This work presents a description of the EXOSMOOTH project, oriented to the benchmarking of lower limb exoskeletons performance. In the field of assisted walking by powered lower limb exoskeletons, the EXOSMOOTH project proposes an experiment that targets two scientific questions. The first question is related to the effectiveness of a novel control strategy for smooth assistance. Current assist strategies are based on controllers that switch the assistance level based on the gait segmentation provided by a finite state machine. The proposed strategy aims at managing phase transitions to provide a smoother assistance to the user, thus increasing the device transparency and comfort for the user. The second question is the role of the actuation at the ankle joint in assisted walking. Many novel exoskeletons devised for industrial applications do not feature an actuated ankle joint. In the EXOSMOOTH project, the ankle joint actuation will be one experimental factor to have a direct assessment of the role of an actuated joint in assisted walking. Preliminary results of 15 healthy subjects walking at different speeds while wearing a lower limb exoskeleton supported the rationale behind this question: having an actuated ankle joint could potentially reduce the torques applied by the user by a maximum value of 85 Nm. The two aforementioned questions will be investigated in a protocol that includes walking on a treadmill and on flat ground, with or without slope, and with a load applied on the back. In addition, the interaction forces measured at the exoskeleton harnesses will be used to assess the comfort of the user and the effectiveness of the control strategy to improve transparency.

Lower limbs exoskeletons provide assistance during standing, squatting, and walking. Gait dynamics, in particular, implies a change in the configuration of the device in terms of contact points, actuation, and system dynamics in general. In order to provide a comfortable experience and maximize performance, the exoskeleton should be controlled smoothly and in a transparent way, which means respectively, minimizing the interaction forces with the user and jerky behavior due to transitions between different configurations. A previous study showed that a smooth control of the exoskeleton can be achieved using a gait phase segmentation based on joint kinematics. Such a segmentation system can be implemented as linear regression and should be personalized for the user after a calibration procedure. In this work, a nonlinear segmentation function based on neural networks is implemented and compared with linear regression. An on-line implementation is then proposed and tested with a subject.

This work presents a system identification procedure based on Convolutional Neural Networks (CNN) for human posture control using the DEC (Disturbance Estimation and Compensation) parametric model. The modular structure of the proposed control model inspired the design of a modular identification procedure, in the sense that the same neural network is used to identify the parameters of the modules controlling different degrees of freedom. In this way the presented examples of body sway induced by external stimuli provide several training samples at once

This work presents a system identification procedure based on Convolutional Neural Networks (CNN) for human posture control using the DEC (Disturbance Estimation and Compensation) parametric model. The modular structure of the proposed control model inspired the design of a modular identification procedure, in the sense that the same neural network is used to identify the parameters of the modules controlling different degrees of freedom. In this way the presented examples of body sway induced by external stimuli provide several training samples at once.

Human posture control models are used to analyse neurological experiments and control of humanoid robots. This work focuses on a well-known nonlinear posture control model, the DEC (Disturbance estimate and Compensation). In order to compensate disturbances, unlike other models, DEC feedbacks signals coming from sensor fusion rather than raw sensory signals. In previous works, the DEC model is shown to predict human behavior and to provide a control system for humanoids. In this work, the stability of the system in the sense of Lyapunov is formally analysed. The theoretical findings are combined with simulation results, in which an external perturbation of the support surface reproduces a typical scenario in posture control experiments.

This paper describes some applications of an incremental implementation of the principal component analysis (PCA). The algorithm updates the transformation coefficients matrix on-line for each new sample, without the need to keep all the samples in memory. The algorithm is formally equivalent to the usual batch version, in the sense that given a sample set the transformation coefficients at the end of the process are the same. The implications of applying the PCA in real time are discussed with the help of data analysis examples. In particular we focus on the problem of the continuity of the PCs during an on-line analysis.

Probabilistic Movement Primitives (ProMPs) are a widely used representation of movements for human-robot interaction. They also facilitate the factorization of temporal and spatial structure of movements. In this work we investigate a method to temporally align observations so that when learning ProMPs, information in the spatial structure of the observed motion is maximized while maintaining a smooth phase velocity. We apply the method on recordings of hand trajectories in a two-dimensional reaching task. A system for simultaneous recognition of movement and phase is proposed and performance of movement recognition and movement reproduction is discussed.

This paper validates experimentally a humanoid posture control concept from neuroscience, called disturbance estimation and compensation, DEC concept. The DEC control system, different from typical state estimation systems, is not including a dynamic model of the body. Also, among human posture control models it is particular in that it uses feedback of multisensory disturbance estimates for compensation, rather than ‘raw’ sensory signals. To this end, the system performs fusions of sensory inputs such as vestibular inputs (IMU) and proprioceptive inputs (joint position and speed). The compensation of external disturbances allows the control to use low loop gain, with human-like tolerance of time delays and mechanical compliance. This paper validates the control concept experimentally, measuring the balancing of biped stance of a humanoid 2 DOF robot, Posturob II, while superimposing on support surface tilt either voluntary trunk bending or push stimuli. The results show that the control concept is robust and able to stabilize the robot's balance in complex disturbance conditions. Furthermore, several human-like features such as hip-ankle coordination emerged from the control concept.

In this work we present a method to produce a model of human motion based on an expansion in functions series. The model is thought to reproduce the learned movements generalizing them to different conditions. We will show, with an example, how the proposed method is capable to produce the model from a reduced set of examples preserving the relevant features of the demonstrations while guaranteeing constraints at boundaries.

We analyze the impact of the delay between the sensors and the visualization on the performance at a catching task in virtual environment. Testers have been asked to interact with a virtual environment displayed on a 2D screen in front of them, catching, with the virtual hand, virtual balls falling from the top of the screen. Two thresholds have been inferred: a behavioral threshold delay beyond which the behavioral organization is sensibly affected and a performance threshold delay beyond which task performance begins to worsen. We found a drop in performance when the total delay is over 70-80 ms, yet without significant changes in the organization of successful catches. Although the presence of a threshold on behavior is still to be further investigated, these two thresholds give a practical indication for the design and the validation of virtual reality based training systems. This is especially important in the context of training human skills in virtual environment: in particular to assess the quality of a training platform simulating the three ball cascade juggling pattern.