Continuous neural control of a bionic limb restores biomimetic gait after amputation

For centuries, the creation of artificial leg replacements that fully replicate the versatility of natural limbs has been a significant challenge. Biological gait necessitates coordinated volitional and reflexive motor control through complex neural interplay, making neuroprosthetic emulation after amputation difficult. Current bionic legs rely on predefined robotic control architectures (finite state machines, pattern recognition) that model cyclic movements into discrete states based on gait phase and terrain. These systems lack continuous neuromodulation from the user, limiting their adaptability. A neuroprosthetic leg fully driven by the human nervous system, without reliance on an intrinsic gait controller, promises superior functionality, requiring high-bandwidth neuromodulation for adaptive foot positioning, shock absorption, and propulsion across diverse terrains. However, standard-of-care amputation procedures result in the loss of essential locomotor peripheral afferents, complicating consistent and biomimetic gait under continuous neural control. Electrical nerve stimulation has shown promise in upper-extremity neuroprostheses, but locomotion's reliance on reflexive spinal neural circuitry and limited plasticity necessitates a different approach. Previous studies with passive and intrinsically controlled bionic legs providing afferent feedback improved gait but lacked full biomimetic gait under continuous human nervous system control. The inherent adaptability of individuals with intact legs, compensating for locomotor disturbances in seconds, suggests that augmenting residual afferent signaling might enable biomimetic neuroprosthetic gait. This study aimed to augment residual muscle afferents—primary feedback modalities for locomotion—using a modified amputation technique, the agonist-antagonist myoneural interface (AMI), which surgically links residual muscles to recreate natural muscle dynamics and leverage native sensory organs to generate biological afferents. Previous studies showed AMI improved volitional free-space control, but stance phase gait remained intrinsically controlled. This study investigated the relationship between muscle afferent augmentation level and biomimetic gait restoration.

Existing literature extensively documents the limitations of current bionic legs. These devices typically employ finite state machines and pattern recognition algorithms to control gait, leading to limited adaptability and responsiveness. While some studies have explored the use of neural inputs for auxiliary control, these remain confined to specific gait phases and unidirectional movements, providing only partial neural control. The complexity of human gait, involving coordinated afferent and efferent signals between supraspinal and spinal neural circuitry, further complicates the development of effective neuroprostheses. The loss of distal tissue during amputation leads to the loss of crucial peripheral afferents, hindering biomimetic control. While electrical nerve stimulation has been investigated, particularly for upper-extremity neuroprostheses, its efficacy in restoring biomimetic gait remains unclear, especially considering the reliance of locomotion on reflexive spinal neural circuitry. Studies involving passive and intrinsically controlled bionic legs with afferent feedback have demonstrated improvements in gait function, but haven't achieved fully biomimetic gait driven by the human nervous system. This highlights the need for a neuroprosthetic system that augments residual afferents and allows for continuous neuromodulation of the bionic limb.

This study, part of a clinical trial (NCT03913273), used a prospective, non-randomized design comparing seven participants with unilateral below-knee AMI amputations (AMI cohort) and seven matched controls with non-AMI amputations (CTL cohort). Matching criteria included age, time since amputation, height, and weight. The AMI procedure surgically connects residual agonist-antagonist muscle pairs to mimic intact biological muscle dynamics, aiming to leverage native sensory organs to generate biomimetic afferents. A custom-designed autonomous bionic limb, incorporating a powered prosthetic ankle, an electromyography (EMG) sensor unit, and flexible surface electrodes, was used. A neuroprosthetic controller enabled continuous ankle torque neuromodulation (dorsiflexion and plantar flexion) throughout the gait cycle using EMG signals from the tibialis anterior (TA) and lateral gastrocnemius (GAS) muscles. The controller did not use predefined robotic algorithms but relied on measured prosthetic ankle joint state to estimate an upper bound for EMG joint torque, avoiding intrinsic gait control techniques. Muscle afferents were estimated using a computational model based on muscle fascicle strains and EMG signals. A range of ambulatory tests were conducted, including level-ground walking at varying speeds, slope and stair navigation, and obstacle crossing. Gait speed, bionic ankle kinematics and kinetics, lower-extremity kinematic (LEK) symmetry, and range of motion (ROM) were analyzed. Correlation analyses examined the relationship between bionic functionality and residual limb muscle afferents. Principal component analysis (PCA) was used to assess overall bionic functionality.

The AMI procedure augmented agonist-antagonist muscle afferents by 18% of biologically intact values. AMI subjects demonstrated a 41% higher maximum walking speed (1.78 ± 0.04 m/s) than CTL subjects (1.26 ± 0.07 m/s), reaching speeds comparable to those of individuals with intact legs. AMI subjects exhibited biomimetic gait neuromodulation, adjusting push-off and foot clearance ankle angles to mimic natural joint kinematics, and increasing bionic ankle peak power and net work with increasing walking speed. In contrast, CTL subjects showed no significant changes in peak power or net work. AMI subjects demonstrated biomimetic terrain adaptation on slopes and stairs, maintaining level-ground walking mechanics while modulating peak power and net work for incline and decline. CTL subjects failed to demonstrate such adaptation. During obstacle crossing, AMI subjects increased dorsiflexion in bionic swing kinematics and generated propulsive power and net work during the recovery step, enabling faster traversal. CTL subjects showed negligible changes or even dysfunctional motor responses. Correlation analysis revealed a strong positive correlation between muscle afferent augmentation and bionic ankle peak power, net work, LEK symmetry, maximum walking speed, and biomimetic slope and stair adaptation. PCA indicated that muscle afferent augmentation significantly enhanced overall bionic limb functionality and perturbation response. Even a modest augmentation of residual muscle afferents restored biomimetic and versatile locomotion and responses to real-world perturbations.

This study's results demonstrate that augmented residual muscle-tendon afferents enable individuals with transtibial amputation to directly neuromodulate biomimetic locomotion. The capacity for adaptation to varying walking speeds, terrains, and perturbations, without relying on predefined intrinsic control frameworks, represents a significant advancement in bionic leg technology. The key to this improved neural controllability is the restoration of proprioceptive afferents through surgically reconstructed agonist-antagonist muscles. The capacity for biomimetic gait neuromodulation, even with a substantial reduction in the total magnitude of afferents compared to intact limbs, highlights the significant sensorimotor adaptability of the human nervous system. The study's findings offer design principles for future neuroprosthetic interfaces, suggesting that partial afferent restoration may suffice for achieving biomimetic neuroprosthetic function under full neuromodulation.

This study demonstrates the successful restoration of biomimetic gait in leg amputees through a novel neuroprosthetic interface that augments residual muscle afferents and enables continuous neural control. The results highlight the importance of proprioceptive feedback in achieving natural and adaptive locomotion. Future research should focus on long-term clinical benefits, patient experiences (confidence, cognitive load, embodiment), and exploration of bionic capabilities in more complex locomotor tasks (sprinting, jumping, single-leg balancing) and dynamic perturbations (rough terrains, balance perturbations). Further investigation into the integration of cutaneous feedback and other surgical interventions, such as osseointegration, might further enhance bionic limb functionality.

The study's non-randomized design and relatively small sample size limit the generalizability of the findings. The specific surgical technique (AMI) might influence the results, and the findings might not be directly transferable to other amputation levels or surgical approaches. The obstacle course was relatively simple, and more complex real-world scenarios should be explored. Finally, long-term effects and potential complications of the AMI procedure are yet to be fully investigated.