Workshop: Human-robot synchronization for assistive technologies - July 17, 2017 

I will summarise human and non-human primate data that highlight the constraints that sensory delays impose on visually guided movements. This leads to theories of predictive control, and the suggestion that forward models underpin much of our behaviour, including coordination and synchronisation. I will end with a summary of evidence that supports our hypothesis that the cerebellum operates as a forward model, anticipating the sensory consequences of actions.

Severe Spinal Cord Injury (SCI) alters the communication between supra-spinal centers and the sensorimotor networks coordinating limb movements, which are usually located below the injury. Epidural electrical stimulation of lumbar segments has shown the ability to enable descending motor control of the lower limbs in rodents and humans with severe paralysis. Using computational models and in vivo experiments in rodents, we found that EES facilitates motor control through the recruitment of muscle spindle feedback circuits. This enables the synchronization and integration of sensory feedbacks and residual motor commands to produce synergistic and coordinated movement. This framework supported the design of brain controlled stimulation strategies that restored locomotion in primates after spinal cord injury, holding promises for applications in humans.

For any motor act, the central nervous system (CNS) must assemble a coordinated pattern of muscle activities both to achieve the intended motion, and to stabilize the joints whose motions are undesired. The computations behind this assembly are extraordinarily complicated, in that the hundreds of muscles spanning the hundreds of joints constitute a large search space of motor commands even for an apparently simple movement. At least for motor activities of daily living, the CNS may circumvent this complexity by generating muscle commands through the combination of a manageable number of pre-specified units, each of whose activation would simultaneously recruit a specific group of muscles. These units, variously called motor primitives, neuromotor modules, or muscle synergies, are in essence a set of neurophysiological mechanisms for synchronizing motoneuronal activities for purposeful motor behaviors. Recent rodent optogenetic experiments of ourselves and others have argued that muscle synergies are encoded by spinal excitatory interneurons that are modulated by both proprioceptive and corticospinal drives. A similar organization may also exist in humans, as indicated by our recent data obtained from human subjects during lower-limb adaptation tasks. Kinematic and EMG recordings from stroke survivors further suggest that specific muscle synergies may serve as stroke recovery biomarkers, and thus could be natural targets of intervention for stroke-rehab assistive technologies.

The development of personalized neurorehabilitation and augmentation technologies requires the profound understanding of the neuro-mechanical processes underlying an individual's motor function, impairment, and recovery. A major challenge is the difficulty of accessing the in vivo neural activity underlying human movement concurrently with the resulting mechanical forces elicited at the musculoskeletal level. Key factors for addressing this challenge are the development of techniques for interfacing with the human nervous system and for the accurate decoding of the resulting motor function under neurophysiological control. In this presentation I will discuss clinically viable methods for accessing the neural information underlying an individual's movement from electrophysiological recordings and the development of subject-specific neuromusculoskeletal modeling formulations that can translate neural inputs into the resulting mechanical output. I will then outline how this paradigm enables establishing effective solutions for replacing or restoring movement in impaired individuals and for developing bio-inspired controllers for wearable technologies.

Cerebral Palsy (CP) could be defined as a disorder that appears in infancy and permanently affect posture and body movement but does not worsen over time. CP is often associated with sensory deficits, cognitive impairments, communication and motor disabilities, behaviour issues, seizure disorder, pain and secondary musculoskeletal problems. Traditionally, robotic strategies have been focused on the Pe- ripheral Nervous System (PNS) supporting patients to perform repetitive movements (a ‘‘Bottom-Up approach’’). However, CP primarily affects brain structures, and thus suggests that both PNS and Central Nervous System (CNS) should be integrated into a physical and cognitive rehabilitation therapy. Current studies manifest that such integration of the CNS into the human–robot loop maximizes the therapeutic effects, especially in children. During this talk I will present and discuss a robot-based training program for gait rehabilitation of pediatric population with Cerebral Palsy. The robotic-based therapies were implement in the CPWalker device and recreates a situation as similar as possible to a real gait scenario, encouraging the patients to control different movements associated with gait: not only individual movements of lower limb joints but also the synergy between them while maintaining a proper posture of upper body. We hypothesize that this interaction between the human and the machine, performed following an appropriate progression of the variables, boosted the rehabilitation of our patients.

Adaptive oscillators are mathematical systems that can be used to extract the main features (like frequency and amplitude) of (quasi-)periodic inputs, while learning these features in dynamic state variables. In this talk, we will review our recent efforts in using adaptive oscillators with humans in the loop. In particular, the dynamic-system-like properties of adaptive oscillators were exploited for doing state observation, state prediction, state differentiation, inverse dynamical modelling, bandpass filtering, and bandstop filtering; each time whith the objective to enhance specific movement features of (disabled) humans.

Human gait is an intrinsically rhythmic movement. Whenever external support is necessary to enable walking, particularly for users with neurological impairments or amputations, this may alter or disrupt natural gait dynamics. In order to preserve or restore such dynamics, hardware and control choices of robotic devices to support walking can either be explicitly tuned to incorporate the desired rhythmicity, or they can be designed in such a way that they favor the emergence of rhythmicity, without any explicit encoding thereof. This talk will focus mainly on the latter strategy and provide examples of several real-world implementations. An outlook will be given on transfer to the inverse problem: Human instructors teaching bipedal robots to walk via rhythmic physical interaction.

In this talk, I will discuss a recent study in which we compared the therapeutic effects of high and low levels of robotic assistance during finger training. Participants (n = 30) with a chronic stroke and moderate hemiparesis actively moved their index and middle fingers to targets to play a musical game similar to GuitarHero three hours/week for three weeks. The FINGER exoskeleton robot provided assistance synchronized to the music; half of the participants were randomized to receive high assistance (causing 82% success at hitting targets), while the other half received low assistance (55% success). High levels of synchronized assistance boosted motivation, as well as secondary motor outcomes (Fugl-Meyer and Lateral Pinch Strength) – particularly for individuals with more severe finger motor deficits. Importantly, individuals with impaired finger proprioception at baseline benefited substantially less from the training. These results show that synchronized robotic assistance can promote key psychological outcomes known to modulate motor learning and retention. Further, the therapeutic effectiveness of synchronized robotic assistance appears to derive at least in part from proprioceptive stimulation, consistent with a Hebbian plasticity model.