Shai Revzen
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Research Interests
Shai Revzen has a total of 26 co-authors. Here are the top co-authors by number of shared publications:
- Matthew D. Kvalheim
University of Maryland, Baltimore County, MD 21250, USA
10 shared publications - Brian Bittner
University of Michigan
5 shared publications - Dan Zhao
Mechanical Engineering, University of Michigan, Ann Arbor, MI, USA
3 shared publications - Ross L. Hatton
Mechanical, Industrial and Manufacturing Engineering, Oregon State University, Corvallis, OR, USA
3 shared publications - Simon Wilshin
Royal Vet College, London NW1 OTU, UK
3 shared publications - Ziyou Wu
Robotics University of Michigan
3 shared publications
Latest Publications
- Open Access
- 26 August 2026
- Published by IOP Publishing in Bioinspiration & Biomimetics
- Vol. 21 (4) , 046040
Walking on rough terrain with any number of legs
Robotics would gain by replicating the remarkable agility of arthropods in navigating complex environments. In this simulation study, we consider the control of ‘multi-legged’ systems which have 6 or more legs. Current multi-legged control strategies in robots include large black-box machine learning models, central pattern generator (CPG) networks, and open-loop feed-forward control with stability arising from the mechanics. Here we present a multi-legged control architecture for rough terrain using a segmental robot with 3 actuators for every 2 legs, which we validated in simulation for 6 to 16 legs. Segments have identical state machines, and each segment also receives input from the segment in front of it. Our design bridges the gap between Walknet -like event cascade controllers and CPG-based controllers: it tightly couples to the ground when present, but produces fictive locomotion when ground contact is missing. It may be useful as an adaptive, computationally light-weight controller for multi-legged robots, and as baseline capability for scaffolding the learning of machine learning controllers.
- 26 August 2026
- Published by IOP Publishing in Bioinspiration & Biomimetics
- Vol. 21 (4) , 046040
Open AccessWalking on rough terrain with any number of legs
Robotics would gain by replicating the remarkable agility of arthropods in navigating complex environments. In this simulation study, we consider the control of ‘multi-legged’ systems which have 6 or more legs. Current multi-legged control strategies in robots include large black-box machine learning models, central pattern generator (CPG) networks, and open-loop feed-forward control with stability arising from the mechanics. Here we present a multi-legged control architecture for rough terrain using a segmental robot with 3 actuators for every 2 legs, which we validated in simulation for 6 to 16 legs. Segments have identical state machines, and each segment also receives input from the segment in front of it. Our design bridges the gap between Walknet -like event cascade controllers and CPG-based controllers: it tightly couples to the ground when present, but produces fictive locomotion when ground contact is missing. It may be useful as an adaptive, computationally light-weight controller for multi-legged robots, and as baseline capability for scaffolding the learning of machine learning controllers.
- Open Access
Sample Efficient Learning of Body-Environment Interaction of an Under-Actuated System
Geometric mechanics provides valuable insights into how biological and robotic systems use changes in shape to move by mechanically interacting with their environment. In high-friction environments it provides that the entire interaction is captured by the ``motility map''. Here we compare methods for learning the motility map from motion tracking data of a physical robot created specifically to test these methods by having under-actuated degrees of freedom and a hard to model interaction with its substrate. We compared four modeling approaches in terms of their ability to predict body velocity from shape change within the same gait, across gaits, and across speeds. Our results show a trade-off between simpler methods which are superior on small training datasets, and more sophisticated methods, which are superior when more training data is available.
Open AccessSample Efficient Learning of Body-Environment Interaction of an Under-Actuated System
Geometric mechanics provides valuable insights into how biological and robotic systems use changes in shape to move by mechanically interacting with their environment. In high-friction environments it provides that the entire interaction is captured by the ``motility map''. Here we compare methods for learning the motility map from motion tracking data of a physical robot created specifically to test these methods by having under-actuated degrees of freedom and a hard to model interaction with its substrate. We compared four modeling approaches in terms of their ability to predict body velocity from shape change within the same gait, across gaits, and across speeds. Our results show a trade-off between simpler methods which are superior on small training datasets, and more sophisticated methods, which are superior when more training data is available.
- 18 November 2025
- Published by MIT Press in Neural Computation
- Vol. 37 (12) , 2158-2204
Estimating Phase From Observed Trajectories Using the Temporal 1-Form
Oscillators are ubiquitous in nature and are usually associated with the existence of an asymptotic phase that governs the long-term dynamics of the oscillator. We show that the asymptotic phase can be estimated using a carefully chosen series expansion that directly computes the phase response curve (PRC) and provides an algorithm for estimating the coefficients of this series. Unlike previously available data-driven phase estimation methods, our algorithm can use observations that are much shorter than a cycle; has proven convergence rate bounds as a function of the properties of measurement noise and system noise; will recover phase within any forward invariant region for which sufficient data are available; recovers the PRCs that govern weak oscillator coupling; and recovers isochron curvature and recovers nonlinear features of isochron geometry. Our method may find application wherever models of oscillator dynamics need to be constructed from measured or simulated time-series.
- 18 November 2025
- Published by MIT Press in Neural Computation
- Vol. 37 (12) , 2158-2204
Estimating Phase From Observed Trajectories Using the Temporal 1-Form
Oscillators are ubiquitous in nature and are usually associated with the existence of an asymptotic phase that governs the long-term dynamics of the oscillator. We show that the asymptotic phase can be estimated using a carefully chosen series expansion that directly computes the phase response curve (PRC) and provides an algorithm for estimating the coefficients of this series. Unlike previously available data-driven phase estimation methods, our algorithm can use observations that are much shorter than a cycle; has proven convergence rate bounds as a function of the properties of measurement noise and system noise; will recover phase within any forward invariant region for which sufficient data are available; recovers the PRCs that govern weak oscillator coupling; and recovers isochron curvature and recovers nonlinear features of isochron geometry. Our method may find application wherever models of oscillator dynamics need to be constructed from measured or simulated time-series.
- 30 October 2024
- Published by Springer Nature in Biological Cybernetics
- Vol. 118 (5-6) , 311-330
Phase response curves and the role of coordinates
The “infinitesimal phase response curve” (PRC) is a common tool used to analyze phase resetting in the natural sciences in general and neuroscience in particular. We make the observation that the PRC with respect to a coordinate v actually depends on the choice of other coordinates. As a consequence, a complete delay embedding reconstruction of the dynamics using v which would allow phase to be computed still does not allow the v PRC to be computed. We give a coordinate-free definition of the PRC making this observation obvious. This leads to an experimental protocol: first collect an appropriate ensemble of measurements by intermittently controlling neuron voltage. Then, for any suitable current carrier dynamic postulated, we show how the ensemble can be used to compute the voltage PRC with that current carrier. The approach extends to many oscillators measured and controlled through a subset of their coordinates.
- 30 October 2024
- Published by Springer Nature in Biological Cybernetics
- Vol. 118 (5-6) , 311-330
Phase response curves and the role of coordinates
The “infinitesimal phase response curve” (PRC) is a common tool used to analyze phase resetting in the natural sciences in general and neuroscience in particular. We make the observation that the PRC with respect to a coordinate v actually depends on the choice of other coordinates. As a consequence, a complete delay embedding reconstruction of the dynamics using v which would allow phase to be computed still does not allow the v PRC to be computed. We give a coordinate-free definition of the PRC making this observation obvious. This leads to an experimental protocol: first collect an appropriate ensemble of measurements by intermittently controlling neuron voltage. Then, for any suitable current carrier dynamic postulated, we show how the ensemble can be used to compute the voltage PRC with that current carrier. The approach extends to many oscillators measured and controlled through a subset of their coordinates.
- 30 September 2024
- Published by ASME International in Journal of Dynamic Systems, Measurement, and Control
- Vol. 147 (3) , 1-17
In-Situ Calibration of Six-Axis Force/Torque Transducers on a Six-Legged Robot
Ground contact modeling for multilegged locomotion is challenging due to the possibility of multiple slipping legs. To understand the interplay of contact forces among multiple legs, we integrated a robot with six high-precision 6 degree-of-freedom (DoF) force-torque sensors, and measured the wrenches (forces and torques) produced in practice. Here, we present an in situ calibration procedure for simultaneously measuring all foot contact wrenches of a hexapod using 6-DoF load cells installed at the hips. We characterized transducer offset, leg gravity offset, and the wrench transformation error in our calibration model. Our calibration reduced the root-mean-square-error (RSME) by 63% for forces and 90% for torques in the residuals of the robot standing in different poses, compared with naive constant offset removal.
- 30 September 2024
- Published by ASME International in Journal of Dynamic Systems, Measurement, and Control
- Vol. 147 (3) , 1-17
In-Situ Calibration of Six-Axis Force/Torque Transducers on a Six-Legged Robot
Ground contact modeling for multilegged locomotion is challenging due to the possibility of multiple slipping legs. To understand the interplay of contact forces among multiple legs, we integrated a robot with six high-precision 6 degree-of-freedom (DoF) force-torque sensors, and measured the wrenches (forces and torques) produced in practice. Here, we present an in situ calibration procedure for simultaneously measuring all foot contact wrenches of a hexapod using 6-DoF load cells installed at the hips. We characterized transducer offset, leg gravity offset, and the wrench transformation error in our calibration model. Our calibration reduced the root-mean-square-error (RSME) by 63% for forces and 90% for torques in the residuals of the robot standing in different poses, compared with naive constant offset removal.
- Open Access
- 6 September 2022
- Published by National Academy of Sciences in Proceedings of the National Academy of Sciences
- Vol. 119 (37)
Walking is like slithering: A unifying, data-driven view of locomotion
Legged movement is ubiquitous in nature and of increasing interest for robotics. Most legged animals routinely encounter foot slipping, yet detailed modeling of multiple contacts with slipping exceeds current simulation capacity. Here we present a principle that unifies multilegged walking (including that involving slipping) with slithering and Stokesian (low Reynolds number) swimming. We generated data-driven principally kinematic models of locomotion for walking in low-slip animals (Argentine ant, 4.7% slip ratio of slipping to total motion) and for high-slip robotic systems (BigANT hexapod, slip ratio 12 to 22%; Multipod robots ranging from 6 to 12 legs, slip ratio 40 to 100%). We found that principally kinematic models could explain much of the variability in body velocity and turning rate using body shape and could predict walking behaviors outside the training data. Most remarkably, walking was principally kinematic irrespective of leg number, foot slipping, and turning rate. We find that grounded walking, with or without slipping, is governed by principally kinematic equations of motion, functionally similar to frictional swimming and slithering. Geometric mechanics thus leads to a unified model for swimming, slithering, and walking. Such commonality may shed light on the evolutionary origins of animal locomotion control and offer new approaches for robotic locomotion and motion planning.
- 6 September 2022
- Published by National Academy of Sciences in Proceedings of the National Academy of Sciences
- Vol. 119 (37)
Open AccessWalking is like slithering: A unifying, data-driven view of locomotion
Legged movement is ubiquitous in nature and of increasing interest for robotics. Most legged animals routinely encounter foot slipping, yet detailed modeling of multiple contacts with slipping exceeds current simulation capacity. Here we present a principle that unifies multilegged walking (including that involving slipping) with slithering and Stokesian (low Reynolds number) swimming. We generated data-driven principally kinematic models of locomotion for walking in low-slip animals (Argentine ant, 4.7% slip ratio of slipping to total motion) and for high-slip robotic systems (BigANT hexapod, slip ratio 12 to 22%; Multipod robots ranging from 6 to 12 legs, slip ratio 40 to 100%). We found that principally kinematic models could explain much of the variability in body velocity and turning rate using body shape and could predict walking behaviors outside the training data. Most remarkably, walking was principally kinematic irrespective of leg number, foot slipping, and turning rate. We find that grounded walking, with or without slipping, is governed by principally kinematic equations of motion, functionally similar to frictional swimming and slithering. Geometric mechanics thus leads to a unified model for swimming, slithering, and walking. Such commonality may shed light on the evolutionary origins of animal locomotion control and offer new approaches for robotic locomotion and motion planning.
- 3 June 2022
- Published by ASME International in Journal of Computational and Nonlinear Dynamics
- Vol. 17 (9)
Representing and Computing the B-Derivative of the Piecewise-Differentiable Flow of a Class of Nonsmooth Vector Fields
This paper concerns first-order approximation of the piecewise-differentiable flow generated by a class of nonsmooth vector fields. Specifically, we represent and compute the Bouligand (or B-)derivative of the piecewise-differentiable flow generated by a vector field with event-selected discontinuities. Our results are remarkably efficient: although there are factorially many “pieces” of the derivative, we provide an algorithm that evaluates its action on a tangent vector using polynomial time and space, and verify the algorithm's correctness by deriving a representation for the B-derivative that requires “only” exponential time and space to construct. We apply our methods in two classes of illustrative examples: piecewise-constant vector fields and mechanical systems subject to unilateral constraints.
- 3 June 2022
- Published by ASME International in Journal of Computational and Nonlinear Dynamics
- Vol. 17 (9)
Representing and Computing the B-Derivative of the Piecewise-Differentiable Flow of a Class of Nonsmooth Vector Fields
This paper concerns first-order approximation of the piecewise-differentiable flow generated by a class of nonsmooth vector fields. Specifically, we represent and compute the Bouligand (or B-)derivative of the piecewise-differentiable flow generated by a vector field with event-selected discontinuities. Our results are remarkably efficient: although there are factorially many “pieces” of the derivative, we provide an algorithm that evaluates its action on a tangent vector using polynomial time and space, and verify the algorithm's correctness by deriving a representation for the B-derivative that requires “only” exponential time and space to construct. We apply our methods in two classes of illustrative examples: piecewise-constant vector fields and mechanical systems subject to unilateral constraints.
- Open Access
Estimating Phase from Observed Trajectories Using the Temporal 1-Form
Oscillators are ubiquitous in nature, and usually associated with the existence of an asymptotic phase that governs the long-term dynamics of the oscillator. % We show that asymptotic phase can be estimated using a carefully chosen series expansion which directly computes the phase response curve and provide an algorithm for estimating the co-efficients of this series. Unlike all previously available data driven phase estimation methods, our algorithm can: (i) use observations that are much shorter than a cycle; (ii) recover phase within any forward invariant region for which sufficient data are available; (iii) recover the phase response curves (PRC-s) that govern weak oscillator coupling; (iv) show isochron curvature, and recover nonlinear features of isochron geometry. Our method may find application wherever models of oscillator dynamics need to be constructed from measured or simulated time-series.
Open AccessEstimating Phase from Observed Trajectories Using the Temporal 1-Form
Oscillators are ubiquitous in nature, and usually associated with the existence of an asymptotic phase that governs the long-term dynamics of the oscillator. % We show that asymptotic phase can be estimated using a carefully chosen series expansion which directly computes the phase response curve and provide an algorithm for estimating the co-efficients of this series. Unlike all previously available data driven phase estimation methods, our algorithm can: (i) use observations that are much shorter than a cycle; (ii) recover phase within any forward invariant region for which sufficient data are available; (iii) recover the phase response curves (PRC-s) that govern weak oscillator coupling; (iv) show isochron curvature, and recover nonlinear features of isochron geometry. Our method may find application wherever models of oscillator dynamics need to be constructed from measured or simulated time-series.
- 24 January 2022
- Published by IOP Publishing in Bioinspiration & Biomimetics
- Vol. 17 (2) , 026004
Data-driven geometric system identification for shape-underactuated dissipative systems
Modeling system dynamics becomes challenging when the properties of individual system components cannot be directly measured, and often requires identification of properties from observed motion. In this paper, we show that systems whose movement is highly dissipative have features which provide an opportunity to more easily identify models and more quickly optimize motions than would be possible with general techniques. Geometric mechanics provides means for reduction of the dynamics by environmental homogeneity, while the dissipative nature minimizes the role of second order (inertial) features in the dynamics. Here we extend the tools of geometric system identification to ‘shape-underactuated dissipative systems (SUDS)’—systems whose motions are more dissipative than inertial, but whose actuation is restricted to a subset of the body shape coordinates. Many animal motions are SUDS, including micro-swimmers such as nematodes and flagellated bacteria, and granular locomotors such as snakes and lizards. Many soft robots are also SUDS, particularly robots that incorporate highly damped series elastic actuators to reduce the rigidity of their interactions with their environments during locomotion and manipulation. We motivate the use of SUDS models, and validate their ability to predict motion of a variety of simulated viscous swimming platforms. For a large class of SUDS, we show how the shape velocity actuation inputs can be directly converted into torque inputs, suggesting that systems with soft pneumatic or dielectric elastomer actuators can be modeled with the tools presented. Based on fundamental assumptions in the physics, we show how our model complexity scales linearly with the number of passive shape coordinates. This scaling offers a large reduction on the number of trials needed to identify the system model from experimental data, and may reduce overfitting. The sample efficiency of our method suggests its use in modeling, control, and optimization in robotics, and as a tool for the study of organismal motion in friction dominated regimes.
- 24 January 2022
- Published by IOP Publishing in Bioinspiration & Biomimetics
- Vol. 17 (2) , 026004
Data-driven geometric system identification for shape-underactuated dissipative systems
Modeling system dynamics becomes challenging when the properties of individual system components cannot be directly measured, and often requires identification of properties from observed motion. In this paper, we show that systems whose movement is highly dissipative have features which provide an opportunity to more easily identify models and more quickly optimize motions than would be possible with general techniques. Geometric mechanics provides means for reduction of the dynamics by environmental homogeneity, while the dissipative nature minimizes the role of second order (inertial) features in the dynamics. Here we extend the tools of geometric system identification to ‘shape-underactuated dissipative systems (SUDS)’—systems whose motions are more dissipative than inertial, but whose actuation is restricted to a subset of the body shape coordinates. Many animal motions are SUDS, including micro-swimmers such as nematodes and flagellated bacteria, and granular locomotors such as snakes and lizards. Many soft robots are also SUDS, particularly robots that incorporate highly damped series elastic actuators to reduce the rigidity of their interactions with their environments during locomotion and manipulation. We motivate the use of SUDS models, and validate their ability to predict motion of a variety of simulated viscous swimming platforms. For a large class of SUDS, we show how the shape velocity actuation inputs can be directly converted into torque inputs, suggesting that systems with soft pneumatic or dielectric elastomer actuators can be modeled with the tools presented. Based on fundamental assumptions in the physics, we show how our model complexity scales linearly with the number of passive shape coordinates. This scaling offers a large reduction on the number of trials needed to identify the system model from experimental data, and may reduce overfitting. The sample efficiency of our method suggests its use in modeling, control, and optimization in robotics, and as a tool for the study of organismal motion in friction dominated regimes.
- Top Cited PapersOpen Access
- 22 December 2021
- Published by The Royal Society in Journal of The Royal Society Interface
- Vol. 18 (185) , 20210686
Challenges in dynamic mode decomposition
Dynamic mode decomposition (DMD) is a powerful tool for extracting spatial and temporal patterns from multi-dimensional time series, and it has been used successfully in a wide range of fields, including fluid mechanics, robotics and neuroscience. Two of the main challenges remaining in DMD research are noise sensitivity and issues related to Krylov space closure when modelling nonlinear systems. Here, we investigate the combination of noise and nonlinearity in a controlled setting, by studying a class of systems with linear latent dynamics which are observed via multinomial observables. Our numerical models include system and measurement noise. We explore the influences of dataset metrics, the spectrum of the latent dynamics, the normality of the system matrix and the geometry of the dynamics. Our results show that even for these very mildly nonlinear conditions, DMD methods often fail to recover the spectrum and can have poor predictive ability. Our work is motivated by our experience modelling multilegged robot data, where we have encountered great difficulty in reconstructing time series for oscillatory systems with intermediate transients, which decay only slightly faster than a period.
- 22 December 2021
- Published by The Royal Society in Journal of The Royal Society Interface
- Vol. 18 (185) , 20210686
Top Cited PapersOpen AccessChallenges in dynamic mode decomposition
Dynamic mode decomposition (DMD) is a powerful tool for extracting spatial and temporal patterns from multi-dimensional time series, and it has been used successfully in a wide range of fields, including fluid mechanics, robotics and neuroscience. Two of the main challenges remaining in DMD research are noise sensitivity and issues related to Krylov space closure when modelling nonlinear systems. Here, we investigate the combination of noise and nonlinearity in a controlled setting, by studying a class of systems with linear latent dynamics which are observed via multinomial observables. Our numerical models include system and measurement noise. We explore the influences of dataset metrics, the spectrum of the latent dynamics, the normality of the system matrix and the geometry of the dynamics. Our results show that even for these very mildly nonlinear conditions, DMD methods often fail to recover the spectrum and can have poor predictive ability. Our work is motivated by our experience modelling multilegged robot data, where we have encountered great difficulty in reconstructing time series for oscillatory systems with intermediate transients, which decay only slightly faster than a period.