to the robotic system will also be set up in order to study the behaviour of the different experimental modalities. This project is carried out in collaboration between three laboratories with complementary skills in biomechanics (LAAS, CRCA), neuroscience (DIMPS-IRISSE), motor control (DIMPS-IRISSE, LAAS, CRCA), motion simulation (LAAS-CRCA) and robotics (LAAS). In accordance with preliminary work integrating the three partners, this project aims in a first step to determine the biomechanical and physiological biomarkers of the optimization of the collaboration between individuals during this complex task as well as the muscular synergies between individuals (WP1). In a second step, we wish to simulate these behaviours by means of optimal or hierarchical control methods (WP2) and finally to transfer the results obtained to humanoid robots of the LAAS-CNRS (WP3). Thus, we have the opportunity to test on Pyrene, one of the most powerful robots of its generation, the validity of the control laws previously determined.

I. The experimental approach in human-human interaction
Preliminary studies were already conducted at CRCA to investigate the collective load carriage task. This work, in which two subjects had to walk with a load held in one hand (per subject) demonstrated the feasibility of the present protocol. Indeed, the present project requires studying the whole body of each subject and to record the reaction forces from the loads, which necessitates the acquisition of new sensors and of forces platforms. The analysis of the synergies, which are thought to be present at the muscular level, will necessitate the acquisition of EMG recording devices.
The methodology and knowledge, necessary to acquire and analyse these data, are well knwon by the members of the team, as evidenced by their previous works (Moretto et al. 2016, Watier et al, 2017, Turpin et al., 2017).
II. Simulation of multibody dynamics
A challenging problem here is to develop a simulation that considers a multi-contact problem when the robot interacts with a partner through the contact created by the load and adjusts the ground reactions to maintain balance and to move. Intense researches are presently performed by members of the LAAS-CNRS on this topic and promising results were already published. The difficulty here is to consider the unknown variation of the forces applied by the subject on the load that will be transferred to the humanoid robot. The model of the force’s variation applied on the robot will be important and could lead to succeed or not in transferring the motion to the robots.
III. Transfer to humanoid robots
The final goal of the project is to control a real robot to make it collaborate with a human being during a handling task. This challenging problem will necessitate very efficient real time calculus to adapt balance and stability of the biped when he is constrained by external and variable forces.

The funding obtained made it possible to equip the experimentation lab located at the CREPS in Toulouse, that of the IRISSE laboratory in La Réunion and the LAAS-CNRS in Toulouse with three force platforms and two Sensix cylindrical force sensors. In La Réunion, a Vicon camera was added to the system already in place, as well as an electromyographic (EMG) acquisition system. The first work consisted in setting up the complex experimental protocol of the study. The experiments could finally be carried out recently. The kinematic, kinetic and electrophysiological data have been acquired and are available on the project's dedicated website.
The simulation was able to benefit from a favourable environment. First of all, we were able to process and analyse experimental data on the locomotion of human subjects specifically carried out at LAAS-CNRS. The inverse optimal control procedure was carried out to determine the invariants of human movement and then to simulate them. In addition, we compared the optimal control simulations to clothoids. Using the Pyrene robot's gait planner we were then able to simulate the robot's locomotor trajectories in order to adapt to human behaviour in real time. This work has been widely communicated and disseminated.
The work on motion generation of the Pyrene robot has also advanced considerably. First of all, we were able to synchronise the robot with the motion capture system, allowing it to locate itself in relation to the individuals and the object to be manipulated. (https://www.youtube.com/watch?v=GSx3bYOrFAI). The simulations of the locomotor trajectories obtained were transferred to the robot's walking pattern generator, which then generated the robot's step positions that respected the desired trajectory in real time. Motion generations with the robot following the clothoid-simulated trajectories have already been generated.

Currently the work of the CoBot project is proceeding at an extremely dynamic pace. Firstly, on the experimental aspects, the inverse dynamics work will allow us to obtain more data for the inverse optimal control algorithms. At this stage, these have only been performed on kinematic data of spontaneous locomotor trajectories. These new results will allow a greater consideration of subject dynamics and interaction forces between subjects. In parallel, important work on muscle synergies should allow a better understanding of the motor pattern during this interaction work. This is an area that has been little explored in the literature. This work will also allow the implementation of innovative processes such as UCM (Uncontrolled Manifold) for the hierarchical control of robots.

Concerning the simulation by optimal control, the work is finalised by carrying out the inverse optimal control on 2 subjects interacting during the table transport. Different scenarios are simulated: robot in front or behind the table to be transported with or without knowledge of the end point. Both leader and follower roles are studied based on the analysis and processing of experimental data. Simulation work based on hierarchical control will be progressively implemented according to the results of experimental data processing.

Finally, important work is in progress concerning the locomotion of the Pyrene robot. Currently, the motion generations have used the walking pattern generator provided by the PAL-Robotics company, which is a walk generator in position. The work in progress concerns the development of a torque controller that will allow greater adaptation to external mechanical actions by providing greater flexibility and ultimately greater safety during interaction.

1 - I. Maroger et al. Human Trajectory Prediction Model and its Coupling with a Walking Pattern Generator of a Humanoid Robot. IEEE/RSJ IROS. 2021, Prague, Czech R.
2 - N. Sghaier et al. (2021) Load Sharing during Team Lifting, CMBBE, Accepté. Congrès de la Société de Biomécanique. St Etienne.
3 - I. Maroger et al. Walking Human Trajectory Models and Their Application to Humanoid Robot Locomotion. IEEE/RSJ IROS, 2020, Las Vegas, US. 10.1109/IROS45743.2020.9341118
4 - M. Boukheddimi et al. Human-like gait generation from a reduced set of tasks using the hierarchical control framework from robotics. Proceeding of the IEEE ROBIO. Dali, China, 2019. 10.1109/ROBIO49542.2019.8961426
5 - M. Boukheddimi et al. Anthropomorphic Gait Generation using Differential Dynamic Programming with a Reduced Number of Cost Criteria, 2020 8th IEEE RAS/EMBS BIOROB, 1036-1042, doi: 10.1109/BioRob49111.2020.9224427.
6 - I. Maroger et al. Description and Assessment of a Human Trajectory Prediction Model during Gait. Accepté. Congrès de la Société de Biomécanique. St Etienne.
1 - N. Sghaier et al. (2020) 3D distribution of the forces applied on a load transported by a dyad, CMBBE, 23:sup1, S282-S284, DOI: 10.1080/10255842.2020.1816293
2 - I. Maroger et al. (2020) Comparison of human experimental trajectories and simulations during gait. CMBBE, 23:sup1, S189-S191, DOI: 10.1080/10255842.2020.1813421
3 - I. Maroger et al.(2021). Human Trajectory Prediction Model and its Coupling with a Walking Pattern Generator of a Humanoid Robot. IEEE RA-L. 10.1109/LRA.2021.3092750
4- Al Abiad et al. (2020). A Mechanical Descriptor of Instability in Human Locomotion: Experimental Findings in Control Subjects and People with Transfemoral Amputation. Appl. Sci., 10, 840. DOI. 10.3390/app10030840
5 - M. Marchitto et al. (2020). Gait analysis comparison of two twins: one healthy and one with spastic cerebral palsy. CMBBE, 23:sup1, S186-S188, DOI: 10.1080/10255842.2020.1816301

Robots appear nowadays as a solution to help humans in the factory, at the hospital or at home but the conditions of efficient secure interactions are not understood yet. Humans are soft, flexible and their behaviours have been optimised through learning mechanisms and through evolution. Despite great advances in robotics, no robots can be involved in secured collaborative load-carriage tasks with humans. At best, the robot is able to follow or to drive the manoeuvre, but in the first case, the robot is no more than a wheelbarrow (but is much more expensive) and in the second case, either the human does not accept to be driven or he has to compensate for the lack of reactivity and manoeuvrability of the robot. To avoid these problems, we must find a way for robots to collaborate safely and in real time with humans.Thus, a major challenge facing roboticists is to developed anthropomorphic systems able to interact safety with humans.

Biomechanicists, especially those interested in human movement try to understand the mechanical parameters that underlie motor control in humans. Over the past decade, swaps between both specialities strengthen exponentially around these issues of modelling, simulations and generation of human motions. Robotics brings powerful tools to compute dynamic equations of treelike poly-articulated systems and researchers in human movement analysis bring key elements for the mechanical parameters which must be taken into account to reduce the complexity of the musculoskeletal system. It’s precisely this whole process of experimentation on the one hand and simulation on the other hand with a steady back and forth between the experts in the field that we wish to undertake here.

We designed this project to explore real-time collaboration strategies used in a load-carriage task performed by two human subjects and to transfer the results to a human-humanoid working pair. As an intermediate step, we will use simulations to test our assumptions regarding the optimization of their collaborative behaviours. This project will bring together three teams with complementary skills in biomechanics (LAAS, CRCA), neuroscience (DIMPS-IRISSE), motor control and collective behavior (DIMPS-IRISSE, LAAS, CRCA), movement simulation (LAAS-CRCA) and robotics (LAAS).

Thanks to high level preliminary works involving the three partners and several co-authored publications, we have the opportunity to collaborate to 1) determine the biomechanical and physiological biomarkers of the movement optimization (i.e., which parameter are optimized) and determine the synergies and collaboration functions, 2) simulate the collective load carriage task using a human-robot model by implementing the cost-functions and synergies previously determined and 3) apply these rules to drive a real humanoid robot. Here we have the opportunity to use Pyrène, the first humanoid robot integrating high power actuators.

Financial grants are request to complete the present experimental setup and conduct an exhaustive full body analysis of the two subjects carrying a load together. To insure a good interdisciplinary, 2 PhD and 3 Master will be recruited and co-supervised. The total amount requested is 372k€ and represents 35% of the total amount dedicated to this project.

Project coordinator

Monsieur Bruno Watier (Laboratoire d'analyse et d'architecture des systèmes)

The author of this summary is the project coordinator, who is responsible for the content of this summary. The ANR declines any responsibility as for its contents.


LAAS-CNRS Laboratoire d'analyse et d'architecture des systèmes

Help of the ANR 372,434 euros
Beginning and duration of the scientific project: March 2019 - 42 Months

Useful links

Sign up for the latest news:
Subscribe to our newsletter