Paolo Tommasino

Upper Extremity Proprioception in Healthy Aging and Stroke Populations, and the Effects of Therapist- and Robot-Based Rehabilitation Therapies on Proprioceptive Function

The world’s population is aging, with the number of people ages 65 or older expected to surpass 1.5 billion people, or 16% of the global total. As people age, there are notable declines in proprioception due to changes in the central and peripheral nervous systems. Moreover, the risk of stroke increases with age, with approximately two-thirds of stroke-related hospitalizations occurring in people over the age of 65. In this literature review, we first summarize behavioral studies investigating proprioceptive deficits in normally aging older adults and stroke patients, and discuss the differences in proprioceptive function between these populations. We then provide a state of the art review the literature regarding therapist- and robot-based rehabilitation of the upper extremity proprioceptive dysfunction in stroke populations and discuss avenues of future research.

H-Man: A planar, H-shape cabled differential robotic manipulandum for experiments on human motor control

In the last decades more robotic manipulanda have been employed to investigate the effect of haptic environments on motor learning and rehabilitation. However, implementing complex haptic renderings can be challenging from technological and control perspectives. We propose a novel robot (H-Man) characterized by a mechanical design based on cabled differential transmission providing advantages over current robotic technology. The H-Man transmission translates to extremely simplified kinematics and homogenous dynamic properties, offering the possibility to generate haptic channels by passively blocking the mechanics, and eliminating stability concerns. We report results of experiments characterizing the performance of the device (haptic bandwidth, Z-width, and perceived impedance). We also present the results of a study investigating the influence of haptic channel compliance on motor learning in healthy individuals, which highlights the effects of channel compliance in enhancing proprioceptive information. The generation of haptic channels to study motor redundancy is not easy for actual robots because of the needs of powerful actuation and complex real-time control implementation. The mechanical design of H-Man affords the possibility to promptly create haptic channels by mechanical stoppers (on one of the motors) without compromising the superior backdriveability and high isotropic manipulability. This paper presents a novel robotic device for motor control studies and robotic rehabilitation. The hardware was designed with specific emphasis on the mechanics that result in a system that is easy to control, homogeneous, and is intrinsically safe for use.

Motor adaptation with passive machines: A first study on the effect of real and virtual stiffness

Motor adaptation to novel force fields is considered as a key mechanism not only for the understanding of skills learning in healthy subjects but also for rehabilitation of neurological subjects. Several studies conducted over the last two decades used active robotic manipulanda to generate force fields capable of perturbing the baseline trajectories of both healthy and impaired subjects. Recent studies showed how motor adaptation to novel force fields can be induced also via virtual environments, whereas the effects of the force are projected onto a virtual hand, while the real hand remains constrained within a channel. This has great potentials of being translated into passive devices, rather than robotic ones, with clear benefits in terms of costs and availability of the devices. However, passive devices and virtual environments have received much less attention at least with regard to motor adaptation. This paper investigates the effects of both the real and virtual stiffness on motor adaptation. In particular, we tested 20 healthy subjects under two different real stiffness conditions (Stiff Channel vs Compliant Channel) and two different virtual conditions (Viscous vs Springy). Our main finding is that compliance of the channel favours a better adaptation featured with less lateral errors and longer retention of the after-effect. We posit that the physical compliance of the channel induces a proprioceptive feedback which is otherwise absent in a stiff condition.

Human-like pointing strategies via non-linear inverse optimization

In movement neuroscience the motor synergy hypothesis has been proposed as the simplifying strategy that the brain adopts when facing redundant tasks. By grouping multiple control variables into synergies, the brain reduces the number of degrees-of-freedom effectively available to solve a certain task. Kinematic, or postural synergies have been identified during the execution of pointing tasks involving either the eye, the head or the wrist and during hand grasping. Postural synergies can be predicted via constrained optimization by hypothesizing the existence of cost functions that the brain would minimize during the execution of redundant tasks. From a computational perspective, in the hypothesis of a correct guess for the cost function, the challenge remains of how to tune the cost parameters so as to predict experimental synergies. In this work a postural model for the wrist-forearm previously proposed in the literature is extended with a non-linear inverse optimization (NIO) approach to tune the discomfort function parameters of the model. An efficient method is proposed to filter and down-sample the experimental data so as to reduce the computational burden of the NIO algorithm. Results show that, after the optimization of the cost parameters, the model can predict with high accuracy six experimental pointing strategies. The proposed approach may in future find applications in human-like motion planning for redundant robots.

Modular and hierarchical brain organization to understand assimilation, accommodation and their relation to autism in reaching tasks: a developmental robotics hypothesis

By assimilation children embody sensorimotor experiences into already built mental structures. Conversely, by accommodation these structures are changed according to the child’s new experience. Despite the intuitive power of these concepts to trace the course of sensorimotor development, they have gradually lost ground in psychology. This likely due to the lack of brain-related views capturing the dynamic mechanisms underlying them. Here we propose that brain modular and hierarchical organization is crucial to understanding assimilation/accommodation. We devise an experiment where a bio-inspired modular and hierarchical mixture-of-experts model guides a simulated robot to learn different reaching tasks by trial-and-error. The model gives a novel interpretation of assimilation/accommodation based on the functional organization of the experts allocated through learning. Assimilation occurs when the model adapts a copy of the expert trained for solving a task, to face another task requiring similar sensorimotor mappings. Experts storing similar sensorimotor mappings belong to the same functional module. Accommodation occurs when the model uses non-trained experts to face tasks requiring different sensorimotor mappings (generating a new functional group of experts). The model also provides a new theoretical framework to investigate assimilation/accommodation impairment, and proposes that such impairment might be related to autism spectrum disorder.

Reinforcement learning algorithms that assimilate and accommodate skills with multiple tasks

Children are capable of acquiring a large repertoire of motor skills and of efficiently adapting them to novel conditions. In a previous work we proposed a hierarchical modular reinforcement learning model (RANK) that can learn multiple motor skills in continuous action and state spaces. The model is based on a development of the mixture-of-expert model that has been suitably developed to work with reinforcement learning. In particular, the model uses a high-level gating network for assigning responsibilities for acting and for learning to a set of low-level expert networks. The model was also developed with the goal of exploiting the Piagetian mechanisms of assimilation and accommodation to support learning of multiple tasks. This paper proposes a new model (TERL - Transfer Expert Reinforcement Learning) that substantially improves RANK. The key difference with respect to the previous model is the decoupling of the mechanisms that generate the responsibility signals of experts for learning and for control. This made possible to satisfy different constraints for functioning and for learning. We test both the TERL and the RANK models with a two-DOFs dynamic arm engaged in solving multiple reaching tasks, and compare the two with a simple, flat reinforcement learning model. The results show that both models are capable of exploiting assimilation and accommodation processes in order to transfer knowledge between similar tasks, and at the same time to avoid catastrophic interference. Furthermore, the TERL model is shown to significantly outperform the RANK model thanks to its faster and more stable specialization of experts.

Preliminary feasibility study of the H-Man planar robot for quantitative motor assessment

Current robotic rehabilitation devices have a high cost-to-benefit ratio, which prevents their large scale adoption by the clinical rehabilitation community. This paper first presents H-Man, a low cost planar robot, as a quantitative assessment and training tool. This is followed by a preliminary study to investigate baseline performance measures for motor assessment during reaching tasks as a step toward replacing conventional ordinal scales with continuous quantitative scales. Thirteen healthy and one participant with upper limb motor impairment participated in the study and performed reaching tasks with their dominant and non-dominant hands in three directions. The results from healthy subjects indicate no significant difference between different directions for both limbs and also between corresponding directions of dominant and non-dominant limbs (p > 0.05, all cases). However, differences in measures can be observed for the impaired subject.

A Reinforcement Learning Architecture that Transfers Knowledge between Skills when Solving Multiple Tasks

When humans learn several skills to solve multiple tasks, they exhibit an extraordinary capacity to transfer knowledge between them. We present here the last enhanced version of a bio-inspired reinforcement-learning (RL) modular architecture able to perform skill-to-skill knowledge transfer and called transfer expert RL (TERL) model. TERL architecture is based on a RL actor–critic model where both actor and critic have a hierarchical structure, inspired by the mixture-of-experts model, formed by a gating network that selects experts specializing in learning the policies or value functions of different tasks. A key feature of TERL is the capacity of its gating networks to accumulate, in parallel, evidence on the capacity of experts to solve the new tasks so as to increase the responsibility for action of the best ones. A second key feature is the use of two different responsibility signals for the experts’ functioning and learning: this allows the training of multiple experts for each task so that some of them can be later recruited to solve new tasks and avoid catastrophic interference. The utility of TERL mechanisms is shown with tests involving two simulated dynamic robot arms engaged in solving reaching tasks, in particular a planar 2-DoF arm, and a 3-D 4-DoF arm.

‘Feel the Painting’: a clinician-friendly approach to programming planar force fields for haptic devices

Haptic force fields are widely used in studies on motor adaptation, motor retention, and motor recovery in both healthy and impaired subjects. In the main paradigm the hand is guided or perturbed along specific paths or channels in order to investigate different aspects underlying the human motor control. Programming such fields for complex haptic environments can be very challenging and is often not feasible for clinicians and therapists. The aim of this paper is to introduce a more intuitive and clinician-friendly programming method capable of transforming a 2D drawing (stored as an image) into a haptic environment or planar force field. By considering the image intensity as a position-dependent potential field, the energy function is approximated through locally weighted projection regression (LWPR). Robot forces are then computed through the gradient of the regressed potential. The proposed method is validated with a two degrees-of-freedom planar manipulandum, the H-Man, and a preliminary shape recognition experiment involving blindfolded healthy subjects.

A novel robot for arm motor therapy with homogeneous mechanical properties

Robotic platforms developed to assist conventional motor therapy and to investigate human motor control have received an increasing interest over the past decades. However, for most of the proposed solutions, bulkiness and expensiveness have limited the use of such devices to specialized clinics that can afford their cost. This paper presents the H-Man, a two degree-of-freedom planar device designed according to three main principles: cost-effectiveness, portability and ease of control. The key component of the device is a planar H-shaped cable differential mechanism which ensures a constant Jacobian and homogeneous perceived inertia over the entire workspace. The paper presents the mechanical design as well as the performance evaluation in terms of perceived impedance.