2025​​Activity reportProject-TeamAUCTUS​​​‌

3.1.1‌​‌ Scientific Context

The purpose​​ of this axis is​​​‌ to provide metrics to‌ assess human behavior. Our‌​‌ human study specifically focuses​​ on industrial operators. We​​​‌ assume the following working‌ hypotheses: the operator's task‌​‌ and environmental conditions are​​ known and circumscribed; the​​​‌ operator is trained in‌ the task, production tools‌​‌ and safety instructions; the​​ task is repeated with​​​‌ more or less frequent‌ intervals. We aim to‌​‌ analyze the following:

• the​​ physical and cognitive fragility​​​‌ of operators to meet‌ assistance needs;
• cognitive biases‌​‌ and physical constraints leading​​ to a loss of​​​‌ the operator safety;
• ergonomics,‌ performance and acceptance of‌​‌ the production tool.

In​​ the industrial context, these​​​‌ questions are tackled through‌ the fields of work‌​‌ ergonomics and cognitive sciences.​​ Four main axes are​​​‌ typically addressed: physiological/biomechanical, cognitive,‌ psychological and sociological studies.‌​‌ We particularly focus on​​ the biomechanical, cognitive and​​​‌ psychological aspects, as described‌ by the ANACT 23‌​‌, 24. The​​ aim is to translate​​​‌ these factors into metrics,‌ optimality criteria or constraints‌​‌ to better analyze, design​​ and control the collaborative​​​‌ robots.

A review of‌ ergonomic workstation evaluation helps‌​‌ in positioning our desired​​ contributions in robotics. Ergonomists​​​‌ evaluate the gesture through‌ the observation of the‌​‌ workstations and, generally, through​​ questionnaires. This requires long​​​‌ periods of field observation,‌ followed by analyses based‌​‌ on ergonomic grids (e.g.​​ RULA 40, REBA​​​‌ 31, LUBA 35‌, OWAS 34,‌​‌ ROSA 54,...). Until​​ then, the use of​​​‌ more complex measurement systems‌ was reserved for laboratories,‌​‌ particularly in biomechanical studies.​​ The advent of cost-effective​​​‌ and minimally intrusive motion‌ capture sensors (Inertial Measurement‌​‌ Units, RGB-D cameras), coupled​​ with advancements in computer​​​‌ vision algorithms, now enables‌ on-site data collection to‌​‌ assess workers' gestures, postures,​​ movements, and physiological states.​​​‌ Some of these systems‌ can be permanently integrated‌​‌ into production lines without​​ disrupting workflow, while simultaneously​​​‌ evaluating worker well-being through‌ biomechanical and physiological metrics.‌​‌ This facilitates longitudinal motion​​ capture studies, similar to​​​‌ the evolution of maintenance‌ practices from reactive to‌​‌ predictive.

Ergonomic robotics has​​ recently taken an interest​​​‌ in this new evaluation‌ paradigm to adapt the‌​‌ robot behavior to reduce​​ ergonomic risks. This ergonomic​​​‌ adaptation complements the conventional‌ approaches that only consider‌​‌ the performance of the​​ action produced by the​​​‌ human in interaction with‌ the robot. However, ergonomic‌​‌ criteria are usually based​​​‌ on the principle that​ the comfort positions are​‌ distant from the human​​ joint limits. These notations​​​‌ are compatible with an​ observation of the human​‌ operator through the eye​​ of the ergonomist. In​​​‌ practice, such evaluations are​ inaccurate and subjective 57​‌. Moreover, they only​​ consider quasi-static human positions,​​​‌ without taking into account​ the evolution of the​‌ person's physical, physiological and​​ psychological state. We aim​​​‌ to extend this approach​ to more reliable and​‌ comprehensive ergonomic metrics with​​ musculoskeletal, fatigue, metabolic consumption​​​‌ modelling. The repetition of​ gestures, and the solicitation​‌ of muscles and joints,​​ are questions that must​​​‌ complete these analyses. A​ method used by ergonomists​‌ to limit biomechanical exposures​​ is to increase variations​​​‌ in motor stress by​ rotating tasks 55.​‌ However, this type of​​ extrinsic method is not​​​‌ always suitable in the​ industrial context 39 where​‌ we place our research​​ efforts.

Through these human​​​‌ analyses, the Auctus team​ aims to revise the​‌ use of collaborative robots​​ in the workplace to​​​‌ vary the operator's environment​ and encourage more appropriate​‌ motor strategies. Biomechanical studies​​ of the intrinsic variability​​​‌ of the motor system​ allowed by the joint​‌ redundancy of the human​​ body should result in​​​‌ the alternation of postures,​ movements and muscle activity​‌ observed in the individual​​ to perform a requested​​​‌ task 55. This​ variation leads to differences​‌ between the motor coordination​​ used by each operator,​​​‌ and conveys the notion​ of motor strategy 32​‌. We aim to​​ provide exhaustive studies of​​​‌ motor strategies in industrial​ setups.

The cognitive dimension​‌ of ergonomics must also​​ be addressed in our​​​‌ approach to reduce the​ mental workload and foster​‌ the wellness of the​​ worker. We believe that​​​‌ known sensorimotor strategies can​ be a physically quantifiable​‌ reflection of the operator's​​ cognitive state. For example,​​​‌ human motion measurements can​ be used to predict​‌ fatigue 52 and, therefore,​​ adjust the robot's assistance.​​​‌ A key challenge here​ is to better analyze​‌ human manual expertise (dexterous​​ and cognitive) to adapt​​​‌ the human-robot interaction. The​ expertise embodies the operators'​‌ decision-making process while perceiving,​​ understanding, and anticipating their​​​‌ gestures to preserve their​ safety, comfort, and performance​‌ in the task. We​​ aim to adapt and​​​‌ refine known human cognitive​ models (multisensory perception 29​‌, situation awareness 28​​) to infer the​​​‌ influence of the task​ context and environment on​‌ operator behavior.

3.1.2 Methodology​​

How can we observe,​​​‌ understand and quantify human​ sensorimotor and cognitive strategies​‌ to better design and​​ control the behavior of​​​‌ the cobotic assistant?

When​ we study systems of​‌ equations (kinematic, static, dynamic,​​ musculoskeletal, etc.) to model​​​‌ human behavior, several problems​ appear and explain our​‌ methodological choices:

• the large​​ dimension of the problems​​​‌ to be considered, due​ to the human body​‌ complexity (eg. joint, muscle​​ redundancy);
• the distribution of​​​‌ anthropometries, force capacities, and​ fatigue resistance among workers,​‌ for example, size, the​​ muscle mass, the possible​​​‌ geometric configurations on a​ workstation (set of possible​‌ trajectories, postures, and placements​​ of the operator), and​​ the forces needed to​​​‌ be produce to execute‌ a task;
• the uncertainties‌​‌ in the measurement and​​ the model approximations.

The​​​‌ idea is to start‌ with a description of‌​‌ redundant workspaces (geometric, static,​​ dynamic...). We use set-theory​​​‌ approaches, based on interval‌ analysis 56, 45‌​‌, which meet the​​ variability requirement and cope​​​‌ with model and measurement‌ uncertainties. Another advantage of‌​‌ such techniques is that​​ they allow the results​​​‌ to be certified, which‌ is essential to address‌​‌ safety issues. Some members​​ of the team have​​​‌ already achieved success in‌ mechanical design for performance‌​‌ certification and robot design​​ 42. By extending​​​‌ these set-theory approaches to‌ our problem, a mapping‌​‌ of ergonomic, efficient, and​​ safe movements can be​​​‌ obtained, in which we‌ project the operators' motor‌​‌ strategies. Biomechanical, ergonomic, and​​ cognitive metrics can, then,​​​‌ be defined and evaluated‌ to quantify the human‌​‌ behavior in specific work​​ situations.

It is therefore​​​‌ necessary to:

• model human‌ capabilities, both at the‌​‌ musculoskeletal and the perceptive/cognitive​​ levels, to allow for​​​‌ global, yet detailed, analyses‌ as well as efficient‌​‌ integration of such knowledge​​ in the control of​​​‌ the collaborative robots.
• contribute‌ to accurate representations of‌​‌ the shoulder joint involved​​ in most cobotic interactions​​​‌ and worker efforts. Its‌ complex range of motion‌​‌ and the numerous muscles​​ involved make proper shoulder​​​‌ modeling a significant challenge‌ for musculoskeletal (MSK) models‌​‌ 8.
• propose new​​ ergonomic, biomechanical, robotic, and​​​‌ cognitive indices that will‌ link different types of‌​‌ performances while taking into​​ account the influence of​​​‌ fatigue, stress, level of‌ expertise, etc.;
• divide the‌​‌ task and the gesture​​ into homogeneous phases: this​​​‌ process is complex and‌ depends on the type‌​‌ of studied index and​​ the techniques being used.​​​‌ We are exploring several‌ methods: inverse optimal control,‌​‌ learning methods, techniques from​​ signal processing;
• develop interval​​​‌ extensions of the proposed‌ indices. The indices are‌​‌ not necessarily the result​​ of a direct model,​​​‌ and algorithms must be‌ developed or adapted to‌​‌ compute them (calculation of​​ manipulability, Uncontrolled Manifold, etc.);​​​‌
• Aggregate proposals into a‌ dedicated interval-analysis library for‌​‌ human behavior studies (use​​ of and contribution to​​​‌ the existing ALIAS-Inria and‌ the open-source IBEX library).‌​‌

The major contribution of​​ the methodology is to​​​‌ embrace in the same‌ model the measurement uncertainties‌​‌ (important for on-site use​​ of measurement equipment), the​​​‌ variability of tasks and‌ trajectories (proper to dexterous‌​‌ industrial operations), and the​​ physiological characteristics of the​​​‌ operators (critical adaptability to‌ every individual). The originality‌​‌ of the approach is​​ to combine biomechanical, ergonomic​​​‌ and cognitive metrics with‌ the usual performance indices‌​‌ to build a comprehensive​​ and objective analysis of​​​‌ human behavior.

Other avenues‌ of research are being‌​‌ explored, particularly around the​​ inverse optimal control 46​​​‌, to project human‌ movements based on the‌​‌ performance or ergonomic indices.​​ Such a projection would​​​‌ offer new interpretations and‌ enhance the analysis of‌​‌ human behaviors.

Finally, the​​ development of a population-scale​​​‌ musculoskeletal model database and‌ the application of anthropometry-robust‌​‌ optimal control represents a​​​‌ new aera of research​ that the team will​‌ explore by 2025. This​​ approach would further explore​​​‌ the inherent variability within​ human workforces. By constructing​‌ a diverse virtual population​​ through statistical and biomechanical​​​‌ principles, we will be​ able to leverage predictive​‌ simulation to identify broadly​​ applicable movement strategies and​​​‌ minimize potential strain across​ a spectrum of worker​‌ morphologies and capabilities. This​​ dual approach moves beyond​​​‌ traditional, one-size-fits-all ergonomic assessments,​ enabling the development of​‌ personalized or, at minimum,​​ broadly robust recommendations for​​​‌ safer and more efficient​ work practices.

3.2.1 Human-Robot interaction​

The first step to​‌ couple the operator and​​ cobot together at work​​​‌ is to provide interaction​ modalities through which the​‌ agents can communicate and​​ coordinate. The interaction can​​​‌ be direct, where the​ robot and operator act​‌ together in the same​​ shared environment, or the​​​‌ operator can remotely perform​ the task with the​‌ robot through a teleoperation​​ system (which reflects the​​​‌ remote interaction and potentially​ corrects for punctual weaknesses).​‌ The level and type​​ of human-robot interactions are​​​‌ chosen with respect to​ the task, the context,​‌ or other human factors.​​ The challenge is, then,​​​‌ to predict the joint​ human-robot behavior and capabilities​‌ for each interaction situation​​ and collaborative context.

The​​​‌ formal computation of joint​ human-robot capabilities can be​‌ given thanks to the​​ models and evaluation indices​​​‌ presented in Axis 1.​ We can focus on​‌ quantifying how the interaction​​ with the robot will​​​‌ impact the human sensorimotor​ strategy (changes in the​‌ posture, positions, forces, etc.,​​ induced by the robot)​​​‌ and recomputing metrics such​ as human fatigue and​‌ motor variability51 and​​ the Mover project. We​​​‌ can further use the​ biomechanical and robotic models​‌ to consider a unified​​ operator-robot entity and to​​​‌ compute their joint abilities​ (e.g. common human-robot force​‌ capabilities 53).

Developing​​ human cognitive and sensorimotor​​​‌ models to account for​ the effect of the​‌ human-robot interaction could provide​​ a valuable tool to​​​‌ evaluate cobotic systems and​ collaborative works. However, the​‌ accuracy of these models​​ must be addressed. We​​​‌ wish to understand how​ the robot influences the​‌ operator's work, and thus​​ how his mental model​​​‌ of the task evolves​ according to the interactions​‌ with the robot. The​​ challenge is, then, to​​​‌ predict the behavior of​ the operator that takes​‌ into account his cognition​​ in the interaction situations.​​​‌ Preliminary literature results have​ shown that key cognitive​‌ mechanisms in human teaming​​ may transfer to human-robot​​​‌ collaboration, such as joint​ human-robot action representation 26​‌ or coordination mechanisms 44​​. However, the situation​​​‌ awareness of the operator​ is modified by the​‌ interaction with the robot​​ 49. Developing a​​ joint mental model that​​​‌ accurately captures the human-robot‌ interaction can later guide‌​‌ the design of relevant​​ interaction modalities to improve​​​‌ the team's understanding 27‌.

3.2.2 Cobot adaptive‌​‌ assistance

Taking into account​​ the coupling between the​​​‌ operator and the robot‌ at the control level‌​‌ is also central to​​ the team's objectives. We​​​‌ wish to demonstrate how‌ a collaborative robot can‌​‌ be used to mediate​​ between a control objective​​​‌ that optimizes task performances,‌ safety and comfort (what‌​‌ we consider as the​​ expertise trinity), on​​​‌ one hand, and the‌ action model of the‌​‌ human interacting with the​​ robot (the inferred human​​​‌ intent), on the other‌ hand. Such an arbitration‌​‌ in the control law​​ adapts the robot's assistive​​​‌ behavior to better collaborate‌ with the operator. This‌​‌ shared-autonomy concept is the​​ focus of part of​​​‌ our research. It can‌ range from a discrete‌​‌ task allocation between the​​ agents to an effectively​​​‌ shared task 43.‌

We are strongly confident‌​‌ that the notion of​​ expertise is central to​​​‌ adjusting the cobot behavior.‌ The robot controller is‌​‌ designed to increase the​​ level of expertise in​​​‌ the operator/robot team: it‌ optimizes the human-centered metrics‌​‌ (safety criteria, biomechanical and​​ cognitive comfort, etc.) and​​​‌ provides a gain in‌ performances (joint human-robot capabilities).‌​‌ But it also aims​​ at preserving the operator's​​​‌ particular expertise and know-how‌ at the center of‌​‌ the activity. Manual expertise​​ of highly skilled operators​​​‌ needs to be analyzed‌ respectively on its dynamic‌​‌ aspects and on the​​ ability to synchronize with​​​‌ other operators in the‌ environment. Understanding better the‌​‌ expertise is envisioned as​​ a way to alleviate​​​‌ the operators of repetitive‌ and easy operations while‌​‌ maintaining their ability to​​ perform expert gestures based​​​‌ on the complexity of‌ the task.

Furthermore, this‌​‌ research axis raises the​​ question of the modification​​​‌ of the work induced‌ by collaborative robots for‌​‌ expert operators. While the​​ overall goal is to​​​‌ make use of robots‌ to punctually or continuously‌​‌ improve the work conditions​​ of these operators (and​​​‌ not to replace them),‌ the presence of these‌​‌ robots necessarily impacts the​​ work referential and thus​​​‌ the expertise itself. One‌ of the central questions,‌​‌ yet to be tackled,​​ relates to the original​​​‌ and core part of‌ the expertise that should‌​‌ remain unchanged. The proposed​​ modeling of the operator/robot​​​‌ coupling and interaction is‌ the first avenue to‌​‌ predict possible changes in​​ the expertise. It can​​​‌ be input to the‌ controller to constraint the‌​‌ robot to let the​​ operator make the expert​​​‌ decisions naturally.

3.3.1‌​‌ Architectural design

Is it​​ necessary to cobotize, robotize​​​‌ or assist the human‌ being? Which mechanical architecture‌​‌ meets the task challenges​​ (a serial cobot, a​​​‌ specific mechanism, an exoskeleton)?‌ What type of interaction‌​‌ (H/R cohabitation, comanipulation, teleoperation)?​​ These questions are the​​​‌ first requests from our‌ industrial partners. For the‌​‌ moment, we have few​​ comprehensive methodological answers to​​​‌ these requests. Choosing a‌ collaborative robot architecture is‌​‌ a difficult problem 36​​​‌. It becomes even​ more complex when the​‌ architectural design is approached​​ from concurrent cognitive ergonomic,​​​‌ biomechanical and robotic perspectives.​ There are major methodological​‌ and conceptual differences in​​ these areas. It is,​​​‌ therefore, necessary to bridge​ these representational gaps and​‌ to propose a global​​ and generic approach that​​​‌ takes into consideration the​ expectations of the robotician​‌ to model and formalize​​ the general properties of​​​‌ a cobotic system as​ well as those of​‌ the ergonomist to define​​ the expectations in terms​​​‌ of an assistance tool.​

To do this, we​‌ propose a user-centered design​​ approach, with a particular​​​‌ focus on human-system interactions.​ From a methodological point​‌ of view, this requires​​ to develop a structured​​​‌ experimental approach. It aims​ at characterizing the task​‌ to be carried out​​ ("system" analysis) but also​​​‌ at capturing the physical​ markers of its realization​‌ (required movements and efforts,​​ ergonomic stress, etc.). This​​​‌ specification must be done​ through the prism of​‌ a systematic study of​​ the exchanged information (type​​​‌ and modality) needed by​ humans to perform the​‌ considered task. Based on​​ these analyses, the main​​​‌ challenge is to define​ a decision support tool​‌ for the choice of​​ the robotic architecture and​​​‌ the specifications of the​ role assigned to the​‌ robot and the operator​​ as well as their​​​‌ interactions.

The evolution of​ the chosen methodology is​‌ for the moment empirical,​​ based on the user​​​‌ cases regularly treated in​ the team (see sections​‌ on contracts and partnerships).​​

The process can be​​​‌ summarized through the following​ steps:

• identify expert or​‌ difficult jobs on industrial​​ sites. This is done​​​‌ through visits and exchanges​ with our partners (manager,​‌ production manager, ergonomist...);
• select​​ some challenging use cases​​​‌ to be studied and,​ then, observe the operator​‌ in its ecological environment.​​ Our motion capture tools​​​‌ allow us to produce​ a force-motion analysis, based​‌ on previously defined ergonomic​​ criteria, and a physical​​​‌ evaluation of the task​ in terms of expected​‌ performances, from both experiments​​ and simulations. In parallel,​​​‌ an evaluation of the​ operator's expertise and cognitive​‌ strategy is throught questionnaires;​​
• synthesize these observatory results​​​‌ into design requirements to​ deduce: the robotic architectures​‌ to be initiated, the​​ key points of human-robot​​​‌ interaction to be developed,​ and the difficulties in​‌ terms of human factors​​ to be taken into​​​‌ account.

The different human​ and task analyses take​‌ advantage of the expertise​​ available within the AUCTUS​​​‌ team. The team has​ already worked on the​‌ currently dominant approach: the​​ use of human models​​​‌ to design the cobotic​ cell through virtual tools.​‌ We would like to​​ gradually introduce the additional​​​‌ evaluation criteria presented above.​ However, the very large​‌ dimensions of the treated​​ problems (modeling of the​​​‌ body's degrees of freedom​ and the constraints applied​‌ to it) makes it​​ difficult to carry out​​​‌ a certified analysis. We​ choose to go through​‌ the calculation of the​​ human workspace and performances,​​​‌ which is not yet​ done in this field.​‌ The idea here is​​ to apply set theory​​ approaches, using interval analysis​​​‌ as already discussed in‌ section 3.1.2. The‌​‌ goal is, then, to​​ extend the human constraints​​​‌ to intervals, which integrate‌ the model variability, and‌​‌ to play them in​​ virtual reality during simulations​​​‌ of the tasks. This‌ would allow the operator‌​‌ to check his trajectories​​ and scenarios not only​​​‌ for a single case‌ study but also for‌​‌ sets of cases. For​​ example, it can be​​​‌ verified that, regardless of‌ the bounded sets of‌​‌ simulated operator physiological capabilities,​​ the physical constraints of​​​‌ a simulated trajectory are‌ not violated. Therefore, the‌​‌ assisted design tools certify​​ cases of use as​​​‌ a whole. Moreover, the‌ intersection between the human‌​‌ and robot workspaces/capabilities provides​​ the necessary constraints to​​​‌ certify the feasibility of‌ a task in the‌​‌ interaction situation. Overall, integrating​​ human and task-related physical​​​‌ constraints in the design‌ process brings to better‌​‌ cobotic systems. In the​​ future, we will similarly​​​‌ develop tools to include‌ human cognitive markers in‌​‌ the design approach.

This​​ research line merges the​​​‌ contributions of the other‌ axes, from the analysis‌​‌ of the human behavior​​ and capabilities in its​​​‌ environment for an identified‌ task, the prediction of‌​‌ the effects of the​​ interaction/coupling strategy with the​​​‌ robot, to the choice‌ of a mechanical architecture‌​‌ from the resulting design​​ constraints. The proposed task-oriented​​​‌ and human-centered methodology is‌ perfectly integrated into an‌​‌ Appropriate Design approach. It​​ can be used for​​​‌ the dimensional design and‌ optimization of robots, again‌​‌ based on interval analysis.​​ The challenges are the​​​‌ change of scale in‌ models that symbiotically consider‌​‌ the human-robot pair, the​​ uncertain, flexible and uncontrollable​​​‌ nature of human behavior,‌ and the many evaluation‌​‌ indices needed to describe​​ them.

It is worth​​​‌ noting that we aim‌ to develop a global‌​‌ mechatronic design approach, which​​ would build upon the​​​‌ design constraints to specify‌ the robot hardware and‌​‌ controller at once. The​​ chosen set-theory computational methodology​​​‌ is particularly appropriate to‌ meet this objective since‌​‌ the interval-based representation of​​ the design constraints can​​​‌ be directly and equally‌ used to set the‌​‌ control constraints.

3.3.2 Control​​ design

The control laws​​​‌ of collaborative robots from‌ the major robot manufacturers‌​‌ differ little or not​​ at all from the​​​‌ existing control laws in‌ the field of conventional‌​‌ industrial robotics. Security is​​ managed a posteriori, as​​​‌ an exception, by a‌ security PLC / PC.‌​‌ It is therefore not​​ an intrinsic property of​​​‌ the controller. This strongly‌ restricts possible physical interactions‌​‌ 1 and leads to​​ suboptimal operation of the​​​‌ robotic system. It is‌ difficult in this context‌​‌ to envision tangible human-robot​​ collaboration. Collaborative operation requires,​​​‌ in this case, a‌ control calculation that integrates‌​‌ safety and ergonomics as​​ a priori constraints.

The​​​‌ control of truly collaborative‌ robots in an industrial‌​‌ context is, from our​​ point of view, underpinned​​​‌ by two main issues.‌ The first one is‌​‌ related to the macroscopic​​ adaptation of the robot's​​​‌ behavior according to the‌ phases of the production‌​‌ process. The second one​​​‌ is related to the​ fine adaptation of the​‌ degree and/or nature of​​ the robot's assistance according​​​‌ to the ergonomic state​ of the operator. If​‌ this second problem is​​ part of a historical​​​‌ dynamics in robotics that​ consists in placing safety​‌ constraints, particularly those related​​ to the presence of​​​‌ a human being, at​ the heart of the​‌ control problem 3041​​, 33, it​​​‌ is not approached from​ the more subtle point​‌ of view of ergonomics​​ where the objective cannot​​​‌ be translated only in​ terms of human life​‌ or death, but rather​​ in terms of long-term​​​‌ respect for their physical​ and mental integrity. Thus,​‌ the simple and progressive​​ adoption by a human​​​‌ operator of the collaborative​ robot intended to assist​‌ him in his gesture​​ requires a self-adaptation in​​​‌ the time of the​ command. This self-adaptation is​‌ a fairly new subject​​ in the literature 47​​​‌, 48.

At​ the macroscopic level, the​‌ task plan to be​​ performed for a given​​​‌ industrial operation can be​ represented by a finite-state​‌ machine. To avoid increasing​​ the human's cognitive load​​​‌ by explicitly asking him​ to manage transitions for​‌ the robot, we propose​​ to develop a decision​​​‌ algorithm that would ensure​ discrete transitions from one​‌ task (and the associated​​ assistance mode) to another​​​‌ based on an online​ estimate of the current​‌ state of the human-robot​​ couple. The associated scientific​​​‌ challenge is to establish​ a link between the​‌ robot's involvement and a​​ given working situation. We​​​‌ propose an incremental approach​ to learn this complex​‌ relationship. Its first stage​​ will consist of identifying​​​‌ the general and relevant​ control variables to conduct​‌ this learning in an​​ efficient and reusable way,​​​‌ regardless of the particular​ calculation method of the​‌ control action. Then, physically​​ realistic simulations and real-world​​​‌ experiments will be used​ to feed this learning​‌ process.

To handle mode​​ transitions, we propose to​​​‌ explore the richness of​ the multi-tasking control formalism​‌ under constraints 38.​​ It would ensure a​​​‌ continuous transition from one​ control mode to another​‌ while guaranteeing compliance with​​ a certain number of​​​‌ robot control constraints. Some​ of these constraints convey​‌ the ergonomic specifications and​​ are dependent on the​​​‌ state of the robot​ and of the human​‌ operator, which, by nature,​​ is difficult to predict​​​‌ accurately. We propose, again,​ to exploit the interval-analysis​‌ paradigm to efficiently formulate​​ ergonomic constraints robust to​​​‌ the various existing uncertainties.​

Purely discrete or reactive​‌ adaptation of the control​​ law would make no​​​‌ sense given the slow​ dynamics of certain physiological​‌ phenomena such as fatigue.​​ Thus, we propose to​​​‌ formulate the control problem​ as a predictive problem​‌ where the impact of​​ the control decision at​​​‌ a time $t$ is​ anticipated at different time​‌ horizons. This requires a​​ prediction of human movement​​​‌ and knowledge of the​ motor variability strategies it​‌ employs. This prediction is​​ possible based on the​​​‌ supervision at all times​ of the operational objectives​‌ (task in progress) in​​ the short term. However,​​ it requires the use​​​‌ of a virtual human‌ model and possibly a‌​‌ dynamic simulation to quantify​​ the impact of these​​​‌ potential movements in terms‌ of performances, including ergonomics.‌​‌ It is impractical to​​ use a predictive command​​​‌ with simulation in the‌ loop with an advanced‌​‌ virtual manikin model. We,​​ therefore, suggest adapting the​​​‌ prediction horizon and the‌ complexity of the corresponding‌​‌ model to guarantee a​​ reasonable computational complexity.

7.1.1 Qontrol

• Name:​​​‌
  Quadratic Optimization coNTROL
• Keywords:​
  Robotics, Control, Optimisation
• Scientific​‌ Description:
  Generic expression of​​ a robot control problem​​​‌ using constrained optimization in​ a dynamic environment, applicable​‌ to robots that can​​ be controlled in terms​​​‌ of torque, speed or​ position.
• Functional Description:
  Qontrol​‌ is a tool for​​ the generic formulation of​​​‌ robotic control problems in​ the form of constrained​‌ optimization problems. It is​​ initially intended for fixed-base​​​‌ polyarticulated robots. It allows​ to easily create tasks​‌ and constraints in the​​ control law.
• News of​​​‌ the Year:
  Declaration to​ the Program protection office.​‌ Appears in "transfert challenge"​​ kind of project (awaiting​​​‌ for finding) Appears in​ a european project (awaiting​‌ for finding)
• URL:
  https://­auctus-team.­gitlabpages.­inria.­fr/­components/­control/­qontrol/­index.­html​​
• Contact:
  Lucas Joseph
• Participants:​​​‌
  Lucas Joseph, Vincent Padois​

7.1.2 pin-actuation

• Keywords:
  Biomechanics,​‌ Robotics, Python
• Functional Description:​​
  Pin-actuation is a Python​​​‌ library for adding actuation​ models to Pinocchio multibody​‌ dynamics models. It extends​​ Pinocchio's core capabilities by​​​‌ enabling the simulation of​ different types of actuators,​‌ including motors and muscles.​​ It also provides tools​​​‌ for parsing OpenSim models​ (version 4) to integrate​‌ their actuation specifications into​​ Pinocchio. This allows for​​​‌ the analysis and control​ of complex biomechanical and​‌ robotic systems with realistic​​ actuation dynamics.
• Release Contributions:​​​‌

  Features Added Freeflyer joint​ option Added Universal Joint​‌ option Added passive actuators​​ Added visual model creation​​​‌ Added model sub-module and​ vtp2stl support Added joint​‌ coordinate_names list Added joint​​ position limits to parsed​​​‌ models

  Improvements & Refactors​ Refactored custom_joint parsing Refactored​‌ cable wire handling General​​ refactors and test maintenance​​​‌ Cleaned OpenSim scouting parser​ Improved child orientation handling​‌ for joint addition Multiple​​ internal refactors (topic/refactor)

  Fixes​​​‌ Fixed child orientation during​ joint addition

  Chore /​‌ Maintenance Added Ruff line-length​​ check to CI Updated​​​‌ README Improved .gitignore (PyCharm​ & cache files) Fancy​‌ .gitignore cleanup

  Contributors Thanks​​ to:

  Pierre PUCHAUD Mégane​​​‌ MILLAN Alexandre SCHORTGEN Maël​ GALLOIS

• URL:
  https://­gitlab.­inria.­fr/­biomeca/­pin-actuation
• Contact:​‌
  Pierre Puchaud

7.1.3 moveit2_trajectory​​

• Keyword:
  Robotics
• Functional Description:​​​‌
  This package provides trajectory​ processing and execution capabilities​‌ for robotic systems. It​​ serves as a bridge​​​‌ between MoveIt2 motion planning​ and robot controllers, offering​‌ trajectory interpolation, Cartesian pose​​ computation, and a ROS​​​‌ 2 action server interface.​
• URL:
  https://­gitlab.­inria.­fr/­auctus-team/­components/­motion-planning/­moveit2_trajectory
• Contact:
  Lucas​‌ Joseph

7.1.4 ik_qontrol

• Keyword:​​
  Robotics
• Functional Description:
  Implementation​​​‌ of a CLIK (Closed-Loop​ Inverse Kinematics) solution based​‌ on Pinocchio and Qontrol.​​ Integration of this solution​​​‌ into a MoveIt plugin,​ enabling its use throughout​‌ the MoveIt framework.
• URL:​​
  https://­gitlab.­inria.­fr/­auctus-team/­people/­erwannlandais/­public/­ros2_ws/­command/­ik_qontrol
• Contact:
  Erwann Landais​​​‌

7.1.5 Active Constraint

• Keywords:​
  Haptic guidance, Haptic, Telerobotics​‌
• Scientific Description:
  The implementation​​ of haptic guidance in​​​‌ a teleoperation context with​ a haptic interface aims​‌ to encourage operators to​​ follow a desired behaviour​​​‌ while allowing a certain​ freedom of action. This​‌ package includes the integration​​ of a generic formula​​​‌ allowing continuous transition between​ different types of generic​‌ guidance model (potential fields,​​ spring-damper and guide tube)​​ as well as adaptable​​​‌ guidance models (ruling guidance,‌ marble cart).
• Functional Description:‌​‌
  active_constraint defines the following​​ parameters: stiffness and damping​​​‌ factors for five different‌ types of guidance (potential‌​‌ fields, spring-damper, guide tube,​​ ruling guidance, and marble​​​‌ cart), as well as‌ threshold distances, parameters for‌​‌ changing models on the​​ fly, and parameters specific​​​‌ to each guidance model.‌ The integration of an‌​‌ RVIZ plugin allows you​​ to switch between each​​​‌ guidance type on the‌ fly and modify the‌​‌ above parameters. The package​​ subscribes to the position​​​‌ and velocity topics of‌ the robot and the‌​‌ haptic interface scaled in​​ the robot's workspace. It​​​‌ also subscribes to a‌ topic to retrieve the‌​‌ target position(s) of the​​ guidance. These positions and​​​‌ velocities are used to‌ generate a guidance force‌​‌ that is sent to​​ the communication hub linked​​​‌ to the haptic interface.‌
• Publications:
  hal-04776071, tel-05418923‌​‌
• Contact:
  Alexis Boulay
• Participants:​​
  Alexis Boulay, Margot Vulliez,​​​‌ David Daney

7.2.1 Grip4All

Participants:‌​‌ Erwann Landais, Loic​​ Mazou, Alexis Boulay​​​‌, Lucas Joseph,‌ David Daney.

This‌​‌ platform consists of two​​ KUKA LBR iiwa 14​​​‌ robotic arms, a conveyor,‌ and a monocular camera.‌​‌ It demonstrates an automated​​ palletizing solution for fresh​​​‌ produce. Product spatial positions‌ are estimated using monocular‌​‌ vision. Depending on the​​ product characteristics, grasping is​​​‌ performed either by a‌ single arm (asynchronous operation)‌​‌ or collaboratively by both​​ arms (synchronous operation).

Advanced​​​‌ control techniques are employed‌ to dynamically manage accessibility‌​‌ constraints, grasping and placement​​ feasibility, grasp stability during​​​‌ transfer, and the overall‌ manipulation capabilities of the‌​‌ robotic system. In particular:​​

• Qontrol: online generation​​​‌ of optimal motions in‌ dynamic contexts using model‌​‌ predictive control and advanced​​ performance optimization tools for​​​‌ robotic systems.
• MegaPose:‌ object recognition and 6D‌​‌ pose estimation (position and​​ orientation in space).

These​​​‌ tools are developed at‌ Inria or within the‌​‌ Auctus team and contribute​​ to showcasing and advancing​​​‌ Inria's expertise in robotics.‌

7.2.2 Explorer

Participants: Esteban‌​‌ Cosserat, Margot Vulliez​​, Vincent Padois,​​​‌ Lucas Joseph.

The‌ Orthopus Explorer is a‌​‌ 6-DOF robotic arm used​​ within the EXTENDER project.​​​‌ The goal of the‌ EXTENDER project is to‌​‌ improve the daily lives​​ of people with disabilities​​​‌ by developing new, adaptable‌ control modes for robotic‌​‌ arms. The objective is​​ to use the Qontrol​​​‌ control software to command‌ the robotic arm developed‌​‌ by Orthopus and to​​ evaluate different control modalities,​​​‌ which will subsequently be‌ tested by users with‌​‌ disabilities.

7.2.3 Courrier

Participants:​​ Raphael Gerin, Margot​​​‌ Vulliez, Vincent Padois‌, Lucas Joseph.‌​‌

The objective of the​​ Courrier project is to​​​‌ evaluate the relevance of‌ communicating a robot's intention‌​‌ to a human operator​​ working in the same​​​‌ workspace. In this context,‌ an experimental platform has‌​‌ been developed involving a​​ robot tasked with sorting​​​‌ colored objects. The operator‌ and the robot share‌​‌ a common goal. The​​ software component allows the​​​‌ selection of different communication‌ modalities used by the‌​‌ robot in order to​​​‌ assess their impact on​ task efficiency and safety.​‌

7.2.4 DMC (Dual Marble​​ Cart)

Participants: Alexis Boulay​​​‌, Margot Vulliez,​ David Daney.

Experimental​‌ simulation platform dedicated to​​ the evaluation of a​​​‌ novel adaptive guidance model​ (dual adaptable) and its​‌ comparison with traditional guidance​​ models through reaching scenarios​​​‌ in dynamic and changing​ environments. This platform combines​‌ a physical haptic device​​ with a simulated robotic​​​‌ system and has already​ been used to conduct​‌ an experimental campaign, which​​ is expected to lead​​​‌ to scientific publications.

7.2.5​ DRPG (Deformable Robot Path​‌ Guidance)

Participants: Alexis Boulay​​, Margot Vulliez,​​​‌ David Daney.

Experimental​ simulation platform for assessing​‌ new adaptable virtual objects​​ (elastic band and predictive​​​‌ guide) and comparing them​ with a fixed virtual​‌ object through reaching scenarios​​ in changing environments. This​​​‌ platform relies on a​ physical haptic device coupled​‌ with a simulated robot​​ and has been used​​​‌ to conduct an experimental​ campaign that is expected​‌ to lead to scientific​​ publications.

7.2.6 OuiJa

Participants:​​​‌ Alicia Barsacq, Lucas​ Joseph, David Daney​‌.

Development of an​​ experimental platform for planar​​​‌ haptic communication between a​ human dyad, aimed at​‌ assessing the sense of​​ agency and understanding force​​​‌ exchange mechanisms. This platform​ makes use of force/torque​‌ sensors available at Inria,​​ as well as a​​​‌ motion capture system, to​ analyze participant behavior and​‌ identify key interaction features.​​

7.2.7 LePre Haptic

Participants:​​​‌ Benjamin Camblor, Hoang-Vy​ Nguyen, Alicia Barsacq​‌, Lucas Joseph,​​ Margot Vulliez.

Experimental​​​‌ platform designed to evaluate​ the impact of legibility​‌ and predictability of haptic​​ guides on performance, sense​​​‌ of agency, and subjective​ measures during an assisted​‌ teleoperation grasping task.

7.2.8​​ Beta

Participants: Jacques Zhong​​​‌, Lucas Joseph,​ Margot Vulliez.

Experimental​‌ simulation platform dedicated to​​ the evaluation and comparison​​​‌ of different dynamic arbitration​ methods. The experimental setup​‌ consists in a simulated​​ 7 DoF robot arm​​​‌ (Panda Franka) implemented in​ Gazebo with communication between​‌ the devices handled in​​ ROS. The robot is​​​‌ teleoperated with a haptic​ interface (omega.7, Force Dimension)​‌ in rate control. The​​ task is a shared-control​​​‌ pick-and-place, where the assistance​ estimate can be non-conflictual​‌ or conflictual with the​​ human targets at each​​​‌ scenario.

7.2.9 VisuoHaptic

Participants:​ Remi Lafitte, Lucas​‌ Joseph, Margot Vulliez​​.

Experimental teleoperation platform​​​‌ to study how Human​ agents perceive visual and​‌ haptic information and to​​ further characterize the rules​​​‌ of multisensory integration. Users​ are asked to perform​‌ a psychometric size-discrimination task.​​ Visual, haptic, and visuo-haptic​​​‌ objects (rectangular 3D shapes)​ are presented by means​‌ of a simulated visual​​ environment (computer screen, Chai3D​​​‌ simulation environment) and a​ haptic interface (Omega 7.0).​‌ Object width, cues reliabilities,​​ and visuo-haptic conflicts can​​​‌ be manipulated.

7.2.10 Bipetto​

Participants: Virgile Batto,​‌ Margot Vulliez.

Prototype​​ of a kid-size humanoid​​​‌ robot, built as an​ open-source, parallel-actuated biped experimental​‌ platform to evaluate proposed​​ co-design and multi-objective optimization​​​‌ methodologies.

8.1.1​​ Personalized Binary Emotion Classification​​​‌ Using Multimodal Smartphone and‌ Wearable Sensor Data

Participants:‌​‌ Jean-Marc Salotti.

The​​ results of Nicolas Simonazzi’s​​​‌ PhD thesis, defended three‌ years ago, were revisited,‌​‌ valorized and published in​​ 2025. This study investigated​​​‌ the binary classification of‌ emotional valence using data‌​‌ collected from motion sensors​​ and keystroke dynamics on​​​‌ a smartphone, as well‌ as from a connected‌​‌ wristband. To this end,​​ we developed a mobile​​​‌ application designed to induce‌ emotions through video stimuli‌​‌ while recording user interactions.​​ A dedicated digital self-assessment​​​‌ tool, adapted from the‌ Geneva Emotion Wheel, was‌​‌ created to help participants​​ report their emotional states.​​​‌ Sensor recordings were labeled‌ according to participants’ self-reports‌​‌ and the experimental video​​ conditions. We propose a​​​‌ method to process the‌ resulting temporal data and‌​‌ to automatically classify the​​ valence of reported emotions​​​‌ using machine learning techniques.‌ Both a general valence‌​‌ classifier trained on all​​ emotions from all participants​​​‌ and a personalized classifier‌ trained on a subset‌​‌ of emotions from a​​ single individual were evaluated.​​​‌ The most promising results‌ were obtained with the‌​‌ personalized model, which achieved,​​ on average across participants,​​​‌ approximately two-thirds accuracy in‌ valence classification using multimodal‌​‌ data fusion. 9.​​

8.1.2 Modeling Human visuo-haptic​​​‌ perception in teleoperation

Participants:‌ Rémi Lafitte, Margot‌​‌ Vulliez, Cécile Scotto​​.

Reliable visual and​​​‌ haptic feedback are known‌ to improve teleoperation tasks.‌​‌ Better understanding human visuo-haptic​​ perception could help to​​​‌ develop feedback strategies that‌ better inform the human‌​‌ operator during human-robot interaction.​​ The project is conducted​​​‌ in collaboration with the‌ CeRCA-CNRS at the University‌​‌ of Poitiers.

Psychophysic experiments​​ have suggested that the​​​‌ brain combines visual and‌ haptic estimates of environmental‌​‌ properties (e.g., object size)​​​‌ in a statistically optimal​ fashion. Whether this sensory​‌ integration still holds in​​ a more challenging environment,​​​‌ such as for teleoperation,​ remains unknown. We developed​‌ a psychophysic experiment, involving​​ 10 young healthy adults​​​‌ (median age = 27​ years old, 5 males)​‌ performing teleoperation size-estimation tasks,​​ to assess the validity​​​‌ of established integration models​ in teleoperation conditions. We​‌ compared the experimental results​​ to the predictions of​​​‌ the optimal integration (Maximum-Likelihood​ Estimate) and of best-cue​‌ switching models. The experimental​​ findings indicate that the​​​‌ observed visuo-haptic judgments could​ not be fully accounted​‌ by the sensory integration​​ or sensory segregation models.​​​‌ In this sense, sensory​ integration in this experiment​‌ could be said “ambiguous”​​ 22.

We further​​​‌ extended the experimental protocol​ to test if integrating​‌ the perceptual estimates of​​ an individual operator in​​​‌ the feedback process, specifically​ by modulating the haptic​‌ feedback, can help the​​ latter to better perform​​​‌ discriminate objects in teleoperation.​

8.1.3 Legibility and predictability​‌ of haptic guidance in​​ teleoperation

Participants: Benjamin Camblor​​​‌, Hoang-Vy Nguyen,​ Alicia Barsacq, Margot​‌ Vulliez.

Haptic guidance​​ is a method to​​​‌ assist a human operator​ performing tasks in teleoperation​‌ through force feedback. Taking​​ into account human factors​​​‌ and individual preferences is​ crucial when designing such​‌ haptic assistance. It can​​ modify the perception and​​​‌ understanding that humans have​ of the robot's assistance​‌ behavior, i.e. the cognitive​​ transparency of the robot.​​​‌ Specifically, making the robot​ actions more predictable and/or​‌ more legible can highly​​ impact the understanding of​​​‌ the robot's behavior during​ human-robot interaction. The predictability​‌ of an action is​​ linked to its similarity​​​‌ to the action that​ the human agent would​‌ expect for a given​​ objective (action deduced from​​​‌ its goal). Conversely, an​ action is legible when​‌ it enables the human​​ to predict the objective​​​‌ or intention it expresses​ (goal deduced from the​‌ action).

The project aims​​ at evaluating the impact​​​‌ of different haptic guidance​ on the legibility and​‌ predictability of the robot's​​ assistance in teleoperation. We​​​‌ designed an experimental protocol​ and platform to study​‌ how different haptic guidance,​​ where we vary the​​​‌ guidance-path curvature, can be​ more or less legible​‌ and predictable during assisted​​ reaching scenarios. We further​​​‌ investigate the effect of​ haptic guidance legibility/predictability on​‌ performance, user comfort, and​​ agency.

8.1.4 Proximal Human-Robot​​​‌ Collaboration: Toward Human-Like Joint​ Agency

Participants: Alicia Barsacq​‌, David Daney,​​ Jean-Christophe Sarrazin.

When​​​‌ interacting with artificial agents​ in a joint task,​‌ human can experience a​​ loss of feeling of​​​‌ control, or a loss​ of agency. The origin​‌ of this phenomenon is​​ not well understood, but​​​‌ have a negative impact​ on the subjective and​‌ objective performance, awareness during​​ the task and involvment.​​​‌ However, interacting with a​ human partner can prevent​‌ this loss of agency​​ in collaborativ joint action.​​​‌

Through a literature review​ of current evaluations of​‌ agency in collaborative human-robot​​ interaction, we show that​​​‌ the embodiement of robots​ can prevent agency loss​‌ in joint tasks in​​ the same way two​​ humans interacting together do.​​​‌ Several barriers preventing the‌ understanding of joint agency‌​‌ in human-robot collaboration were​​ identified, and guidelines for​​​‌ roboticists were provided to‌ overcome them.

This thesis‌​‌ is co-supervised with ONERA​​ as part of the​​​‌ PEPR O2R program.

8.2.1‌​‌ Study of Motor Variability​​

Participants: David Daney,​​​‌ Vincent Padois, Pauline‌ Maurice, Jonathan Savin‌​‌.

This long term​​ research project led to​​​‌ a main action in‌ 2024: the analysis of‌​‌ the experimental results obtained​​ during the experimental campaign​​​‌ of 2023 (MOVER). These‌ results are encouraging as‌​‌ they demonstrate that motor​​ variability depends on some​​​‌ features of the task,‌ especially pace and direction‌​‌ of the movement. Moreover,​​ the observed motor variability​​​‌ is high enough to‌ induce changes in biomechanical‌​‌ risk factors as estimated​​ through standard ergonomic scores​​​‌ 50. The analysis‌ of the results will‌​‌ be pursued in 2025,​​ potentially within the framework​​​‌ of the INTRO project‌ submitted to the ANR‌​‌ yearly call in collaboration​​ with INRS and the​​​‌ LARSEN team in Nancy.‌

8.2.2 Searching for best-fitting‌​‌ musculoskeletal models approximating an​​ individual's upper limb force​​​‌ capacities

Participants: Gautier Laisné‌, Jean-Marc Salotti,‌​‌ Nasser Rezzoug.

8.2.3​​ Parametric Identification of Metabolic​​​‌ Models of Fatigue

Participants:‌ Pierre Puchaud, Simon‌​‌ Bernier, Jérémy Briand​​​‌.

In the context​ of analyzing human physical​‌ capacity and fatigue. We​​ developed a computational pipeline​​​‌ using Bioptim to identify​ individualized physiological parameters from​‌ mechanical power and oxygen​​ consumption. This framework provides​​​‌ a robust method for​ extracting internal physiological characteristics​‌ and fatigue metrics, expanding​​ the team's capabilities in​​​‌ human monitoring beyond pure​ kinematics and dynamics. The​‌ problem aims to identify​​ time-independent parameters, specifically maximal​​​‌ oxygen uptake, maximal lactate​ production rate, and active​‌ muscle mass percentage, while​​ simultaneously reconstructing hidden metabolic​​​‌ states such as glycogen​ depletion and muscle lactate​‌ accumulation. The formulation satisfy​​ non-linear biochemical dynamics derived​​​‌ from equilibrium equations (e.g.,​ creatine-phosphate reaction, Hill equations​‌ for oxidative phosphorylation). This​​ work is a collaboration​​​‌ with Université de Montréal.​

8.2.4 Modeling of human-human​‌ physical interaction

Participants: Alicia​​ Barsacq, David Daney​​​‌, Jean-Christophe Sarrazin.​

Using control theory tools​‌ such as feeback control​​ and MPC, a physical​​​‌ model of the OuiJa​ platform was developped. This​‌ model allow for a​​ simulation of the behaviour​​​‌ of one or two​ humans collaborating via haptic​‌ information exchange. The model​​ relies on traditionnal implementation​​​‌ of computationnal neuroscience, taking​ into account muscle dynamics,​‌ sensory delays and noise.​​ It produces simulated data​​​‌ such as force, movement​ of the OuiJa platform​‌ (position, velocity). Those simulated​​ data can be compared​​​‌ to the experimental one​ in different conditions. Finally,​‌ this model allow for​​ a computationnal model of​​​‌ agency to be tested​ via simulated prediction error​‌ in different conditions.

This​​ thesis is co-supervised with​​​‌ ONERA as part of​ the PEPR O2R program.​‌

8.3.1 Re-expression of manual​​​‌ expertise through manual control​ of a teleoperated system​‌

Participants: Erwann Landais,​​ Vincent Padois, Nasser​​​‌ Rezzoug.

In the​ thesis of Erwann Landais​‌ which was defended in​​ December 37, we​​​‌ studied how teleoperation can​ allow for the remote​‌ expression of technical gestures,​​ with chemistry as a​​​‌ potential applicative domain. Indeed,​ teleoperation enables a task​‌ to be carried out​​ remotely by a human​​​‌ expert. This remote control​ is often a guarantee​‌ of greater safety and​​ comfort, or simply of​​​‌ feasibility in hazardous environments.​ However, it can also​‌ mean a loss of​​ efficiency, or added complexity.​​​‌ To avoid these pitfalls,​ it is necessary to​‌ consider 1) what constitutes​​ an operator's expertise for​​​‌ a given task, 2)​ the constraints encountered in​‌ carrying it out, and​​ 3) the form of​​​‌ a teleoperation system adapted​ to it. An example​‌ of a task that​​ could benefit from teleoperation​​​‌ is the task of​ finding solvents for chemical​‌ compounds, which is one​​ of Syensqo's areas of​​​‌ expertise. This involves characterizing​ the solubility of a​‌ solute in a set​​ of solvents, in order​​​‌ to determine the optimum​ solvent for that solute.​‌ This task, based on​​ visual, cognitive and manual​​​‌ expertise, is performed by​ a small number of​‌ expert technicians. Performing this​​ task relies on an​​​‌ empirical, tedious and sometimes​ dangerous process, motivating the​‌ distancing of technicians from​​ the experimental environment through​​ robotic assistance. Using this​​​‌ task as a case‌ study, the thesis of‌​‌ Erwann Landais aimed at​​ answering the following questions​​​‌ : on the one‌ hand, how can an‌​‌ operator's expertise be preserved​​ when performing a task​​​‌ using a teleoperated system‌ ? And on the‌​‌ other hand, how can​​ the suitability of a​​​‌ teleoperation solution for performing‌ an expert task be‌​‌ assessed ? To answer​​ these questions, a broad-spectrum​​​‌ literature review was carried‌ out and two experimental‌​‌ studies were conducted. While​​ the first study studied​​​‌ semi-automatic manipulation protocols, in‌ the second study, which‌​‌ restults were analysed in​​ 2024, a robot was​​​‌ designed to achieve a‌ range of motion similar‌​‌ to that observed in​​ technicians, and intuitive interfaces​​​‌ were used to define‌ the desired movement of‌​‌ the vial in real​​ time. The study shows​​​‌ that controlling the robot‌ via these interfaces does‌​‌ not achieve an efficiency​​ similar to that of​​​‌ the manual mode. However,the‌ performance achieved is encouraging,‌​‌ and the study identifies​​ several avenues of improvement​​​‌ for the efficient and‌ reliable deporting of the‌​‌ technical gesture in chemistry.​​ Finally, beyond the application​​​‌ framework, this work establishes‌ a comprehensive methodology for‌​‌ evaluating the performance of​​ teleoperation modalities.

8.3.2 Adaptive​​​‌ haptic guidance to assist‌ human during teleoperation

Participants:‌​‌ Alexis Boulay, Margot​​ Vulliez, David Daney​​​‌.

This work is‌ within the framework of‌​‌ the collaboration with the​​ Farm3 company and addresses​​​‌ issues of remote vertical‌ farming.Performing remote tasks through‌​‌ a teleoperation system can​​ be assisted through haptic​​​‌ guidance, a force feedback‌ computed based on a‌​‌ virtual geometric constraints to​​ help the user to​​​‌ follow a given behavior‌ (task trajectory, safety area,...).‌​‌

A user study showed​​ that the choice of​​​‌ guidance model depends on‌ the interaction context: task‌​‌ complexity, environmental clutter, user​​ preference, etc. 25.​​​‌ This initial result motivated‌ the development of a‌​‌ generic, adaptable online haptic​​ guidance that follows the​​​‌ dynamics of the interaction‌ to optimize human comfort.‌​‌

8.3.3 Unified‌ human-robot simulation for modulation‌​‌ of muscle activation

Participants:​​ Pierre Schegg, Pierre​​​‌ Puchaud, François Bailly‌.

Current robot-assisted physiotherapy‌​‌ often lacks explicit biomechanical​​​‌ integration in control systems.​ To address this, we​‌ develope a unified framework​​ co-simulating a musculoskeletal model​​​‌ of the upper limb​ and a collaborative robotic​‌ arm. Using Bioptim, we​​ formulated a single optimal​​​‌ control problem encompassing both​ human and robot variables​‌ to define assistive torque​​ trajectories. This approach allows​​​‌ for the modulation of​ specific muscle activations—specifically the​‌ Supraspinatus in a rotator​​ cuff rehabilitation scenario—via a​​​‌ single hyperparameter. Preliminary results​ demonstrate that the formulation​‌ reliably adapts the robot’s​​ control law. Rather than​​​‌ uniformly scaling torques, the​ controller non-trivially redistributes the​‌ load across robot joints​​ to achieve targeted reductions​​​‌ in muscle activation. This​ work validates the use​‌ of optimal control to​​ integrate biomechanical constraints into​​​‌ human-robot interaction, paving the​ way for personalized movement​‌ assistance in cases such​​ as muscle dystrophy or​​​‌ hemiplegia.

8.3.4 A Riemannian​ approach for Inverse Optimal​‌ Control

Participants: Ahmed-Manaf Dahamni​​, David Daney,​​​‌ François Charpillet.

The​ methods for solving Inverse​‌ Optimal Control that are​​ currently available in the​​​‌ literature are either computationally​ slow like the commonly​‌ used Bilevel, or approximative​​ methods that do not​​​‌ recovered the closest trajectory​ to the observation. To​‌ try and solve this​​ problem a new approach​​​‌ called the Projected IOC​ has been created. However,​‌ this method has encountered​​ solvability issues using modern​​​‌ solver. In our research,​ we have discovered that​‌ the solvability prolems stem​​ from a violation of​​​‌ the Mangasarian–Fromovitz constraint qualification​ (MFCQ).

Following this discovery,​‌ we have developed a​​ new method based on​​​‌ Riemannian optimization that has​ working principles similar to​‌ the Bilevel approach with​​ a faster run time.​​​‌ This method can compute​ the jacobian of the​‌ Direct Optimal Control problem​​ if proper conditions are​​​‌ met for the jacobian​ to exist. Gradient descent​‌ is then used to​​ minimize a distance to​​​‌ the observation.

8.4.1 Model​‌ Predictive Control blending and​​ adaptive assistance for shared​​​‌ human-robot teleoperation

Participants: Elio​ Jabbour, Margot Vulliez​‌, Vincent Padois,​​ Célestin Préault.

Shared​​​‌ control aims at assisting​ human operators using robots​‌ in physically and cognitively​​ demanding tasks which cannot​​​‌ be automated as they​ require human expertise and​‌ deliberative abilities. Sharing control​​ for a given task​​​‌ typically involves blending algorithms​ that combine human control​‌ inputs and (pre)planed assistance​​ trajectories. Over the last​​​‌ year, we proposed a​ shared control architecture that​‌ combines a predictive approach​​ to blending and adaptation​​​‌ mechanisms to correct imperfect​ assistance inputs.

8.4.2 Dynamic‌​‌ authority distribution in haptic​​ shared control

Participants: Jacques​​​‌ Zhong, Margot Vulliez‌, Jee-Hwan Ryu.‌​‌

An important challenge of​​ shared control lies in​​​‌ how two agents (i.e.‌ the human and the‌​‌ assistance) share the control​​ of the teleoperated robot,​​​‌ set through the level‌ of authority. However, there‌​‌ is currently no consensus​​ on the most relevant​​​‌ metric to use to‌ adjust this level of‌​‌ authority between the human​​ and the robot, as​​​‌ this is mainly context-dependent.‌ The arbitration is often‌​‌ set through an arbitrary​​ criteria defined by-hand and​​​‌ no global arbitration method‌ exists.

We proposed a‌​‌ novel optimization-based framework that​​ dynamically arbitrates between different​​​‌ authorithy distibutions, with an‌ online adaptation mechanism with‌​‌ respect to the user​​ and the task. The​​​‌ arbitration policy is formulated‌ as a Quadratic Progam‌​‌ (QP) to combine different​​ metrics under constraints over​​​‌ the authority level evolution.‌ Bayesian optimization is used‌​‌ as a sample-efficient global​​ optimizer to adjust the​​​‌ policy parameters in a‌ few evaluations. We evaluated‌​‌ the proposed method on​​ a shared-controlled reaching task​​​‌ involving two typical situations:‌ non-conflictual and conflictual assistance.‌​‌ Compared to existing methods,​​ results show that the​​​‌ method yields significantly better‌ interaction quality (from both‌​‌ objective and subjective measures)​​ for non-conflictual scenarios, but​​​‌ not for conflictual scenarios‌ where the assistance goal‌​‌ is not aligned with​​ the actual human target.​​​‌

8.4.3 Scenario-based Model Predictive‌ Control for safe and‌​‌ effective human-robot collaboration

Participants:​​ Vincent Padois, Sebastien​​​‌ Kleff, Kloe Bonnet‌, Tianyi Jin,‌​‌ Raphael Gerin.

Human-robot​​ collaboration requires both safety​​​‌ and effectivness to be‌ considered as viable solution‌​‌ in situtations where human​​ working conditions could benefit​​​‌ from a robotic assistance.‌ These two objectives are‌​‌ somewhat antinomic. Achieving both​​ requires both an optimal​​​‌ control approach (i.e. one‌ that considers the long‌​‌ term consquences of the​​ control actions), online replanning​​​‌ (to constantly adapt to‌ changing objectives), as well‌​‌ as a way to​​​‌ consider the variability (and​ associated uncertainties) in the​‌ human motion strategies. To​​ combine these three characteristics,​​​‌ we have started to​ explore Scenario-based Model Predictive​‌ Control (SMPC) as a​​ way to formulate the​​​‌ control problem faced when​ considering Human-Robot collaboration. This​‌ work has been initiated​​ within the framework of​​​‌ Kloe Bonnet internship and​ is being pursued as​‌ part of the ANR​​ COURRIER research program with​​​‌ the PhD work of​ Tianyi Jin , co-advised​‌ by Sebastien Kleff and​​ Vincent Padois . Raphael​​​‌ Gerin provides support with​ the implementation of the​‌ proposed control approach within​​ an experimental set-up shared​​​‌ by the partners of​ the ANR project. He​‌ is also in charge​​ of implementing state-of-the art,​​​‌ lower-level, task-based control strategies​ allowing for the robust​‌ achievement of the task​​ and motion planning strategy​​​‌ emerging from the higher​ level SMPC controller. With​‌ this architecture he has​​ provided a ready-to-use simulation​​​‌ software to one of​ the COURRIER project partner​‌ (Onera) performing robot motion​​ prediction tests with humans.​​​‌

8.4.4 A Unified Tactile​ Servoing Framework based on​‌ Hybrid Force-Position Control

Participants:​​ Sebastien Kleff, Vincent​​​‌ Padois, Lucas Joseph​.

In robotics, traditional​‌ force control lacks local​​ contact information. Tactile sensors​​​‌ provide rich feedback on​ physical interaction, but remain​‌ notably difficult to integrate​​ into real-time control loops.​​​‌ The research we are​ initiating towards tactile based​‌ control and its application​​ to human-robot collaboration requires​​​‌ a solid grounding and​ we have proposed a​‌ unified tactile servoing formulation​​ that allows explicit control​​​‌ of contact pose and​ force at the Center​‌ of Pressure (CoP). Unlike​​ conventional tactile servoing techniques,​​​‌ which are tightly coupled​ to a specific sensor​‌ and often treat the​​ contact wrench as disturbance​​​‌ to be rejected, our​ approach relies on a​‌ generic and physically grounded​​ feature space. We derive​​​‌ a hybrid force-position control​ law based on the​‌ Jacobian at the CoP​​ that naturally decouples force​​​‌ and motion subspaces, ensuring​ geometric consistency during the​‌ contact interaction. Our framework,​​ validated on a robotic​​​‌ manipulator with vision-based tactile​ sensing, demonstrates robust contact​‌ maintenance and force tracking​​ capabilities, outperforming standard imagebased​​​‌ and pose-based controllers in​ tasks requiring precise regulation​‌ of the physical interaction.​​ This work has been​​​‌ submitted to the IEEE​ Robotics and Automation Letters​‌ 16.

8.4.5 Extended​​ Friction Models for the​​​‌ Physics Simulation of Servo​ Actuators

Participants: Marc Duclusaud​‌, Grégoire Passault,​​ Vincent Padois, Olivier​​​‌ Ly.

Accurate physical​ simulation is crucial for​‌ the development and validation​​ of control algorithms in​​​‌ robotic systems. Recent works​ in Reinforcement Learning (RL)​‌ take notably advantage of​​ extensive simulations to produce​​​‌ efficient robot control. State-of-the-art​ servo actuator models generally​‌ fail at capturing the​​ complex friction dynamics of​​​‌ these systems. This limits​ the transferability of simulated​‌ behaviors to real-world applications.​​ In this work, we​​​‌ present extended friction models​ that allow to more​‌ accurately simulate servo actuator​​ dynamics. We propose a​​​‌ comprehensive analysis of various​ friction models, present a​‌ method for identifying model​​ parameters using recorded trajectories​​ from a pendulum test​​​‌ bench, and demonstrate how‌ these models can be‌​‌ integrated into physics engines.​​ The proposed friction models​​​‌ are validated on four‌ distinct servo actuators and‌​‌ tested on 2R manipulators,​​ showing significant improvements in​​​‌ accuracy over the standard‌ Coulomb-Viscous model. Our results‌​‌ highlight the importance of​​ considering advanced friction effects​​​‌ in the simulation of‌ servo actuators to enhance‌​‌ the realism and reliability​​ of robotic simulations. This​​​‌ work has been published‌ at IEEE ICRA 2025‌​‌ 11.

8.4.6 Simulation,​​ Observability Analysis, and AI​​​‌ Methods for Tracking

Participants:‌ Jimmy Étienne, David‌​‌ Daney, François Charpillet​​.

Over the past​​​‌ year, I developed and‌ validated simulation and diagnostic‌​‌ tools for estimation, tracking,​​ and decision-making. Key contributions​​​‌ include the FIM Observability‌ Tool, an interactive simulator‌​‌ for analyzing Fisher Information​​ Matrix observability under bearing-only​​​‌ measurements, and a Passive‌ Hydrophone Array Simulator that‌​‌ synthesizes towed-array sonar scenes​​ with multi-hydrophone time series,​​​‌ bearing measurements, and ground-truth‌ metadata to accelerate acoustic‌​‌ tracking and localization prototyping.​​ In parallel, I contributed​​​‌ to early work on‌ a new platform for‌​‌ mobile robotics and delivered​​ new results through POMmeDaPy​​​‌ (Python POMDP implementations including‌ naive solvers, value functions,‌​‌ value iteration, and HSVI)​​ and machine-learning experiments using​​​‌ diffusion processes for trajectory‌ generation.

10.1.1​​ Inria associate team not​​​‌ involved in an IIL​ or an international program​‌

10.1.2 Visits to international​​​‌ teams

10.2.1 Other european programs/initiatives​​

10.3.1‌ PEPR O2R - AS2,‌​‌ Robot motion with physical​​ interactions and social adaptation​​​‌

Participants: David Daney,‌ Vincent Padois, Margot‌​‌ Vulliez, Alicia Barsacq​​, Jean-Christophe Sarrazin.​​​‌

The objective of this‌ structuring action is to‌​‌ rethink the problem of​​ motion generation of robotic​​​‌ systems by addressing it‌ in its globality and‌​‌ by redefining the research​​ objectives in connection with​​​‌ the Human and Social‌ Sciences. It aims on‌​‌ the one hand to​​ develop technological solutions, innovative​​​‌ methods and software to‌ provide these new generation‌​‌ robots with advanced planning​​ and control capabilities of​​​‌ their movements and on‌ the other hand to‌​‌ guarantee that the motor​​ actions produced by these​​​‌ systems will be well‌ adapted to humans. These‌​‌ systems will have to​​ be able to interact​​​‌ physically with their environment‌ and with humans to‌​‌ perform a wide panel​​ of tasks ranging from​​​‌ agile locomotion to dexterous‌ manipulation through collaborative tasks.‌​‌ Among the key objectives​​ linked to these developments,​​​‌ these robots will have‌ to be able to‌​‌ anticipate their movements but​​ also to adapt them​​​‌ to react to unforeseen‌ events and to implement‌​‌ robust control strategies to​​ guarantee the successful execution​​​‌ of tasks and safety‌ for the human. The‌​‌ question of movement autonomy​​ of the machine and​​​‌ sharing of control during‌ collaborative tasks will also‌​‌ be essential with regard​​​‌ to applications and needs.​ In all cases, it​‌ will be necessary to​​ ensure the sustainability of​​​‌ the approaches developed with​ regard to environmental and​‌ societal challenges.

Project in​​ a nutshell:

• Consortium: AUCTUS@Inria,​​​‌ Rainbow@inria, willow@inria, Onera, Lirmm,​ univ. Picardie, Cerca, Pprime​‌
• Funding: France 2030 (PEPR​​ O2R)
• Duration: 2024-2032

10.3.2​​​‌ PEPR Robotics : Dexterous​ Robotic Manipulation for Industry​‌

Participants: David Daney,​​ Vincent Padois, Margot​​​‌ Vulliez, Lucas Joseph​.

The PEPR Robotics​‌ program aims to develop​​ the foundations of high-performance,​​​‌ frugal, and responsible robotics​ to support societal and​‌ industrial transformations, by integrating​​ energy and environmental challenges​​​‌ while improving the productivity​ and sustainability of human​‌ activities. By accelerating research​​ from its earliest stages,​​​‌ the program seeks to​ develop a set of​‌ functional, software, and hardware​​ building blocks, enabling the​​​‌ emergence of innovative robotic​ systems or functions capable​‌ of delivering pre-maturation results​​ and fostering valorization within​​​‌ the strategic sectors targeted​ by France 2030.

Robotic​‌ manipulation capabilities are crucial​​ for industry but remain​​​‌ limited when dealing with​ complex, fragile, deformable, or​‌ size-varying objects. The DRMI​​ project aims to overcome​​​‌ these limitations by developing​ advanced manipulation solutions suited​‌ to constrained industrial environments.​​ It relies on the​​​‌ design of new versatile​ grippers integrating intelligent perception​‌ and adaptive control, capable​​ of continuously adjusting position,​​​‌ speed, force, and stiffness.​ The project revisits the​‌ entire design chain of​​ robotic manipulation systems, from​​​‌ mechanical structures to control​ algorithms. Aligned with a​‌ circular and frugal economy​​ approach, DRMI particularly targets​​​‌ recycling and remanufacturing applications,​ which require dexterous manipulation​‌ at multiple scales, from​​ millimeter-sized components to large​​​‌ objects.

Project in a​ nutshell:

• Consortium: AUCTUS@Inria, Rainbow@Inria,​‌ LIRMM, PPRIME, CEA List,​​ ENSAM, Marie et Louis​​​‌ Pasteur Univ., Icube IMT​ Atlantique, Institut Pascal,
• Funding:​‌ France 2030 (PEPR Robotique)​​
• Duration: 2025-2032

10.3.3 ANR​​​‌ COURRIER, COopération hUmain Robot​ - Rôle des Intentions​‌ Exprimées par le Robot​​

Participants: Vincent Padois,​​​‌ Margot Vulliez, Lucas​ Joseph, Raphael Gerin​‌, Tianyi Jin,​​ Kloe Bonnet, Sebastien​​​‌ Kleff.

The COURRIER​ project addresses the challenge​‌ of safe and efficient​​ human-robot collaboration in industrial​​​‌ environments. While collaborative robots​ are increasingly deployed, their​‌ opacity and limited predictability​​ remain major barriers to​​​‌ effective coordination and safety.​ Rather than relying solely​‌ on isolation or fully​​ autonomous intelligence, COURRIER proposes​​​‌ a complementary approach that​ leverages the human operator's​‌ ability to anticipate and​​ coordinate with robotic systems,​​​‌ provided that their behavior​ and intentions are intelligible.​‌ Grounded in cognitive science​​ and theories of joint​​​‌ action, the project investigates​ how the readability of​‌ a robot's intentions—at strategic,​​ task-specific, and motor levels—affects​​​‌ its predictability and, in​ turn, human performance, cognitive​‌ load, learning, and trust.​​ COURRIER combines robotics, control,​​​‌ cognitive psychology, and neuroscience​ to identify the minimal​‌ information required for humans​​ to build accurate mental​​​‌ models of robotic behavior.​ The project develops both​‌ methods for communicating robot​​ intentions through motion and​​​‌ control strategies, and objective​ metrics to quantify the​‌ impact of predictability on​​ coordination quality and cognitive​​ cost. The methodology relies​​​‌ on controlled human-robot collaboration‌ experiments, centered on a‌​‌ shared object-sorting task using​​ a collaborative robotic arm.​​​‌ Performance, subjective measures (trust,‌ perceived control, usability), and‌​‌ physiological indicators (EEG, eye​​ tracking) are jointly analyzed.​​​‌ The project is structured‌ around five work packages‌​‌ covering coordination, intention communication,​​ expressive control, performance evaluation,​​​‌ and learning and trust.‌ Inria, through the Auctus‌​‌ team, contributes core expertise​​ in robot control, predictive​​​‌ and constrained motion generation,‌ and human-robot interaction. Auctus‌​‌ is responsible for the​​ design and implementation of​​​‌ control architectures enabling the‌ expression and communication of‌​‌ robot intentions, as well​​ as the experimental robotic​​​‌ platform and simulation tools.‌ The team plays a‌​‌ central role in bridging​​ control theory, experimental validation,​​​‌ and cognitive modeling, ensuring‌ strong integration between robotics‌​‌ developments and human-centered evaluation.​​

Project in a nutshell:​​​‌

• Consortium: ICNA/Onera (coordinator -‌ Jean-Christophe Sarrazin), INCIA/CNRS, Mnemosyne/Inria,‌​‌ Auctus/Inria (PI: Vincent Padois​​ )
• Funding: ANR (367​​​‌ 800 Euros for Inria)‌
• Duration: 2024-2028

10.3.4 Défi‌​‌ Transfert robotique, GRIP4ALL

Participants:​​ David Daney, Vincent​​​‌ Padois, Margot Vulliez‌, Lucas Joseph,‌​‌ Alexis Boulay, Erwann​​ Landais, Loic Mazou​​​‌.

As companies adapt‌ to changes in the‌​‌ Factory of the Future,​​ they need to modernise​​​‌ their working environments, in‌ particular to improve logistics‌​‌ and transform industrial sites​​ into flexible spaces incorporating​​​‌ intelligent machines and shared‌ with human operators. The‌​‌ aim of the Grip4All​​ project is to make​​​‌ industry more competitive by‌ developing a new palletising‌​‌ cell adapted to the​​ severe constraints imposed on​​​‌ the logistics flow when‌ handling mixed products (of‌​‌ varying dimensions and weight)​​ and arranging them on​​​‌ a pallet, without having‌ to sort them manually‌​‌ upstream. This new type​​ of palletising meets a​​​‌ strong demand from a‌ number of sectors, notably‌​‌ mass distribution and the​​ food industry. It meets​​​‌ the demand for handling‌ heterogeneous products without imposing‌​‌ constraints on their packaging,​​ which significantly improves productivity​​​‌ and eliminates tedious human‌ tasks. No similar solution‌​‌ currently exists on the​​ market. The flexible robotics​​​‌ issues addressed will be‌ transposable to other logistics‌​‌ processes in the factory​​ of the future. Grip4All​​​‌ is proposing a logistics‌ line design that breaks‌​‌ with the state of​​ the art, where lines​​​‌ are usually built by‌ aggregating specialised cells: by‌​‌ their very nature, these​​ limit the possibilities for​​​‌ logistics to evolve and‌ adapt to a wide‌​‌ variety of products, and​​ therefore also their sustainability​​​‌ in a context of‌ sustainable development. The Grip4All‌​‌ approach is based on​​ the use of the​​​‌ technological building blocks of‌ robotics and new technologies‌​‌ in gripping, dynamic movement​​ planning, AI perception and​​​‌ scheduling. The palletising cell‌ is made flexible by‌​‌ using one or more​​ 'augmented' manipulator arms with​​​‌ adaptive grippers, and by‌ arranging and controlling the‌​‌ manipulator arms collaboratively in​​ a dynamic environment. The​​​‌ system is based on‌ a reactive and predictive‌​‌ control mode, as well​​ as perception and scheduling​​​‌ algorithms that detect heterogeneous‌ products and calculate their‌​‌ positioning according to their​​​‌ nature. The aim is​ to be able to​‌ handle a wide variety​​ of products on the​​​‌ same pallet in real​ time, with adaptive gripping​‌ capabilities compatible with a​​ wide range of products​​​‌ (weight, shape, nature, packaging),​ including two-arm gripping, and​‌ with vision-based detection and​​ perception capabilities coupled with​​​‌ dynamic generation of movements​ for gripping and depositing​‌ products on the pallet.​​ To meet this challenge,​​​‌ the project brings together​ 4 complementary partners: 2​‌ industrial partners and 2​​ academic partners. The Fives​​​‌ industrial group, through its​ subsidiary FIVES Syleps, world​‌ leader in computerised and​​ robotised mixed palletising, is​​​‌ seeking to stay ahead​ of its competitors and​‌ open up new markets.​​ Subcontracting industrial partner Kannon​​​‌ MSD specialises in vision-based​ perception. The INRIA teams​‌ will contribute their technological​​ building blocks in the​​​‌ form of software dedicated​ to reactive control in​‌ dynamic and human environments​​ by Auctus, and to​​​‌ product scheduling by the​ Edge team. RoBioSS, a​‌ leading team in the​​ development of high dexterity​​​‌ robotic hands and grippers,​ will contribute its mechatronic​‌ design bricks to increase​​ robot flexibility. To ensure​​​‌ that the solution is​ acceptable and that human​‌ factors are taken into​​ account, the consortium will​​​‌ be supported by the​ CERCA laboratory, which is​‌ working closely with the​​ RoBioSS team has been​​​‌ working on this subject​ for over ten years.​‌

Project in a nutshell:​​

• Consortium: AUCTUS@Inria (coordinator -​​​‌ David Daney), RoBioSS@Pprime (CNRS),​ Syleps@Fives
• Funding: France 2030​‌ (952 000 Euros for​​ Auctus)
• Duration: 2024-2027

10.3.5​​​‌ Défi Transfert robotique, Extender​

Participants: Esteban Cosserat,​‌ Vincent Padois, Margot​​ Vulliez, Lucas Joseph​​​‌.

Project in a​ nutshell: The EXTENDER project​‌ aims to help wheelchair​​ users perform daily tasks​​​‌ using a robotic arm​ by developing solutions that​‌ adapt to diverse sensory-motor,​​ cognitive, and socio-economic profiles.​​​‌ The project follows a​ co-design approach involving robotics​‌ labs, the industrial partner​​ ORTHOPUS, end users, medical​​​‌ teams, and socio-anthropology researchers.​ EXTENDER targets the growing​‌ assistive technology market, leveraging​​ recent advances in robot​​​‌ control and human-robot interaction​ to improve safety, robustness,​‌ efficiency, and ease of​​ use. The project integrates​​​‌ multimodal interfaces—joysticks, touch, motion,​ eye tracking, voice, and​‌ AR—and evaluates solutions through​​ lab experiments, healthcare center​​​‌ trials, and international competitions​ (Cybathlon 2025-2026), complemented by​‌ socio-anthropological observations. Inria's Auctus​​ team leads the adaptation​​​‌ of advanced control and​ interaction methods, ensuring safety,​‌ intuitiveness, and robustness in​​ real-world use. They also​​​‌ contribute to evaluation, personalization,​ and technology transfer, bridging​‌ research and industrial deployment​​ alongside ORTHOPUS and clinical​​​‌ partners.

• Consortium: ISIR/Sorbonne Université​ (coordinator: Guillaume Morel), Orthopus,​‌ AUCTUS@Inria (PI: Vincent Padois​​ ), LAAS/CNRS, ESEAN, Université​​​‌ de Clermont
• Funding: France​ 2030 (324 000 Euros​‌ for Auctus)
• Duration: 2024-2027​​

10.3.6 ANR ASAP-HRC, Autonomy​​​‌ for Shared Action and​ Perception in Human-Robot Collaboration​‌

Participants: Vincent Padois,​​ Margot Vulliez, Jacques​​​‌ Zhong, Elio Jabbour​, Benjamin Camblor,​‌ Remi Lafitte, Célestin​​ Préault, Cécile Scotto​​​‌.

This collaborative ANR​ project started in 2021​‌ between the AUCTUS team​​ at Inria, the RoBioSS​​ team at the Pprime​​​‌ Institute (CNRS), and the‌ CeRCA laboratory (CNRS). It‌​‌ aims at rethinking autonomy​​ for shared action and​​​‌ perception in Human-Robot Collaboration,‌ through transverse studies in‌​‌ robotics and cognitive sciences.​​ More particularly, three scientific​​​‌ axes must be addressed‌ to develop a human-centered‌​‌ and generic shared-autonomy framework:​​ 1) study key features​​​‌ of Human-Robot perception-action mechanisms‌ and identify multisensory integration‌​‌ processes involved in Human-Robot​​ interaction. These human studies​​​‌ should constitute the baseline‌ of robotic developments and‌​‌ shape the shared-autonomy scheme;​​ 2) develop a shared​​​‌ perception between the different‌ actors (humans and collaborative‌​‌ robots),according to their sensory​​ data and involvement in​​​‌ the task. This shared‌ perception will be based‌​‌ on a multimodal (haptic,​​ visual) feedback mixture conveying​​​‌ information about the task,‌ the environment, and the‌​‌ collaborators; 3) combine Human-Robot​​ commands into a joint​​​‌ action toward the task‌ goal. The human inputs‌​‌ will first be used​​ to infer the operator​​​‌ intent and adapt the‌ robot behavior. Then, the‌​‌ shared action will combine​​ robot skills and human​​​‌ commands into a unified‌ and consistent control objective.‌​‌

Project in a nutshell:​​

• Consortium: AUCTUS@Inria (coordinator -​​​‌ Margot Vulliez), RoBioSS@Pprime (CNRS),‌ Interactions@CeRCA (CNRS)
• Funding: ANR‌​‌ Funding (287 840 Euros)​​
• Duration: 2021-2026

10.3.7 LAAS-AUCTUS​​​‌ collaborations

Participants: Thomas Flayols‌, Nicolas Mansard,‌​‌ Vincent Bonnet, Virgile​​ Batto, Margot Vulliez​​​‌, David Daney.‌

We have built a‌​‌ close scientific relationship with​​ the Gepetto team at​​​‌ LAAS CNRS (Toulouse) these‌ past few years, through‌​‌ several collaborative projects:

Legged-robot​​ codesign (2022-2025): This PhD​​​‌ project aimed at developing‌ a generic codesign approach‌​‌ to cover the hardware​​ specification and dimensioning and​​​‌ the control strategy and‌ requirements at once. We‌​‌ proposed to leverage mastered​​ AI-based methods (simulation, planning,​​​‌ optimization) to guide the‌ mechatronic design cycles and‌​‌ to provide tools to​​ assist designers. The transversal​​​‌ approach was applied to‌ the design and prototyping‌​‌ of BIPETTO a new​​ parallel-actuated kid-size legged robot.​​​‌ Inverse Optimal Control: A‌ collaboration on Inverse Optimal‌​‌ Control problem has started​​ in pursuit of the​​​‌ results obtained previously.

Project‌ in a nutshell:

• Consortium:‌​‌ AUCTUS@Inria, GEPETTO@LAAS (CNRS)
• Funding:​​ None
• Duration: 2021-Ongoing

10.3.8​​​‌ National visits to the‌ team

• Ludovic de Mattéis,‌​‌ LAAS-CNRS, PhD Student, Toulouse,​​ gave on April 28th​​​‌ 2025 a presentation on‌ "Optimal control of walkers‌​‌ with parallel actuation."

10.4.1 AAPR​​​‌ Perception-HRI, Improving perceptual information‌ in human-robot interaction

Participants:‌​‌ Margot Vulliez, Benjamin​​ Camblor, Remi Lafitte​​​‌, Cécile Scotto.‌

This regional project completes‌​‌ the ASAP-HRC ANR objectives​​ with additional cognitive studies​​​‌ to improve the exchange‌ of perceptive information during‌​‌ Human-Robot interactions. Such an​​ exchange of information between​​​‌ the agents is required‌ to communicate and coordinate‌​‌ together. We particularly focus​​ on visual and haptic​​​‌ feedbacks, related to the‌ task, the context, or‌​‌ the robot assistance, and​​ given through a teleoperation​​​‌ device to perform an‌ industrial task. Only a‌​‌ fine analysis and modeling​​ of the human multisensory​​​‌ perception and integration processes‌ can provide practical guidelines‌​‌ to determine the optimal​​​‌ mixture of feedbacks to​ implement in the human-robot​‌ interface. The project therefore​​ aims at developing personalized​​​‌ mathematical models of the​ perceptive and sensorimotor integration​‌ of visuo-haptic informations, in​​ interaction scenarios with a​​​‌ robot. We additionally investigate​ how different haptic guidance​‌ can be perceived by​​ a user, particularly studying​​​‌ the effect of legibility​ and predictability of a​‌ guidance path on performances,​​ comfort or trust in​​​‌ the assistance.

Project in​ a nutshell:

• Consortium: AUCTUS@Inria​‌ (coordinator - Margot Vulliez),​​ RoBioSS@Pprime (CNRS), Interactions@CeRCA (CNRS)​​​‌
• Funding: Région Nouvelle Aquitaine​
• Duration: 2022-2025

11.1.1 Scientific events: organisation​​​‌

11.1.2​​​‌ Scientific events: selection

11.1.3 Journal​​​‌

11.1.4​​​‌ Invited talks

• Margot Vulliez​ : Joint GDR day,​‌ GDR Robotique and GDR​​ IHM, November. Toward human-centered​​​‌ haptic design.
• Marot Vulliez​ , Alexis Boulay ,​‌ Elio Jabbour : GDR​​ day on Shared Control,​​​‌ TS4 Humans and Robots,​ January. Unified methodologies for​‌ adaptive shared control and​​ haptic guidance.
• Alicia Barsacq​​​‌ : JNRR, Rennes, 30​ September, nvolvement of Predictive​‌ Mechanisms in Human-Robot Joint​​ Action: the Lens of​​​‌ Agency.
• Sebastien Kleff :​ LAAS, Gepetto Team, Toulouse,​‌ May. From force-feedback MPC​​ to tactile servoing.
• Sebastien​​​‌ Kleff : LIRMM, IDH​ Team, Montpellier, December. From​‌ force-feedback MPC to tactile​​ servoing.
• Sebastien Kleff :​​​‌ CRISTAL, DEFROST Team, Lille,​ December. From force-feedback MPC​‌ to tactile servoing.
• Jacques​​ Zhong : IRiS lab,​​​‌ KAIST, Daejeon (South Korea),​ October. Adapting robotic systems​‌ to humans through dynamic​​ shared control and multi-morphological​​​‌ simulation of virtual humans.​
• Vincent Padois : Nov​‌ 2025 Invited seminar at​​ Longlab, Atlantic Technological University​​​‌ of Galway: Getting the​ Human in the control​‌ loop of the collaborative​​ robot (and vice-versa). Organizer:​​ Prof. Philip Long (ATU).​​​‌ Galway, Ireland
• Vincent Padois‌ : Oct 2025 Invited‌​‌ seminar at the JNRR​​ workshop session (Journées Nationales​​​‌ de la Recherche en‌ Robotique): Getting the Human‌​‌ in the control loop​​ of the collaborative robot​​​‌ (and vice-versa). Organizer: Dr.‌ Claudio Pachierotti (CNRS). Rennes,‌​‌ France

11.1.5 Leadership within​​ the scientific community

• Margot​​​‌ Vulliez is co-coordinator of‌ the Support Action "Hardware"‌​‌ of the PEPR Robotics​​ Acceleration together with Abderrahmane​​​‌ Kheddar (LIRMM) and Olivier‌ Ly (Labri Rhoban).
• Pierre‌​‌ Puchaud is part of​​ the biomechanics groups within​​​‌ INRIA, responsible for software‌
• Pierre Puchaud is in‌​‌ charge of the Bioptim​​ library: Maintainer, conducting weekly​​​‌ development reviews to guarantee‌ software robustness.
• Vincent Padois‌​‌ : Leader of the​​ advisory committee for the​​​‌ creation of a University‌ Hub for Advanced Robotics‌​‌ Sciences (PULSAR) under the​​ SIN department of the​​​‌ University of Bordeaux.
• Vincent‌ Padois member of the‌​‌ GDR Robotics thesis award​​ committee.
• Vincent Padois Representative​​​‌ of the Inria program‌ agency and member of‌​‌ the creation committee of​​ the PEPR “Accélération Robotique”.​​​‌
• Vincent Padois Co-leader of‌ the Priority Axis 1‌​‌ “Robotique et Sobriété” of​​ the GDR Robotics.
• David​​​‌ Daney is a member‌ of the Scientific Council‌​‌ of the GDR Robotics,​​ the main organization for​​​‌ robotics researchers in France.‌
• David Daney David participates‌​‌ in the Inria Master​​ Class.

11.1.6 Scientific expertise​​​‌

• David Daney is a‌ member of the executive‌​‌ board of the "Aquitaine​​ robotics" cluster, which brings​​​‌ together robotics players in‌ Nouvelle-Aquitaine.
• David Daney represents‌​‌ Inria on the board​​ of directors the "Aquitaine​​​‌ robotics" cluster.
• David Daney‌ is the coordinator of‌​‌ 2 working groups (GT)​​ on the evaluation and​​​‌ on the creation of‌ 2 Inria Projet-Teams for‌​‌ the Inria Evaluation Committee.​​
• Jean-Marc Salotti represents Ensc​​​‌ on the board of‌ directors of the "Aquitaine‌​‌ robotics" cluster.
• Margot Vulliez​​ was external reviewer for​​​‌ the evaluation of a‌ research proposal at the‌​‌ Swiss National Science Foundation​​ SNSF, 2025
• Vincent Padois​​​‌ was external reviewe for‌ the UP-SQUARED research call,‌​‌ PIA4 ExcellenceS Univ. Poitiers.​​
• Jean-Marc Salotti was a​​​‌ scientific advisor for the‌ Cap Sciences exhibition dedicated‌​‌ to the exploration of​​ the Moon.

11.1.7 Research​​​‌ administration

• David Daney is‌ the Head of Science‌​‌ at Inria centre at​​ the University of Bordeaux​​​‌
• David Daney is a‌ member of of the‌​‌ Inria Evaluation Committee.
• David​​ Daney is a member​​​‌ of the France 2030‌ Robotics Committee.
• David Daney‌​‌ is member of the​​ "Commission des Emplois de​​​‌ Recherche" for the Inria‌ centre at the University‌​‌ of Bordeaux.
• David Daney​​ is member of the​​​‌ "Comité de Centre" of‌ the Inria centre at‌​‌ the University of Bordeaux.​​
• David Daney is a​​​‌ member of the executive‌ board of R4, a‌​‌ regional robotics network involving​​ 12 research entities in​​​‌ the region of Nouvelle-Aquitaine,‌ France.
• David Daney is‌​‌ the principal investigator of​​ DTR Grip4All ANR/France 2030​​​‌
• David Daney is the‌ co-investigator of Topic Priority‌​‌ DRMI within the PEPR​​ Robotics
• Margot Vulliez is​​​‌ member of the "Commission‌ des Emplois de Recherche"‌​‌ for Inria centre at​​​‌ the University of Bordeaux,​ since January 2024.
• Pierre​‌ Puchaud is member of​​ the CUMI (Communauté des​​​‌ Utilisateurs des Moyens Informatiques):​ Representative for the Auctus​‌ team at the INRIA​​ Bordeaux center.
• Vincent Padois​​​‌ is president of the​ Technological Developments Commission (CDT)​‌ at Inria Bordeaux.

11.2.1​‌ Supervision

11.2.2 Juries​​​‌

11.2.3 Educational​​ and pedagogical outreach

• As​​​‌ part of the "Fête‌ de la Science" week,‌​‌ Jean-Marc Salotti offered about​​​‌ thirty high school students​ a robotics game designed​‌ to help them better​​ understand the degrees of​​​‌ freedom of a robotic​ arm.

11.3.1​‌ Specific official responsibilities in​​ science outreach structures

• Vincent​​​‌ Padois : Assigned to​ scientific mediation for Inria​‌ Bordeaux and member of​​ the steering committee of​​​‌ the Aquitaine external scientific​ circuit

11.3.2 Productions (articles,​‌ videos, podcasts, serious games,​​ ...)

• Jean-Marc Salotti took​​​‌ part in a video​ on the Curieux YouTube​‌ channel as part of​​ the presentation of the​​​‌ Cap Sciences exhibition dedicated​ to the exploration of​‌ the Moon. Youtube video​​
• Margot Vulliez was interviewed​​​‌ for the content of​ Cap.Sciences workshop "Les Femmes​‌ Scientifiques" on Sciences.live, September​​ 2025
• Margot Vulliez was​​​‌ interviewed for the Inria​ portrait series « Elles​‌ font le numérique »,​​ June 2025
• Pierre Puchaud​​​‌ was speaker at the​ INRIA Diggest seminar, "Un​‌ parcours en Biomécanique pour​​ rejoindre Auctus", July 2025​​​‌

11.3.3 Participation in Live​ events

International journals

• 6 article‌O.Olivier David,‌​‌ V.Victor Dhédin,​​ J.Julie Dumora,​​​‌ V.Vincent Padois,‌ A.Adib Rebbouh and‌​‌ F.Ferdinando Milella.​​ Using programming by demonstration​​​‌ tools for DEMO maintenance:‌ a framework proposal with‌​‌ an experimental validation.​​Fusion Engineering and Design​​​‌2152025, 114935‌HALDOI
• 7 article‌​‌É.Étienne Fournier-Aubret,​​ A.Aurélie Landry,​​​‌ B.Beatrice Piras,‌ D.Damien Pellier,‌​‌ H.Humbert Fiorino,​​ D.David Daney and​​​‌ C.Christine Jeoffrion.‌ Proof of concept of‌​‌ a cobotic system in​​ a constrained work environment​​​‌.Applied Ergonomics125‌May 2025, 104472‌​‌HALDOI
• 8 article​​N.Nasser Rezzoug,​​​‌ A.Antun Skuric,‌ V.Vincent Padois and‌​‌ D.David Daney.​​ Simulation Study of the​​​‌ Upper-Limb Isometric Wrench Feasible‌ Set With Glenohumeral Joint‌​‌ Constraints.Journal of​​ Biomechanical Engineering1472​​​‌February 2025, 024501‌HALDOIback to‌​‌ text
• 9 articleN.​​Nicolas Simonazzi, J.-M.​​​‌Jean-Marc Salotti, C.‌Caroline Dubois and P.‌​‌Philippe Le Goff.​​ Automatic classification of emotions​​​‌ using motion sensors and‌ keystroke dynamics on smartphones‌​‌.Ingénierie cognitique8​​1July 2025,​​​‌ 1-23HALDOIback‌ to text
• 10 article‌​‌A.Antun Skuric,​​ N. T.Nicolas Torres​​​‌ Alberto, L.Lucas‌ Joseph, V.Vincent‌​‌ Padois and D.David​​ Daney. Online approach​​​‌ to near time-optimal task-space‌ trajectory planning.IEEE‌​‌ Transactions on RoboticsDecember​​ 2025, 1-16HAL​​​‌DOIback to text‌

International peer-reviewed conferences

• 11‌​‌ inproceedingsM.Marc Duclusaud​​, G.Grégoire Passault​​​‌, V.Vincent Padois‌ and O.Olivier Ly‌​‌. Extended Friction Models​​ for the Physics Simulation​​​‌ of Servo Actuators.‌ICRA 2025 - IEEE/RAS‌​‌ International Conference on Robotics​​​‌ and AutomationAtlanta, United​ States2025HALback​‌ to text
• 12 inproceedings​​J.-M.Jean-Marc Salotti.​​​‌ Qualitative Analysis of Mars​ Entry and Descent Options​‌ for Manned Missions.​​Proceedings of the 11th​​​‌ European Conference for AeroSpace​ SciencesEUCASS 2025 -​‌ 11th European Conference for​​ AeroSpace SciencesRome, Italy​​​‌June 2025, 2797-2804​HAL

Edition (books, proceedings,​‌ special issue of a​​ journal)

• 13 proceedingsJ.-M.​​​‌Jean-Marc Salotti, eds.​ Proceedings of the 2025​‌ European Mars Conference, EMC2025​​.EMC 2025 -​​​‌ European Mars ConferenceParis,​ FranceNovember 2025HAL​‌

Doctoral dissertations and habilitation​​ theses

• 14 thesisA.​​​‌Alexis Boulay. Assisting​ humans through adaptive haptic​‌ guidance, application to remote-controlled​​ vertical farming.Université​​​‌ de BordeauxSeptember 2025​HALback to text​‌
• 15 thesisE.Elio​​ Jabbour. Shared-autonomy control​​​‌ for improving Human-Robot collaboration​ in haptic teleoperation.​‌Université de BordeauxNovember​​ 2025HAL

Reports &​​​‌ preprints

• 16 miscS.​Sébastien Kleff, L.​‌Lucas Joseph and V.​​Vincent Padois. A​​​‌ Unified Tactile Servoing Framework​ based on Hybrid Force-Position​‌ Control.January 2026​​HALback to text​​​‌
• 17 miscG.Guillaume​ de Mathelin de Papigny​‌, F.Francesco Gassibe​​ and V.Vincent Padois​​​‌. F-RRT: an Efficient​ Algorithm for Semi-Constrained Path​‌ Planning Problems.November​​ 2025HAL
• 18 misc​​​‌T. D.Thanh D​ V Nguyen, V.​‌Vincent Bonnet, P.​​Pierre Fernbach, D.​​​‌David Daney and F.​Florent Lamiraux. Humanoid​‌ Robot Whole-body Geometric Calibration​​ with Embedded Sensors and​​​‌ a Single Plane.​July 2025HAL
• 19​‌ miscK.Krzysztof Wojciechowski​​, E.Ege Gursoy​​​‌, A.Arthur Haffemayer​, S.Sébastien Kleff​‌, V.Vincent Bonnet​​, F.Florent Lamiraux​​​‌ and N.Nicolas Mansard​. Learning-Guided Force-Feedback Model​‌ Predictive Control with Obstacle​​ Avoidance for Robotic Deburring​​​‌.October 2025HAL​

Other scientific publications

• 20​‌ inproceedingsA.Alicia Barsacq​​, D.David Daney​​​‌ and J.-C.Jean-Christophe Sarrazin​. Involvement of Predictive​‌ Mechanisms in Human-Robot Joint​​ Action: the Lens of​​​‌ Agency.JJCR 2025​ - Journée des Jeunes​‌ Chercheuses et Jeunes Chercheurs​​ en RobotiqueRennes, France​​​‌September 2025HAL
• 21​ inproceedingsA.Alexis Boulay​‌ and R.Romain Schmitt​​. Toward a human-centered​​​‌ haptic guidance.Journée​ TS4 du GdR Robotique​‌Paris, FranceJanuary 2025​​HAL
• 22 inproceedingsR.​​​‌Rémi Lafitte, M.​Margot Vulliez and C.​‌Cécile Scotto. Ambiguous​​ visuo-haptic integration in a​​​‌ teleoperation environment.IMRF​ 2025 - 23rd International​‌ Multisensory Research ForumDurham​​ (GB), United KingdomJuly​​​‌ 2025HALback to​ text