Deep reinforcement learning for modeling human locomotion control in neuromechanical simulation

Abstract

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

Precision strength training: Data-driven artificial intelligence approach to strength and conditioning

“…We disruptive AI-driven companies taking similar marketing-heavy paths [218,283,545,730]. [673,707,758,890], but with collaborative interest from sports scientists, the models developed e.g. with OpenSIM-RL [404] could be easily retrained for athlete populations.…”

“…In self-supervised learning approaches that are shown to learn good generic features [932], to be used for netuning with smaller domain-specic datasets (e.g. KIMORE [111], MoVi [267], Learn to Move [758]) for your desired task.…”

Perspective on musculoskeletal modelling and predictive simulations of human movement to assess the neuromechanics of gait

“…Alternatively, reinforcement learning has been used to solve for control policies. Since 2017, the NeurIPS 'Learn to move' competition series has accelerated the adoption of reinforcement learning techniques to simulate human locomotion based on neuromusculoskeletal models (for detailed reviews, see [27,28]). Overall, approaches that solve for gait control policies were based on simpler neuro-musculoskeletal models than trajectory optimization methods, probably because of computational efficiency.…”

ASE: Large-Scale Reusable Adversarial Skill Embeddings for Physically Simulated Characters

“…However, these heuristics are generally not broadly applicable for all skills, and different skills often require different carefully curated sets of heuristics in order to produce life-like behaviors. Incorporating more biologically accurate simulation models can also improve motion quality [Geijtenbeek et al 2013;Jiang et al 2019;Wang et al 2012], but may nonetheless produce unnatural behaviors without the appropriate objective functions [Song et al 2020].…”