Title: Guiding with Touch: Haptic Feedback for Enhancing Human Motor Performance

Abstract: Recent advances in virtual reality simulation and robotics have changed the way that motor skills are trained, yet the feedback that most trainees receive when working in these environments is still delayed, subjective, and qualitative, which does not provide the maximum support for rapid acquisition of motor skill. In this talk, I will describe our research on identifying objective and quantitative metrics that capture motor skills in a few different tasks, and I will present our methods for employing real-time haptic feedback and guidance. We have shown that different implementations of haptic guidance provided via kinesthetic feedback can have either negative, neutral, or positive effects on motor performance and skill acquisition. More recently, we have shifted to provide cutaneous haptic feedback for guidance or performance feedback. We have shown that low-level properties of movements (e.g., smoothness) made in the performance of several motor tasks—including surgery in both virtual and robotic environments—are highly correlated with high-level performance outcomes. We encode these movement properties as simple haptic cues that are conveyed to the trainee in real-time, and evaluate the effect on skill acquisition. This talk will highlight our progress over the past decade in implementing real-time haptic feedback via wearable devices to train complex motor skills.

Title: Human Control of Dynamically Complex Objects

Abstract: Despite having a slow and noisy neuromotor system, humans manage to dexterously interact with complex (nonlinear, underactuated, non-rigid) objects on a daily basis, vastly outperforming robotic manipulators. This feat of human sensorimotor control is not yet fully understood, with much of the current research in human motor control being focused on highly simplified movements. Similar to human motor control, research in primate neurophysiology has mostly focused on simple feedforward tasks that do not involve interaction with the environment. The first part of this talk will present a series of experimental studies inspired by how humans manipulate a complex object, such as transporting a cup of coffee. Different experimental manipulations of this cup-of-coffee task reveal the different strategies that humans use to computationally simplify this control problem. Using tools from dynamical systems theory, control theory, and information theory we show that humans seek control solutions that are 1) stable, 2) predictable, and 3) afford simple internal models. The neuroscience portion of the talk will conclude by presenting a recent primate study where we similarly start to investigate complex, feedback-driven motor behaviors in monkeys. To close the loop between neuroscience and robotics, the second part of the talk will be dedicated to introducing a neuroscience-inspired robot learning method. This new approach uses insights from human observational learning to teach robots manipulation skills by having them observe multiple human demonstrators with varying levels of expertise.

Title: Information Measures in Assistive Autonomy 

Abstract: In physical human-robot interaction, information is communicated through mechanical variables such as configurations, velocities, and forces, and these signals are used to determine assistive mechanical actions. Many tasks are characterized by a state trajectory, and measures of motion quality are errors expressed in that state; for instance, it is common to assess the error when a subject is reaching for a target. The goal of an assistive device could then be to minimize the trajectory error, a traditional engineering goal that is potentially ill-suited to the needs of a person. Indeed, many tasks associated with daily life are not well described by trajectory error profiles. Instead, other descriptions of motion, such as dynamic stability and spatial coverage, may be more important. This talk will discuss the use of information measures and dynamic response for assessing motion and designing assistive interventions. Experimental results include evaluation of measures that assess assistance, motor learning, and motor impairment due to brain injury. I will also discuss control strategies for closing the loop during assistance.

Title: Serial Motor Decisions in a Manual Balancing Task

Abstract: This talk presents investigations of three balancing tasks – bipedal stance, manually controlled self-balancing, and manual object balancing – comparative analysis of which permits partitioning peripheral reflexive and reflex-like mechanisms from higher level mechanisms.  Evidence will be presented that these tasks rely, to different degrees, on both alignment with respect to an external reference frame as well as reference-free intrinsic dynamic stabilization.  We have observed that the self- and object-balancing tasks are spontaneously executed with commands dominated by discrete versus proportional control, and can be performed when proportional control is experimentally prohibited.  A novel discovery is that up to 20% of discrete commands in the self- and object-balancing tasks are frankly destabilizing – tending to accelerate falls – when the task conditions are novel or difficult.  The mixture of discrete corrections and diametrically opposed destabilizations suggests a decision-like process.  The evolution of this decision-like process becomes critical in the phase of the self- and object-balancing where the time to a fall is imminent.  The evolution of this failure-sensitive, decision-like process differs between age groups, between novices and experts, and between nominal versus sensory deprived conditions.  Predicting forthcoming falls in the self- and object-balancing tasks is best in AI algorithms which embody the discrete, decision-like mode which humans exhibit.  Sensory augmentation of balancing performance improves the likelihood of time-critical corrective versus destabilizing actions.  Initial investigations of lessons from self- and object-balancing for bipedal stance will be described.

Title: Learning Locomotion Control at the Spinal Cord Level

Abstract: The importance of the spinal cord to locomotion control is well established. However, it is less well understood how much the spinal circuitry is involved in learning and adaptation during locomotion. We will explore this question with the help of computational models and neurophysiological experiments.  In the first part of this talk, we will use neuromuscular models of human gait to suggest that the spinal control circuitry may be learned through a transfer of control from the brain to the spinal cord. In these models, the transfer is realized by a continual heterosynaptic modulation of muscle reflex gains. In the second part, we will address if such a continual modulation of human muscle reflexes can be observed at all. To this end, we measured H-reflex changes of major leg muscles during split belt locomotion and found that when humans are exposed to novel mechanical environments, the H-reflexes continually adapt in synchrony with the observed changes in gait asymmetry. We further provide evidence that the reflex adaptations are the cause rather than the effect of these gait changes. The overall picture emerging from this research is that the spinal cord may be much more directly involved in locomotion learning and adaptation than normally assumed.

Title: Non-invasive Mapping of Human Cortical Motor Topography: Technical Innovation, Evidence of Modularity, and Reorganization Following Stroke

Abstract: Understanding corticospinal organization of muscle activation is a long-standing goal of motor neuroscience and has profound implications for interventions aimed at enhancing motor performance in health and disease. In this talk I will overview our work using transcranial magnetic stimulation (TMS) to explore motor modularity in healthy individuals and to track reorganization of motor topography in the acute period following stroke. I will present key evidence that modularity of muscle responses to TMS resembles modularity observed during voluntary activation. I will also discuss our recent work to develop a novel experimental and computational framework for modeling multi-muscle responses to TMS of the human motor cortex using machine learning approaches.