Pavlovian Control of Intraspinal Microstimulation to Produce Over-Ground Walking

Abstract

ABSTRACTObjectiveNeural interface technologies are more commonly used in people with neural injury. To achieve a symbiotic relationship between device and user, the control system of the device must augment remaining function and adapt to day-to-day changes. The goal of this study was to develop predictive control strategies to produce alternating, over-ground walking in a cat model of hemisection spinal cord injury (SCI) using intraspinal microstimulation (ISMS).ApproachEight cats were anaesthetized and placed in a sling over a walkway. The residual function of a hemisection SCI was mimicked by manually moving one hind-limb through the walking cycle over the walkway. ISMS targeted motor networks in the lumbosacral enlargement to activate muscles in the other limb using low levels of current (< 130 µA). Four different people took turns to move the “intact” limb. Two control strategies, which used ground reaction force and angular velocity information about the manually moved limb to control the timing of the transitions of the other limb, were compared. The first strategy, reaction-based control, used thresholds on the sensor values to initiate state transitions. The other strategy used a reinforcement learning strategy, Pavlovian control, to learn predictions about the sensor values. Thresholds on the predictions were used to initiate transitions.Main ResultsBoth control strategies were able to produce alternating, over-ground walking. Reaction-based control required manual tuning of the thresholds for each person to produce walking, whereas Pavlovian control did not. We demonstrate that learning occurs quickly during walking. Predictions of the sensor signals were learned quickly, initiating transitions in no more than 4 steps. Pavlovian control was resilient to transitions between people walking the limb, between cat experiments, and recovered from mistakes during walking.SignificanceThis work demonstrates that Pavlovian control can augment remaining function and allow for personalized walking with minimal tuning requirements.