Learning agile soccer skills for a bipedal robot with deep reinforcement learning.

We investigated whether deep reinforcement learning (deep RL) is able to synthesize sophisticated and safe movement skills for a low-cost, miniature humanoid robot that can be composed into complex behavioral strategies. We used deep RL to train a humanoid robot to play a simplified one-versus-one soccer game. The resulting agent exhibits robust and dynamic movement skills, such as rapid fall recovery, walking, turning, and kicking, and it transitions between them in a smooth and efficient manner. It also learned to anticipate ball movements and block opponent shots. The agent's tactical behavior adapts to specific game contexts in a way that would be impractical to manually design. Our agent was trained in simulation and transferred to real robots zero-shot. A combination of sufficiently high-frequency control, targeted dynamics randomization, and perturbations during training enabled good-quality transfer. In experiments, the agent walked 181% faster, turned 302% faster, took 63% less time to get up, and kicked a ball 34% faster than a scripted baseline.

Simultaneous Floating-Base Estimation of Human Kinematics and Joint Torques.

The paper presents a stochastic methodology for the simultaneous floating-base estimation of the human whole-body kinematics and dynamics (i.e., joint torques, internal and external forces). The paper builds upon our former work where a fixed-base formulation had been developed for the human estimation problem. The presented approach is validated by presenting experimental results of a health subject equipped with a wearable motion tracking system and a pair of shoes sensorized with force/torque sensors while performing different motion tasks, e.g., walking on a treadmill. The results show that joint torque estimates obtained by using floating-base and fixed-base approaches match satisfactorily, thus validating the present approach.

Whole-Body Human Inverse Dynamics with Distributed Micro-Accelerometers, Gyros and Force Sensing.

Human motion tracking is a powerful tool used in a large range of applications that require human movement analysis. Although it is a well-established technique, its main limitation is the lack of estimation of real-time kinetics information such as forces and torques during the motion capture. In this paper, we present a novel approach for a human soft wearable force tracking for the simultaneous estimation of whole-body forces along with the motion. The early stage of our framework encompasses traditional passive marker based methods, inertial and contact force sensor modalities and harnesses a probabilistic computational technique for estimating dynamic quantities, originally proposed in the domain of humanoid robot control. We present experimental analysis on subjects performing a two degrees-of-freedom bowing task, and we estimate the motion and kinetics quantities. The results demonstrate the validity of the proposed method. We discuss the possible use of this technique in the design of a novel soft wearable force tracking device and its potential applications.

Motor contagion during human-human and human-robot interaction.

Motor resonance mechanisms are known to affect humans' ability to interact with others, yielding the kind of "mutual understanding" that is the basis of social interaction. However, it remains unclear how the partner's action features combine or compete to promote or prevent motor resonance during interaction. To clarify this point, the present study tested whether and how the nature of the visual stimulus and the properties of the observed actions influence observer's motor response, being motor contagion one of the behavioral manifestations of motor resonance. Participants observed a humanoid robot and a human agent move their hands into a pre-specified final position or put an object into a container at various velocities. Their movements, both in the object- and non-object- directed conditions, were characterized by either a smooth/curvilinear or a jerky/segmented trajectory. These trajectories were covered with biological or non-biological kinematics (the latter only by the humanoid robot). After action observation, participants were requested to either reach the indicated final position or to transport a similar object into another container. Results showed that motor contagion appeared for both the interactive partner except when the humanoid robot violated the biological laws of motion. These findings suggest that the observer may transiently match his/her own motor repertoire to that of the observed agent. This matching might mediate the activation of motor resonance, and modulate the spontaneity and the pleasantness of the interaction, whatever the nature of the communication partner.

The ITALK project: a developmental robotics approach to the study of individual, social, and linguistic learning.

This article presents results from a multidisciplinary research project on the integration and transfer of language knowledge into robots as an empirical paradigm for the study of language development in both humans and humanoid robots. Within the framework of human linguistic and cognitive development, we focus on how three central types of learning interact and co-develop: individual learning about one's own embodiment and the environment, social learning (learning from others), and learning of linguistic capability. Our primary concern is how these capabilities can scaffold each other's development in a continuous feedback cycle as their interactions yield increasingly sophisticated competencies in the agent's capacity to interact with others and manipulate its world. Experimental results are summarized in relation to milestones in human linguistic and cognitive development and show that the mutual scaffolding of social learning, individual learning, and linguistic capabilities creates the context, conditions, and requisites for learning in each domain. Challenges and insights identified as a result of this research program are discussed with regard to possible and actual contributions to cognitive science and language ontogeny. In conclusion, directions for future work are suggested that continue to develop this approach toward an integrated framework for understanding these mutually scaffolding processes as a basis for language development in humans and robots.

Muscle synergies in neuroscience and robotics: from input-space to task-space perspectives.

In this paper we review the works related to muscle synergies that have been carried-out in neuroscience and control engineering. In particular, we refer to the hypothesis that the central nervous system (CNS) generates desired muscle contractions by combining a small number of predefined modules, called muscle synergies. We provide an overview of the methods that have been employed to test the validity of this scheme, and we show how the concept of muscle synergy has been generalized for the control of artificial agents. The comparison between these two lines of research, in particular their different goals and approaches, is instrumental to explain the computational implications of the hypothesized modular organization. Moreover, it clarifies the importance of assessing the functional role of muscle synergies: although these basic modules are defined at the level of muscle activations (input-space), they should result in the effective accomplishment of the desired task. This requirement is not always explicitly considered in experimental neuroscience, as muscle synergies are often estimated solely by analyzing recorded muscle activities. We suggest that synergy extraction methods should explicitly take into account task execution variables, thus moving from a perspective purely based on input-space to one grounded on task-space as well.

Evidence for composite cost functions in arm movement planning: an inverse optimal control approach.

An important issue in motor control is understanding the basic principles underlying the accomplishment of natural movements. According to optimal control theory, the problem can be stated in these terms: what cost function do we optimize to coordinate the many more degrees of freedom than necessary to fulfill a specific motor goal? This question has not received a final answer yet, since what is optimized partly depends on the requirements of the task. Many cost functions were proposed in the past, and most of them were found to be in agreement with experimental data. Therefore, the actual principles on which the brain relies to achieve a certain motor behavior are still unclear. Existing results might suggest that movements are not the results of the minimization of single but rather of composite cost functions. In order to better clarify this last point, we consider an innovative experimental paradigm characterized by arm reaching with target redundancy. Within this framework, we make use of an inverse optimal control technique to automatically infer the (combination of) optimality criteria that best fit the experimental data. Results show that the subjects exhibited a consistent behavior during each experimental condition, even though the target point was not prescribed in advance. Inverse and direct optimal control together reveal that the average arm trajectories were best replicated when optimizing the combination of two cost functions, nominally a mix between the absolute work of torques and the integrated squared joint acceleration. Our results thus support the cost combination hypothesis and demonstrate that the recorded movements were closely linked to the combination of two complementary functions related to mechanical energy expenditure and joint-level smoothness.

Finger impedance evaluation by means of hand exoskeleton.

Modulation of arm mechanical impedance is a fundamental aspect for interaction with the external environment and its regulation is essential for stability preservation during manipulation. Even though past research on human arm movements has suggested that models of human finger impedance would benefit the study of neural control mechanisms and the design of novel hand prostheses, relatively few studies have focused on finger and hand impedance. This article touches on the two main aspects of this research topic: first it introduces a mechanical refinement of a device that can be used to effectively measure finger impedance during manipulation tasks; then, it describes a pilot study aimed at identifying the inertia of the finger and the viscous and elastic properties of finger muscles. The proposed wearable exoskeleton, which has been designed to measure finger posture and impedance modulation while leaving the palm free, is capable of applying fast displacements while monitoring the interaction forces between the human finger and the robotic links. Moreover, due to the relatively small inertia of the fingers, it allows us to meet some stringent specifications, performing relatively large displacements (~45°) before the stretch reflex intervenes (~25 ms). The results of measurements on five subjects show that inertia, damping, and stiffness can be effectively identified and that the parameters obtained are comparable with values from previous studies.

Manifold reaching paradigm: how do we handle target redundancy?

How the central nervous system coordinates the many intrinsic degrees of freedom of the musculoskeletal system is a recurrent question in motor control. Numerous studies addressed it by considering redundant reaching tasks such as point-to-point arm movements, for which many joint trajectories and muscle activations are usually compatible with a single goal. There exists, however, a different, extrinsic kind of redundancy that is target redundancy. Many times, indeed, the final point to reach is neither specified nor unique. In this study, we aim to understand how the central nervous system tackles such an extrinsic redundancy by considering a reaching-to-a-manifold paradigm, more specifically an arm pointing to a long vertical bar. In this case, the endpoint is not defined a priori and, therefore, subjects are free to choose any point on the bar to successfully achieve the task. We investigated the strategies used by subjects to handle this presented choice. Our results indicate both intersubject and intertrial consistency with respect to the freedom provided by the task. However, the subjects' behavior is found to be more variable than during classical point-to-point reaches. Interestingly, the average arm trajectories to the bar and the structure of intertrial endpoint variations could be explained via stochastic optimal control with an energy/smoothness expected cost and signal-dependent motor noise. We conclude that target redundancy is first overcome during movement planning and then exploited during movement execution, in agreement with stochastic optimal feedback control principles, which illustrates how the complementary problems of goal and movement selection may be resolved at once.

The iCub humanoid robot: an open-systems platform for research in cognitive development.

We describe a humanoid robot platform--the iCub--which was designed to support collaborative research in cognitive development through autonomous exploration and social interaction. The motivation for this effort is the conviction that significantly greater impact can be leveraged by adopting an open systems policy for software and hardware development. This creates the need for a robust humanoid robot that offers rich perceptuo-motor capabilities with many degrees of freedom, a cognitive capacity for learning and development, a software architecture that encourages reuse & easy integration, and a support infrastructure that fosters collaboration and sharing of resources. The iCub satisfies all of these needs in the guise of an open-system platform which is freely available and which has attracted a growing community of users and developers. To date, twenty iCubs each comprising approximately 5000 mechanical and electrical parts have been delivered to several research labs in Europe and to one in the USA.

A control theory approach to the analysis and synthesis of the experimentally observed motion primitives.

Recent experiments on frogs and rats, have led to the hypothesis that sensory-motor systems are organized into a finite number of linearly combinable modules; each module generates a motor command that drives the system to a predefined equilibrium. Surprisingly, in spite of the infiniteness of different movements that can be realized, there seems to be only a handful of these modules. The structure can be thought of as a vocabulary of "elementary control actions". Admissible controls, which in principle belong to an infinite dimensional space, are reduced to the linear vector space spanned by these elementary controls. In the present paper we address some theoretical questions that arise naturally once a similar structure is applied to the control of nonlinear kinematic chains. First of all, we show how to choose the modules so that the system does not loose its capability of generating a "complete" set of movements. Secondly, we realize a "complete" vocabulary with a minimal number of elementary control actions. Subsequently, we show how to modify the control scheme so as to compensate for parametric changes in the system to be controlled. Remarkably, we construct a set of modules with the property of being invariant with respect to the parameters that model the growth of an individual. Robustness against uncertainties is also considered showing how to optimally choose the modules equilibria so as to compensate for errors affecting the system. Finally, the motion primitive paradigm is extended to locomotion and a related formalization of internal (proprioceptive) and external (exteroceptive) variables is given.