KWKeenon Werling

Keenon Werling

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Keenon Werling has a total of 35 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    From Skin to Skeleton: Towards Biomechanically Accurate 3D Digital Humans

    by Marilyn Keller, Keenon Werling, Soyong Shin, Scott Delp, Sergi Pujades, C. Karen Liu, Michael J. Black

    Great progress has been made in estimating 3D human pose and shape from images and video by training neural networks to directly regress the parameters of parametric human models like SMPL. However, existing body models have simplified kinematic structures that do not correspond to the true joint locations and articulations in the human skeletal system, limiting their potential use in biomechanics. On the other hand, methods for estimating biomechanically accurate skeletal motion typically rely on complex motion capture systems and expensive optimization methods. What is needed is a parametric 3D human model with a biomechanically accurate skeletal structure that can be easily posed. To that end, we develop SKEL, which re-rigs the SMPL body model with a biomechanics skeleton. To enable this, we need training data of skeletons inside SMPL meshes in diverse poses. We build such a dataset by optimizing biomechanically accurate skeletons inside SMPL meshes from AMASS sequences. We then learn a regressor from SMPL mesh vertices to the optimized joint locations and bone rotations. Finally, we re-parametrize the SMPL mesh with the new kinematic parameters. The resulting SKEL model is animatable like SMPL but with fewer, and biomechanically-realistic, degrees of freedom. We show that SKEL has more biomechanically accurate joint locations than SMPL, and the bones fit inside the body surface better than previous methods. By fitting SKEL to SMPL meshes we are able to "upgrade" existing human pose and shape datasets to include biomechanical parameters. SKEL provides a new tool to enable biomechanics in the wild, while also providing vision and graphics researchers with a better constrained and more realistic model of human articulation. The model, code, and data are available for research at https://skel.is.tue.mpg.de..

  • AddBiomechanics Dataset: Capturing the Physics of Human Motion at Scale

    by Keenon Werling, Janelle Kaneda, Tian Tan, Rishi Agarwal, Six Skov, Tom Van Wouwe, Scott Uhlrich, Nicholas Bianco, Carmichael Ong, Antoine Falisse, et al.

    While reconstructing human poses in 3D from inexpensive sensors has advanced significantly in recent years, quantifying the dynamics of human motion, including the muscle-generated joint torques and external forces, remains a challenge. Prior attempts to estimate physics from reconstructed human poses have been hampered by a lack of datasets with high-quality pose and force data for a variety of movements. We present the AddBiomechanics Dataset 1.0, which includes physically accurate human dynamics of 273 human subjects, over 70 h of motion and force plate data, totaling more than 24 million frames. To construct this dataset, novel analytical methods were required, which are also reported here. We propose a benchmark for estimating human dynamics from motion using this dataset, and present several baseline results. The AddBiomechanics Dataset is publicly available at addbiomechanics.org/download/_data.html.


  • Top Cited Papers
    Open Access

    From Skin to Skeleton: Towards Biomechanically Accurate 3D Digital Humans

    by Marilyn Keller, Keenon Werling, Soyong Shin, Scott Delp, Sergi Pujades, C. Karen Liu, Michael J. Black

    Great progress has been made in estimating 3D human pose and shape from images and video by training neural networks to directly regress the parameters of parametric human models like SMPL. However, existing body models have simplified kinematic structures that do not correspond to the true joint locations and articulations in the human skeletal system, limiting their potential use in biomechanics. On the other hand, methods for estimating biomechanically accurate skeletal motion typically rely on complex motion capture systems and expensive optimization methods. What is needed is a parametric 3D human model with a biomechanically accurate skeletal structure that can be easily posed. To that end, we develop SKEL, which re-rigs the SMPL body model with a biomechanics skeleton. To enable this, we need training data of skeletons inside SMPL meshes in diverse poses. We build such a dataset by optimizing biomechanically accurate skeletons inside SMPL meshes from AMASS sequences. We then learn a regressor from SMPL mesh vertices to the optimized joint locations and bone rotations. Finally, we re-parametrize the SMPL mesh with the new kinematic parameters. The resulting SKEL model is animatable like SMPL but with fewer, and biomechanically-realistic, degrees of freedom. We show that SKEL has more biomechanically accurate joint locations than SMPL, and the bones fit inside the body surface better than previous methods. By fitting SKEL to SMPL meshes we are able to "upgrade" existing human pose and shape datasets to include biomechanical parameters. SKEL provides a new tool to enable biomechanics in the wild, while also providing vision and graphics researchers with a better constrained and more realistic model of human articulation. The model, code, and data are available for research at https://skel.is.tue.mpg.de.


  • Top Cited Papers
    Open Access

    AddBiomechanics: Automating model scaling, inverse kinematics, and inverse dynamics from human motion data through sequential optimization

    by Keenon Werling, Nicholas A. Bianco, Michael Raitor, Jon Stingel, Jennifer L. Hicks, Steven H. Collins, Scott L. Delp, C. Karen Liu

    Creating large-scale public datasets of human motion biomechanics could unlock data-driven breakthroughs in our understanding of human motion, neuromuscular diseases, and assistive devices. However, the manual effort currently required to process motion capture data and quantify the kinematics and dynamics of movement is costly and limits the collection and sharing of large-scale biomechanical datasets. We present a method, called AddBiomechanics, to automate and standardize the quantification of human movement dynamics from motion capture data. We use linear methods followed by a non-convex bilevel optimization to scale the body segments of a musculoskeletal model, register the locations of optical markers placed on an experimental subject to the markers on a musculoskeletal model, and compute body segment kinematics given trajectories of experimental markers during a motion. We then apply a linear method followed by another non-convex optimization to find body segment masses and fine tune kinematics to minimize residual forces given corresponding trajectories of ground reaction forces. The optimization approach requires approximately 3-5 minutes to determine a subject’s skeleton dimensions and motion kinematics, and less than 30 minutes of computation to also determine dynamically consistent skeleton inertia properties and fine-tuned kinematics and kinetics, compared with about one day of manual work for a human expert. We used AddBiomechanics to automatically reconstruct joint angle and torque trajectories from previously published multi-activity datasets, achieving close correspondence to expert-calculated values, marker root-mean-square errors less than 2 cm, and residual force magnitudes smaller than 2% of peak external force. Finally, we confirmed that AddBiomechanics accurately reproduced joint kinematics and kinetics from synthetic walking data with low marker error and residual loads. We have published the algorithm as an open source cloud service at AddBiomechanics.org, which is available at no cost and asks that users agree to share processed and de-identified data with the community. As of this writing, hundreds of researchers have used the prototype tool to process and share about ten thousand motion files from about one thousand experimental subjects. Reducing the barriers to processing and sharing high-quality human motion biomechanics data will enable more people to use state-of-the-art biomechanical analysis, do so at lower cost, and share larger and more accurate datasets.


  • Open Access

    AddBiomechanics: Automating model scaling, inverse kinematics, and inverse dynamics from human motion data through sequential optimization

    by Keenon Werling, Nicholas August Bianco, Michael Raitor, Jon Stingel, Jennifer L. Hicks, Steven H. Collins, Scott L. Delp, C. Karen Liu

    Creating large-scale public datasets of human motion biomechanics could unlock data-driven breakthroughs in our understanding of human motion, neuromuscular diseases, and assistive devices. However, the manual effort currently required to process motion capture data and quantify the kinematics and dynamics of movement is costly and limits the collection and sharing of large-scale biomechanical datasets. We present a method, called AddBiomechanics, to automate and standardize the quantification of human movement dynamics from motion capture data. We use linear methods followed by a non-convex bilevel optimization to scale the body segments of a musculoskeletal model, register the locations of optical markers placed on an experimental subject to the markers on a musculoskeletal model, and compute body segment kinematics given trajectories of experimental markers during a motion. We then apply a linear method followed by another non-convex optimization to find body segment masses and fine tune kinematics to minimize residual forces given corresponding trajectories of ground reaction forces. The optimization approach requires approximately 3-5 minutes to determine a subject’s skeleton dimensions and motion kinematics, and less than 30 minutes of computation to also determine dynamically consistent skeleton inertia properties and fine-tuned kinematics and kinetics, compared with about one day of manual work for a human expert. We used AddBiomechanics to automatically reconstruct joint angle and torque trajectories from previously published multi-activity datasets, achieving close correspondence to expert-calculated values, marker root-mean-square errors less than 2 cm, and residual force magnitudes smaller than 2% of peak external force. Finally, we confirmed that AddBiomechanics accurately reproduced joint kinematics and kinetics from synthetic walking data with low marker error and residual loads. We have published the algorithm as an open source cloud service at AddBiomechanics.org , which is available at no cost and asks that users agree to share processed and de-identified data with the community. As of this writing, hundreds of researchers have used the prototype tool to process and share about ten thousand motion files from about one thousand experimental subjects. Reducing the barriers to processing and sharing high-quality human motion biomechanics data will enable more people to use state-of-the-art biomechanical analysis, do so at lower cost, and share larger and more accurate datasets.


  • Open Access

    Real-time Model Predictive Control and System Identification Using Differentiable Physics Simulation

    by Sirui Chen, Keenon Werling, Albert Wu, C. Karen Liu

    Developing robot controllers in a simulated environment is advantageous but transferring the controllers to the target environment presents challenges, often referred to as the "sim-to-real gap". We present a method for continuous improvement of modeling and control after deploying the robot to a dynamically-changing target environment. We develop a differentiable physics simulation framework that performs online system identification and optimal control simultaneously, using the incoming observations from the target environment in real time. To ensure robust system identification against noisy observations, we devise an algorithm to assess the confidence of our estimated parameters, using numerical analysis of the dynamic equations. To ensure real-time optimal control, we adaptively schedule the optimization window in the future so that the optimized actions can be replenished faster than they are consumed, while staying as up-to-date with new sensor information as possible. The constant re-planning based on a constantly improved model allows the robot to swiftly adapt to the changing environment and utilize real-world data in the most sample-efficient way. Thanks to a fast differentiable physics simulator, the optimization for both system identification and control can be solved efficiently for robots operating in real time. We demonstrate our method on a set of examples in simulation and show that our results are favorable compared to baseline methods.

  • Open Access

    Rapid bilevel optimization to concurrently solve musculoskeletal scaling, marker registration, and inverse kinematic problems for human motion reconstruction

    by Keenon Werling, Michael Raitor, Jon Stingel, Jennifer L. Hicks, Steve Collins, Scott L. Delp, C. Karen Liu

    Creating large-scale public datasets of human motion biomechanics could unlock data-driven breakthroughs in our understanding of human motion, neuromuscular diseases, and assistive devices. However, the manual effort currently required to process motion capture data is costly and limits the collection and sharing of large-scale biomechanical datasets. We present a method to automate and standardize motion capture data processing: bilevel optimization that is able to scale the body segments of a musculoskeletal model, register the locations of optical markers placed on an experimental subject to the markers on a musculoskeletal model, and compute body segment kinematics given trajectories of experimental markers during a motion. The optimization requires less than five minutes of computation to process a subject’s motion capture data, compared with about one day of manual work for a human expert. On a sample of 34 trials of experimental data, the root-mean-square marker reconstruction error (RMSE) was 1.38 cm, approximately 40% lower than the 2.58 cm achieved manually by 3 experts. Optimization solutions reconstructed known joint angle trajectories from four diverse motion trials of synthetic data to an average of 0.79 degrees RMSE. We have published an open source cloud service at AddBiomechanics.org to process experimental motion capture data, which is available at no cost and asks that users agree to share processed and de-identified data with the community. Reducing the barriers to processing and sharing high-quality human motion biomechanics data will enable more people to engage in state-of-the-art biomechanical analysis in their work, do so at lower cost, and share larger and more accurate datasets. Author summary Creating large-scale public datasets of human motion could unlock data-driven breakthroughs in our understanding of neuromuscular diseases, assistive devices, and human motion more broadly. The manual effort currently required to process these motion datasets is costly and limits the collection and sharing of large-scale datasets. Our cloud-based software tool, called AddBiomechanics, uses state-of-the-art optimization techniques to automatically scale the body segments of a musculoskeletal model to match the subject of interest, and then compute body segment kinematics during a motion. The optimization requires less than five minutes of computation to process a subject’s motion capture data, compared with about one day of manual work for a human expert. The accuracy of the approach in quantifying the body segment kinematics is as good or better than the results achieved manually by experts. Reducing the barriers to processing and sharing high-quality human motion biomechanics data will enable more people to engage in state-of-the-art biomechanical analysis, do so at lower cost, and share larger and more accurate datasets.


  • Open Access

    Robust Subgraph Generation Improves Abstract Meaning Representation Parsing

    by Keenon Werling, Gabor Angeli, Christopher Manning

    The Abstract Meaning Representation (AMR) is a representation for open-domain rich semantics, with potential use in fields like event extraction and machine translation. Node generation, typically done using a simple dictionary lookup, is currently an important limiting factor in AMR parsing. We propose a small set of actions that derive AMR subgraphs by transformations on spans of text, which allows for more robust learning of this stage. Our set of construction actions generalize better than the previous approach, and can be learned with a simple classifier. We improve on the previous state-of-the-art result for AMR parsing, boosting end-to-end performance by 3 F$_1$ on both the LDC2013E117 and LDC2014T12 datasets.