Balance is an essential feature of humanoids but, despite a strong understanding of its laws and dynamics, it remains an open problem for control applications. Optimization-based control approaches explicitly include balance dynamics and constraints in the control problem in order to capture at best the behavior of the system and fully exploit it to reach complex control objectives. Although theoretically appealing, these approaches intrinsically induce a significant computational burden. In practice, this implies to resort to simplifications on the model and problem complexities, which limits the capacity to actually generate complex behaviors. In this chapter, an overview of the balance problem is first proposed. A general, abstract formulation of the balance control problem as an optimal control one is then derived. Three major approaches can be found in the literature, coping with the computational complexity of the general balance optimization problem. They range from offline motion planning to reactive whole-body control and are presented in the remainder of the chapter.

This paper proposes a novel mechanical design of a lower limb exoskeleton device which prevents the residual stresses due to arthro-kinematics movements of synovial joints and by the way allows effective compensation for dynamic disturbances in osteo-kinematic movements of the wearer. Here, the exoskeleton is only actuated at the knee joints to provide assistive torques, which are required to assist the anatomical joint motion and to increase the transparency of the device. Dynamic simulations of a virtual human equipped with this exoskeleton are used to quantify the disturbances induced by the device during locomotion and to show the benefit of passive mechanisms introduced in the mechanical attaches as well. The authors also demonstrated how the device's transparency can be improved by providing the motor torques in order to compensate the inertial and gravitational effects. This can be done rely on the knowledge of the locomotion movement phases. A robust gait phase detection method was implemented on the experimental device in order to identify specific gait phases in real time. This method exploits the K-nearest neighbors algorithm to identify the k-closest trained vectors, coupling with a discrete time Markov chain to determine the phases shift probability during the gait cycle. This gait detection algorithm was tested with a percentage of success of more than 95% when the subjects walked with constant and variable stride lengths.

This paper considers the central question of transparency in a lower limb exoskeleton, designed for knee rehabilitation. The device is aimed to provide torque assistance for the user's knee joint during walking. The mechanical design is realized according to the methodology established for exoskeletons, which consists of using passive linkages to connect the device's external rigid structure to the human body. Torque controllers are implemented at the knee joints. A gait phase detection method is investigated in order to provide an effective control of the system. First simulations and experiments were conducted, showing the effect of the device on the wearer during locomotion.

Rising to the challenge of motor control for systems involved in multi-objective and highly-constrained activities is a requirement to enable the emergence of efficient and robust behaviors; the elaboration of complex motor coordination strategies is critical in ensuring performance, feasibility and safety.Although multi-objective predictive approaches enable the definition of complex and constrained strategies coordinating the motor activity of the system, their computational cost is a critical drawback from practical applications.The work presented in this dissertation aims at considering multi-objective predictive control for feasible and practical applications to humanoid robotics.A control architecture is proposed to this purpose as a multi-objective, two-layered controller exploiting the respective advantages of predictive and instantaneous formulations.The contribution of this work takes the form of the validation of the benefits from such an approach in its development for practical challenges and applications, in simulation and real-time implementation, on the iCub and TORO robots and virtual human models.Computational demand of the predictive level is contained with the introduction of reduced multi-objective predictive problems, enabling computationally-favorable formulations of the control problem using mixed-integer programming and sequential and parallel distributions.Despite the resulting approximations on the dynamics of the system at the predictive level, complex behaviors are emerging, exploiting elaborate coordination strategies between conflicting objectives and constraints to increase performance and robustness against disturbances.

A deep understanding of human activity is key to successful human-robot interaction (HRI). The translation of sensed human behavioural signals/cues and context descriptors into an encoded human activity remains a challenge because of the complex nature of human actions. In this paper, we propose a multilayer framework for the understanding of human activity to be implemented in a mobile robot. It consists of a perception layer which exploits a D-RGB-based skeleton tracking output used to simulate a physical model of virtual human dynamics in order to compensate for the inaccuracy and inconsistency of the raw data. A multi-support vector machine (MSVM) model trained with features describing the human motor coordination through temporal segments in combination with environment descriptors (object affordance) is used to recognize each sub-activity (classification layer). The interpretation of sequences of classified elementary actions is based on discrete hidden Markov models (DHMMs) (interpretation layer). The framework assessment was performed on the Cornell Activity Dataset (CAD-120) [1]. The performances of our method are comparable with those presented in [2] and clearly show the relevance of this model-based approach.

Balance strategies range from continuous postural adjustments to discrete changes in contacts: their simultaneous execution is required to maintain postural stability while considering the engaged walking activity. In order to compute optimal time, duration and position of footsteps along with the center of mass trajectory of a humanoid, a novel mixed-integer model of the system is presented. The introduction of this model in a predictive control problem brings the definition of a Mixed-Integer Quadratic Program, subject to linear constraints. Simulation results demonstrate the simultaneous adaptation of the gait pattern and posture of the humanoid, in a walking activity under large disturbances, to efficiently compromise between task performance and balance. In addition, a push recovery scenario displays how, using a single balance-performance ratio, distinct behaviors of the humanoid can be specified.

This chapter proposes an original Model Predictive Control approach to the walking control for humanoid robots, which allows to generate stable walking motions without the prior definition of footsteps positions and instants. Both the instant and amplitude of the changes in the supporting surface are part of the walking motion generation problem, and are described by a set of highly-constrained integer and real variables. Combined with the center of mass trajectory of the robot, this description leads to the formulation of a Mixed-Integer Quadratic Program in a Model Predictive Control framework aiming at reaching high-level objectives, such as velocity tracking and tip-over riskminimization. The contribution of this approach is illustrated by the simulation of two scenarii, demonstrating the validity of the steps and trajectories computed in push-recovery and walking velocity tracking cases.

A novel formulation of the synthesis of motor coordination for humanoid whole-body motion is proposed in this paper, in order to ensure robust control of postural stability. It relies on the distributed model predictive control framework to coordinate, in an optimal way, several objectives. The effectiveness of this control technique to maintain postural stability of a biped against strong external disturbances is shown. Control of the horizontal dynamics of the center of mass can withstand limited perturbations. Thus postural stability criteria are specified with respect to the robot center of mass vertical and horizontal dynamics, and to the angular dynamics of its torso. Formulating the balance problem in a predictive form and distributing at different time scales significantly increases the robustness of the system to external disturbances, in terms of both tip-over and slippage risks. This original control architecture is validated through the simulation of an iCub robot performing a walking activity under unknown external actions.

This paper proposes a preview control method for the whole-body motion of humanoid robots ensuring high performance in both interaction and postural balance tasks under large physical perturbations.By previewing the reduced coupled models of upper-limb interaction and postural balance dynamics, the proposed controller adapts simultaneously the impedance of the arms and the center of mass trajectory with respect to known external perturbations.Here, we show how the ZMP preview control formulation can be extended to account for disturbances resulting from the interacting arms dynamics of which control parameters are adapted online in order to maximize both interaction and balance performances over a preview horizon.The validity of this formulation is assessed through simulation considering a force applied at the humanoid hand level when it is walking.

This paper summarizes the motivation in mutual understanding of human activities and those of the robot. We then introduce successively control techniques for creating a repertoire of purely reactive sensorimotor functions (based on convex optimization techniques) and anticipatory from distributed predictive control methods. We then consider how to develop the robotic system the foresight to more or less long term to develop action plans and their implementation plan based on the principles of ideomotor integration.

This paper proposes a robust whole-body control formulation for biped balance in disturbed conditions by manipulation tasks. In order to include the effects of the interaction of the robot with its environment, required by the manipulation task in the balance control, we introduce a distributed preview control which captures both balance and manipulation behaviors and enables the regulation of the interaction impedance. The initial ZMP preview control is extended to take into account the disturbance resulting from the manipulation task and the preview control of adaptive impedances used to drive the upper-limbs. The resulting behavior is illustrated in a simple scenario. Its aptitude to dynamically extract an optimal control strategy improving tracking performances of both manipulation and balance tasks is also assessed when complex perturbations have to be compensated.

des établissements d'enseignement et de recherche français ou étrangers, des laboratoires publics ou privés.