Deep reinforcement learning for modeling human locomotion control in neuromechanical simulation

Journal of NeuroEngineering and Rehabilitation, Год журнала: 2021, Номер 18(1)

Опубликована: Авг. 16, 2021

Abstract Modeling human motor control and predicting how humans will move in novel environments is a grand scientific challenge. Researchers the fields of biomechanics have proposed evaluated models via neuromechanical simulations, which produce physically correct motions musculoskeletal model. Typically, researchers developed that encode physiologically plausible hypotheses compared resulting simulation behaviors to measurable motion data. While such were able simulate explain many basic locomotion (e.g. walking, running, climbing stairs), modeling higher layer controls processing environment cues, planning long-term strategies, coordinating skills navigate dynamic complex environments) remains Recent advances deep reinforcement learning lay foundation for these processes controlling diverse repertoire movement; however, has been rarely applied model control. In this paper, we review current state along with fundamentals learning, as it applies locomotion. We also present competition accompanying software platform, organized accelerate use simulations. This “Learn Move” was an official at NeurIPS conference from 2017 2019 attracted over 1300 teams around world. Top adapted state-of-the-art techniques produced motions, quick turning walk-to-stand transitions, not demonstrated before simulations without utilizing reference close discussion future opportunities intersection movement our plans extend Learn Move further facilitate interdisciplinary collaboration rehabilitation research

Язык: Английский

---

Comparing optimized exoskeleton assistance of the hip, knee, and ankle in single and multi-joint configurations

Wearable Technologies, Год журнала: 2021, Номер 2

Опубликована: Янв. 1, 2021

Abstract Exoskeletons that assist the hip, knee, and ankle joints have begun to improve human mobility, particularly by reducing metabolic cost of walking. However, direct comparisons optimal assistance these joints, or their combinations, not yet been possible. Assisting multiple may be more beneficial than sum individual effects, because muscles often span less effective, single-joint can indirectly aid other joints. In this study, we used a hip–knee–ankle exoskeleton emulator paired with human-in-the-loop optimization find single-joint, two-joint, whole-leg maximally reduced Hip-only ankle-only walking 26 30% relative in device unassisted, confirming both are good targets for ( N = 3). Knee-only 13%, demonstrating effective knee is possible Two-joint between 33 42%, largest improvements coming from hip-ankle all three 50%, showing at least half energy expended during saved through 4). Changes kinematics muscle activity indicate assisted such improvement was smaller its parts. entire limb maximum effect, but single well-chosen joint efficient when considering additional factors as weight cost.

Язык: Английский

---

General variability leads to specific adaptation toward optimal movement policies

Current Biology, Год журнала: 2022, Номер 32(10), С. 2222 - 2232.e5

Опубликована: Май 1, 2022

Язык: Английский

---

Lower limb biomechanics of fully trained exoskeleton users reveal complex mechanisms behind the reductions in energy cost with human-in-the-loop optimization

Frontiers in Robotics and AI, Год журнала: 2024, Номер 11

Опубликована: Янв. 31, 2024

Exoskeletons that assist in ankle plantarflexion can improve energy economy locomotion. Characterizing the joint-level mechanisms behind these reductions cost lead to a better understanding of how people interact with devices, as well improved device design and training protocols. We examined biomechanical responses exoskeleton assistance users trained lengthened protocol. Kinematics at unassisted joints were generally unchanged by assistance, which has been observed other studies. Peak angle increased led total biological mechanical power despite decreases joint torque whole-body net metabolic cost. Ankle plantarflexor activity also decreased assistance. Muscles act about for large levels this response should be investigated over long-term use prevent overuse injuries.

Язык: Английский

---

Minimum effort simulations of split-belt treadmill walking exploit asymmetry to reduce metabolic energy expenditure

Journal of Neurophysiology, Год журнала: 2023, Номер 129(4), С. 900 - 913

Опубликована: Март 8, 2023

Behavioral observations of split-belt treadmill adaptation have been inconclusive toward its underlying causes. To estimate gait patterns when driven exclusively by one these possible causes, we simulated walking with a musculoskeletal model that minimized summed muscle excitations. Our took significantly longer steps on the fast belt and reduced metabolic rate below tied-belt walking, unlike experimental observations. This suggests asymmetry is energetically optimal, but human involves additional factors.

Язык: Английский

---

Intermittent hypoxia-induced enhancements in corticospinal excitability predict gains in motor learning and metabolic efficiency

Опубликована: Апрель 24, 2024

Acute intermittent hypoxia (AIH) enhances human motor function after incomplete spinal cord injury. Although the underlying mechanisms in humans are unknown, emerging evidence indicates that AIH facilitates corticospinal excitability to upper limb. However, functional relevance of this plasticity remains unexplored, and it is unclear whether similar can be induced for lower limb areas. We recently demonstrated improves learning metabolic efficiency during split-belt walking. Thus, we hypothesized increases these enhancements would predict magnitude corresponding reductions net power. assessed tibialis anterior (TA) using transcranial magnetic stimulation quantified changes spatiotemporal asymmetries power response speed perturbations. show TA excitability, facilitation positively correlates with greater adaptation. Notably, demonstrate a novel association between increased reduced savings. Together, our results suggest AIH-induced gains both associated efficiency. Determining indices improvements performance critical optimizing its therapeutic reach.

Язык: Английский

---

Enhanced motor learning and motor savings after acute intermittent hypoxia are associated with a reduction in metabolic cost

The Journal of Physiology, Год журнала: 2023, Номер 602(21), С. 5879 - 5899

Опубликована: Ноя. 20, 2023

Abstract Breathing mild bouts of low oxygen air (i.e. acute intermittent hypoxia, AIH) has been shown to improve locomotor function in humans after a spinal cord injury. How AIH‐induced gains motor performance are achieved remains unclear. We examined the hypothesis that AIH augments learning and retention during adaptation task. further hypothesized will be associated with reductions net metabolic power, consistent acquisition energetically favourable mechanics. Thirty healthy individuals were randomly allocated into either control group or an group. utilized split‐belt treadmill characterize adaptations unexpected belt speed perturbation equal magnitude initial exposure second exposure. Adaptation was characterized by changes spatiotemporal step asymmetry, anterior–posterior force power. While both groups adapted reducing spatial only significant double support time asymmetry propulsive exposures perturbation. Net power also significantly lower group, from second. These results provide first evidence mediates improvements retention. Further, our suggest continue optimized upon subsequent driven more temporal coordination strategies. Our observation facilitates can leveraged design rehabilitation interventions promote functional recovery. image Key points Brief air, known as hypoxia (AIH), improves injury, but it unclear how achieved. In this study, we tested induces enhancements quantifying interlimb coordination, symmetry cost show savings newly learned strategies compared untreated individuals. demonstrate elicits greater continues learning. findings facilitated

Язык: Английский

---

Exploration-based learning of a stabilizing controller predicts locomotor adaptation

Nature Communications, Год журнала: 2024, Номер 15(1)

Опубликована: Ноя. 3, 2024

Humans adapt their locomotion seamlessly in response to changes the body or environment. It is unclear how such adaptation improves performance measures like energy consumption symmetry while avoiding falling. Here, we model locomotor as interactions between a stabilizing controller that reacts quickly perturbations and reinforcement learner gradually controller's through local exploration memory. This predicts time-varying many settings: walking on split-belt treadmill (i.e. with both feet at different speeds), asymmetric leg weights, using exoskeletons — capturing learning generalization phenomena ten prior experiments two model-guided conducted here. The measure of minimization minor cost for asymmetry captures broad range can act alongside other mechanisms reducing sensory prediction error. Such model-based understanding guide rehabilitation wearable robot control. People learn walk better over time novel situations, new shoes. authors show adaptive behavior relies stabilizer keep walker from falling, explaining this modified improve performance.

Язык: Английский

---

Patterns of asymmetry and energy cost generated from predictive simulations of hemiparetic gait

PLoS Computational Biology, Год журнала: 2022, Номер 18(9), С. e1010466 - e1010466

Опубликована: Сен. 9, 2022

Hemiparesis, defined as unilateral muscle weakness, often occurs in people post-stroke or with cerebral palsy, however it is difficult to understand how this hemiparesis affects movement patterns presents alongside a variety of other neuromuscular impairments. Predictive musculoskeletal modeling an opportunity investigate impairments affect gait performance assuming particular cost function. Here, we use predictive simulation quantify the spatiotemporal asymmetries and changes metabolic that emerge when strength unilaterally reduced reducing symmetry cost. We modified 2-D model by uniformly peak isometric force unilaterally. then solved optimal control simulations walking across range speeds minimizing sum cubed excitations. Lastly, ran additional optimizations test if asymmetry would result increase Our results showed magnitude direction effort-optimal depends on both speed level weakness. Also, was 1.25 m/s for symmetrical 20% weakness models but slower (1.00 m/s) 40% 60% models, suggesting can account portion seen hemiparesis. Modifying function minimize resulted small increases (~4%) Overall, our indicate may be Additionally, effect help explain well-known heterogenous distribution observed clinic. Future work could extend testing effects strategies, therefore build more comprehensive understanding clinical populations.

Язык: Английский

---

Step velocity asymmetry rather than step length asymmetry is updated in split-belt treadmill adaptation

Journal of Biomechanics, Год журнала: 2025, Номер 182, С. 112564 - 112564

Опубликована: Фев. 6, 2025

Язык: Английский

---