Search for: All records

This paper presents a comprehensive study on using deep reinforcement learning (RL) to create dynamic locomotion controllers for bipedal robots. Going beyond focusing on a single locomotion skill, we develop a general control solution that can be used for a range of dynamic bipedal skills, from periodic walking and running to aperiodic jumping and standing. Our RL-based controller incorporates a novel dual-history architecture, utilizing both a long-term and short-term input/output (I/O) history of the robot. This control architecture, when trained through the proposed end-to-end RL approach, consistently outperforms other methods across a diverse range of skills in both simulation and the real world. The study also delves into the adaptivity and robustness introduced by the proposed RL system in developing locomotion controllers. We demonstrate that the proposed architecture can adapt to both time-invariant dynamics shifts and time-variant changes, such as contact events, by effectively using the robot’s I/O history. Additionally, we identify task randomization as another key source of robustness, fostering better task generalization and compliance to disturbances. The resulting control policies can be successfully deployed on Cassie, a torque-controlled human-sized bipedal robot. This work pushes the limits of agility for bipedal robots through extensive real-world experiments. We demonstrate a diverse range of locomotion skills, including: robust standing, versatile walking, fast running with a demonstration of a 400-meter dash, and a diverse set of jumping skills, such as standing long jumps and high jumps. 

Abstract Modeling human motor control and predicting how humans will move in novel environments is a grand scientific challenge. 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 was an official competition at the NeurIPS conference from 2017 to 2019 and attracted over 1300 teams from around the world. Top teams adapted state-of-the-art deep reinforcement learning techniques and produced 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 

Generalization is a central challenge for the deployment of reinforcement learning (RL) systems in the real world. In this paper, we show that the sequential structure of the RL problem necessitates new approaches to generalization beyond the well-studied techniques used in supervised learning. While supervised learning methods can generalize effectively without explicitly accounting for epistemic uncertainty, we describe why appropriate uncertainty handling can actually be essential in RL. We show that generalization to unseen test conditions from a limited number of training conditions induces a kind of implicit partial observability, effectively turning even fully-observed MDPs into POMDPs. Informed by this observation, we recast the problem of generalization in RL as solving the induced partially observed Markov decision process, which we call the epistemic POMDP. We demonstrate the failure modes of algorithms that do not appropriately handle this partial observability, and suggest a simple ensemble-based technique for approximately solving the partially observed problem. Empirically, we demonstrate that our simple algorithm derived from the epistemic POMDP achieves significant gains in generalization over current methods on the Procgen benchmark suite. 

Machine learning systems deployed in the wild are often trained on a source distribution but deployed on a different target distribution. Unlabeled data can be a powerful point of leverage for mitigating these distribution shifts, as it is frequently much more available than labeled data and can often be obtained from distributions beyond the source distribution as well. However, existing distribution shift benchmarks with unlabeled data do not reflect the breadth of scenarios that arise in real-world applications. In this work, we present the WILDS 2.0 update, which extends 8 of the 10 datasets in the WILDS benchmark of distribution shifts to include curated unlabeled data that would be realistically obtainable in deployment. These datasets span a wide range of applications (from histology to wildlife conservation), tasks (classification, regression, and detection), and modalities (photos, satellite images, microscope slides, text, molecular graphs). The update maintains consistency with the original WILDS benchmark by using identical labeled training, validation, and test sets, as well as identical evaluation metrics. We systematically benchmark state-of-the-art methods that use unlabeled data, including domain-invariant, self-training, and self-supervised methods, and show that their success on WILDS is limited. To facilitate method development, we provide an open-source package that automates data loading and contains the model architectures and methods used in this paper.