2024
Zaichyk, H.; Biess, A.; Kontorovich, A.; Makarychev, Y.
Efficient Kirszbraun Extension with Applications to Regression Journal Article
In: Mathematical Programming (MAPR), 2024.
@article{/b,
title = {Efficient Kirszbraun Extension with Applications to Regression},
author = {H. Zaichyk and A. Biess and A. Kontorovich and Y. Makarychev},
doi = {https://doi.org/10.1007/s10107-024-02056-5},
year = {2024},
date = {2024-01-20},
journal = {Mathematical Programming (MAPR)},
abstract = {We introduce a framework for performing regression between two Hilbert spaces. This is done based on Kirszbraun's extension theorem, to the best of our knowledge, the first application of this technique to supervised learning. We analyze the statistical and computational aspects of this method. We decompose this task into two stages: training (which corresponds operationally to smoothing/regularization) and prediction (which is achieved via Kirszbraun extension). Both are solved algorithmically via a novel multiplicative weight updates (MWU) scheme, which, for our problem formulation, achieves a quadratic runtime improvement over the state of the art. Our empirical results indicate a dramatic improvement over standard off-the-shelf solvers in our setting.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
We introduce a framework for performing regression between two Hilbert spaces. This is done based on Kirszbraun's extension theorem, to the best of our knowledge, the first application of this technique to supervised learning. We analyze the statistical and computational aspects of this method. We decompose this task into two stages: training (which corresponds operationally to smoothing/regularization) and prediction (which is achieved via Kirszbraun extension). Both are solved algorithmically via a novel multiplicative weight updates (MWU) scheme, which, for our problem formulation, achieves a quadratic runtime improvement over the state of the art. Our empirical results indicate a dramatic improvement over standard off-the-shelf solvers in our setting.
2023
Le, A. T.; Chalvatzaki, G.; Biess, A.; J.Peters,
Accelerating Motion Planning via Optimal Transport Conference
Advances in Neural Information Processing Systems (NeurIPS), 2023.
@conference{/c,
title = {Accelerating Motion Planning via Optimal Transport},
author = {A.T. Le and G. Chalvatzaki and A. Biess and J.Peters},
year = {2023},
date = {2023-12-01},
journal = {Advances in Neural Information Processing Systems (NeurIPS)},
publisher = {Advances in Neural Information Processing Systems (NeurIPS)},
abstract = {Motion planning is still an open problem for many disciplines, e.g., robotics, autonomous driving, due to their need for high computational resources that hinder real-time, efficient decision-making. A class of methods striving to provide smooth solutions is gradient-based trajectory optimization. However, those methods usually suffer from bad local minima, while for many settings, they may be inapplicable due to the absence of easy-to-access gradients of the optimization objectives. In response to these issues, we introduce Motion Planning via Optimal Transport (MPOT)—a gradient-free method that optimizes a batch of smooth trajectories over highly nonlinear costs, even for high-dimensional tasks, while imposing smoothness through a Gaussian Process dynamics prior via the planning-as-inference perspective. To facilitate batch trajectory optimization, we introduce an original zero-order and highly-parallelizable update rule—-the Sinkhorn Step, which uses the regular polytope family for its search directions. Each regular polytope, centered on trajectory waypoints, serves as a local cost-probing neighborhood, acting as a trust region where the Sinkhorn Step “transports” local waypoints toward low-cost regions. We theoretically show that Sinkhorn Step guides the optimizing parameters toward local minima regions of non-convex objective functions. We then show the efficiency of MPOT in a range of problems from low-dimensional point-mass navigation to high-dimensional whole-body robot motion planning, evincing its superiority compared to popular motion planners, paving the way for new applications of optimal transport in motion planning.},
keywords = {},
pubstate = {published},
tppubtype = {conference}
}
Motion planning is still an open problem for many disciplines, e.g., robotics, autonomous driving, due to their need for high computational resources that hinder real-time, efficient decision-making. A class of methods striving to provide smooth solutions is gradient-based trajectory optimization. However, those methods usually suffer from bad local minima, while for many settings, they may be inapplicable due to the absence of easy-to-access gradients of the optimization objectives. In response to these issues, we introduce Motion Planning via Optimal Transport (MPOT)—a gradient-free method that optimizes a batch of smooth trajectories over highly nonlinear costs, even for high-dimensional tasks, while imposing smoothness through a Gaussian Process dynamics prior via the planning-as-inference perspective. To facilitate batch trajectory optimization, we introduce an original zero-order and highly-parallelizable update rule—-the Sinkhorn Step, which uses the regular polytope family for its search directions. Each regular polytope, centered on trajectory waypoints, serves as a local cost-probing neighborhood, acting as a trust region where the Sinkhorn Step “transports” local waypoints toward low-cost regions. We theoretically show that Sinkhorn Step guides the optimizing parameters toward local minima regions of non-convex objective functions. We then show the efficiency of MPOT in a range of problems from low-dimensional point-mass navigation to high-dimensional whole-body robot motion planning, evincing its superiority compared to popular motion planners, paving the way for new applications of optimal transport in motion planning.
2022
Emuna, R.; Duffney, R.; Borowsky, A.; Biess, A.
Example-guided learning of stochastic human driving policies using deep reinforcement learning Journal Article
In: Neural Computing and Applications , vol. 35, pp. 16791–16804, 2022.
@article{Emuna2022,
title = {Example-guided learning of stochastic human driving policies using deep reinforcement learning},
author = {R. Emuna and R. Duffney and A. Borowsky and A. Biess},
year = {2022},
date = {2022-12-23},
journal = {Neural Computing and Applications },
volume = {35},
pages = {16791–16804},
abstract = {Deep reinforcement learning has been successfully applied to the generation of goal-directed behavior in artificial agents. However, existing algorithms are often not designed to reproduce human-like behavior, which may be desired in many environments, such as human–robot collaborations, social robotics and autonomous vehicles. Here we introduce a model-free and easy-to-implement deep reinforcement learning approach to mimic the stochastic behavior of a human expert by learning distributions of task variables from examples. As tractable use-cases, we study static and dynamic obstacle avoidance tasks for an autonomous vehicle on a highway road in simulation (Unity). Our control algorithm receives a feedback signal from two sources: a deterministic (handcrafted) part encoding basic task goals and a stochastic (data-driven) part that incorporates human expert knowledge. Gaussian processes are used to model human state distributions and to assess the similarity between machine and human behavior. Using this generic approach, we demonstrate that the learning agent acquires human-like driving skills and can generalize to new roads and obstacle distributions unseen during training.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Deep reinforcement learning has been successfully applied to the generation of goal-directed behavior in artificial agents. However, existing algorithms are often not designed to reproduce human-like behavior, which may be desired in many environments, such as human–robot collaborations, social robotics and autonomous vehicles. Here we introduce a model-free and easy-to-implement deep reinforcement learning approach to mimic the stochastic behavior of a human expert by learning distributions of task variables from examples. As tractable use-cases, we study static and dynamic obstacle avoidance tasks for an autonomous vehicle on a highway road in simulation (Unity). Our control algorithm receives a feedback signal from two sources: a deterministic (handcrafted) part encoding basic task goals and a stochastic (data-driven) part that incorporates human expert knowledge. Gaussian processes are used to model human state distributions and to assess the similarity between machine and human behavior. Using this generic approach, we demonstrate that the learning agent acquires human-like driving skills and can generalize to new roads and obstacle distributions unseen during training.
Kshirsagar, A.; Faibish, T.; Hoffman, G.; Biess, A.
Lessons learned from utilizing Guided Policy Search for human-robot handovers with a collaborative robot Proceedings
IEEE 2022 2nd International Conference on Robotics, Automation and Artificial Intelligence (RAAI), 2022, (oral presentation).
@proceedings{/,
title = {Lessons learned from utilizing Guided Policy Search for human-robot handovers with a collaborative robot},
author = {A. Kshirsagar and T. Faibish and G. Hoffman and A. Biess},
year = {2022},
date = {2022-12-09},
publisher = {2022 2nd International Conference on Robotics, Automation and Artificial Intelligence (RAAI)},
organization = {IEEE},
abstract = {We evaluate the performance of Guided Policy Search (GPS), a model-based reinforcement learning method, for generating the handover reaching motions of a collaborative robot arm. In a previous work, we evaluated GPS for the same task but only in a simulated environment. This paper provides a replication of the findings in simulation, along with new insights on GPS when used on a physical robot platform. First, we find that a policy learned in simulation does not transfer readily to the physical robot due to differences in model parameters and existing safety constraints on the real robot. Second, in order to successfully train a GPS model, the robot’s workspace needs to be severely reduced, owing to the joint-space limitations of the physical robot. Third, a policy trained with moving targets results in large worst-case errors even in regions spatially close to the training target locations. Our findings motivate further research towards utilizing GPS in humanrobot interaction settings, especially where safety constraints are imposed.},
note = {oral presentation},
keywords = {},
pubstate = {published},
tppubtype = {proceedings}
}
We evaluate the performance of Guided Policy Search (GPS), a model-based reinforcement learning method, for generating the handover reaching motions of a collaborative robot arm. In a previous work, we evaluated GPS for the same task but only in a simulated environment. This paper provides a replication of the findings in simulation, along with new insights on GPS when used on a physical robot platform. First, we find that a policy learned in simulation does not transfer readily to the physical robot due to differences in model parameters and existing safety constraints on the real robot. Second, in order to successfully train a GPS model, the robot’s workspace needs to be severely reduced, owing to the joint-space limitations of the physical robot. Third, a policy trained with moving targets results in large worst-case errors even in regions spatially close to the training target locations. Our findings motivate further research towards utilizing GPS in humanrobot interaction settings, especially where safety constraints are imposed.
2021
Manela, B; Biess, A
Curriculum learning with Hindsight Experience Replay for sequential object manipulation tasks Journal Article
In: Neural Networks, vol. 145, pp. 260-270, 2021, ISSN: 0893-6080.
@article{MANELA2022260,
title = {Curriculum learning with Hindsight Experience Replay for sequential object manipulation tasks},
author = {Manela, B and Biess, A},
doi = {https://doi.org/10.1016/j.neunet.2021.10.011},
issn = {0893-6080},
year = {2021},
date = {2021-11-11},
journal = {Neural Networks},
volume = {145},
pages = {260-270},
abstract = {Learning complex tasks from scratch is challenging and often impossible for humans as well as for artificial agents. Instead, a curriculum can be used, which decomposes a complex task – the target task – into a sequence of source tasks. Each source task is a simplified version of the next source task with increasing complexity. Learning then occurs gradually by training on each source task while using knowledge from the curriculum’s prior source tasks. In this study, we present a new algorithm that combines curriculum learning with Hindsight Experience Replay (HER), to learn sequential object manipulation tasks for multiple goals and sparse feedback. The algorithm exploits the recurrent structure inherent in many object manipulation tasks and implements the entire learning process in the original simulation without adjusting it to each source task. We test our algorithm on three challenging throwing tasks in simulation and show significant improvements compared to vanilla-HER.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Learning complex tasks from scratch is challenging and often impossible for humans as well as for artificial agents. Instead, a curriculum can be used, which decomposes a complex task – the target task – into a sequence of source tasks. Each source task is a simplified version of the next source task with increasing complexity. Learning then occurs gradually by training on each source task while using knowledge from the curriculum’s prior source tasks. In this study, we present a new algorithm that combines curriculum learning with Hindsight Experience Replay (HER), to learn sequential object manipulation tasks for multiple goals and sparse feedback. The algorithm exploits the recurrent structure inherent in many object manipulation tasks and implements the entire learning process in the original simulation without adjusting it to each source task. We test our algorithm on three challenging throwing tasks in simulation and show significant improvements compared to vanilla-HER.
Manela, B; Biess, A
Bias-reduced hindsight experience replay with virtual goal prioritization Journal Article
In: Neurocomputing, vol. 451, pp. 305-315, 2021, ISSN: 0925-2312.
@article{arXiv:1905.05498v1,
title = { Bias-reduced hindsight experience replay with virtual goal prioritization},
author = {Manela, B and Biess, A},
url = {https://arxiv.org/abs/1905.05498
https://youtu.be/xjAiwJiSeLc},
doi = {https://doi.org/10.1016/j.neucom.2021.02.090},
issn = {0925-2312},
year = {2021},
date = {2021-05-10},
journal = {Neurocomputing},
volume = {451},
pages = {305-315},
abstract = {Hindsight Experience Replay (HER) is a multi-goal reinforcement learning algorithm for sparse reward functions. The algorithm treats every failure as a success for an alternative (virtual) goal that has been achieved in the episode. Virtual goals are randomly selected, irrespective of which are most instructive for the agent. In this paper, we present two improvements over the existing HER algorithm. First, we prioritize virtual goals from which the agent will learn more valuable information. We call this property the instructiveness of the virtual goal and define it by a heuristic measure, which expresses how well the agent will be able to generalize from that virtual goal to actual goals. Secondly, we reduce existing bias in HER by the removal of misleading samples. To test our algorithms, we built two challenging environments with sparse reward functions. Our empirical results in both environments show vast improvement in the final success rate and sample efficiency when compared to the original HER algorithm},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Hindsight Experience Replay (HER) is a multi-goal reinforcement learning algorithm for sparse reward functions. The algorithm treats every failure as a success for an alternative (virtual) goal that has been achieved in the episode. Virtual goals are randomly selected, irrespective of which are most instructive for the agent. In this paper, we present two improvements over the existing HER algorithm. First, we prioritize virtual goals from which the agent will learn more valuable information. We call this property the instructiveness of the virtual goal and define it by a heuristic measure, which expresses how well the agent will be able to generalize from that virtual goal to actual goals. Secondly, we reduce existing bias in HER by the removal of misleading samples. To test our algorithms, we built two challenging environments with sparse reward functions. Our empirical results in both environments show vast improvement in the final success rate and sample efficiency when compared to the original HER algorithm
Kshirsagar, A.; Hoffman, G.; Biess, A.
Evaluating Guided Policy Search for human-robot handovers Journal Article
In: IEEE Robotics and Automation Letters, vol. 6, no. 2, pp. 3933-3940, 2021.
@article{,
title = {Evaluating Guided Policy Search for human-robot handovers},
author = {A. Kshirsagar and G. Hoffman and A. Biess },
doi = {10.1109/LRA.2021.3067299},
year = {2021},
date = {2021-04-01},
journal = {IEEE Robotics and Automation Letters},
volume = {6},
number = {2},
pages = {3933-3940},
abstract = {We present a robot controller for the reach phase of human-robot object handovers, a key competency for collaborative and assistive robots. We use Guided Policy Search (GPS), a reinforcement learning method, to train this robot controller in simulation. In contrast to existing handover controllers, GPS is data efficient and does not require prior knowledge of robot dynamics. We first formulate the reach phase of handovers as a reinforcement learning problem and then train a collaborative robot arm in a simulation environment. Our results indicate that GPS can be used to train a robot reach a moving target and the learnt controller works well even for large changes in the robot's mass, but is limited in the spatial generalizability to variations in the target trajectories. This work also contributes to the GPS literature, as handovers present previously unexplored challenges, including large spatial variations in test locations, moving targets and changes in the robot's mass.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
We present a robot controller for the reach phase of human-robot object handovers, a key competency for collaborative and assistive robots. We use Guided Policy Search (GPS), a reinforcement learning method, to train this robot controller in simulation. In contrast to existing handover controllers, GPS is data efficient and does not require prior knowledge of robot dynamics. We first formulate the reach phase of handovers as a reinforcement learning problem and then train a collaborative robot arm in a simulation environment. Our results indicate that GPS can be used to train a robot reach a moving target and the learnt controller works well even for large changes in the robot's mass, but is limited in the spatial generalizability to variations in the target trajectories. This work also contributes to the GPS literature, as handovers present previously unexplored challenges, including large spatial variations in test locations, moving targets and changes in the robot's mass.
2020
von Eschenbach, M. Ebner; Manela, B.; Peters, J.; Biess, A.
Bi-directional correspondence matrix for imitation learning between dissimilar manipulators Journal Article
In: 2020, (submitted).
@article{,
title = {Bi-directional correspondence matrix for imitation learning between dissimilar manipulators},
author = {M. Ebner von Eschenbach and B. Manela and J. Peters and A. Biess },
year = {2020},
date = {2020-02-25},
abstract = {A major challenge in imitation learning is the correspondence problem: how to establish corresponding states and actions between the expert and learner, when the embodiments of the agents are different (morphology, dynamics, degrees of freedom, etc.)? Many existing approaches in imitation learning circumvent the correspondence problem, for example, by using kinesthetic teaching or teleoperation. In this study, the correspondence problem is addressed by automatically establishing correspondence between links of different embodiments using a bi-directional correspondence matrix. A distance measure between embodiments is then defined by the average of all mutual link distances weighted by the correspondence matrix. This metric-based approach is applied to static pose and movement imitation tasks between dissimilar planar manipulators and anthropomorphic robotic arms in simulation. We find that the approach is well suited for describing the similarity between embodiments and for learning imitation policies by distance minimization.},
note = {submitted},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
A major challenge in imitation learning is the correspondence problem: how to establish corresponding states and actions between the expert and learner, when the embodiments of the agents are different (morphology, dynamics, degrees of freedom, etc.)? Many existing approaches in imitation learning circumvent the correspondence problem, for example, by using kinesthetic teaching or teleoperation. In this study, the correspondence problem is addressed by automatically establishing correspondence between links of different embodiments using a bi-directional correspondence matrix. A distance measure between embodiments is then defined by the average of all mutual link distances weighted by the correspondence matrix. This metric-based approach is applied to static pose and movement imitation tasks between dissimilar planar manipulators and anthropomorphic robotic arms in simulation. We find that the approach is well suited for describing the similarity between embodiments and for learning imitation policies by distance minimization.
2018
Litvak, Y; Biess, A*; Bar-Hillel, A*
Learning a high-precision robotic assembly task from simulated depth images (* equal contribution) Journal Article
In: 2018, (Proceedings ICRA 2019).
@article{Biess2018,
title = {Learning a high-precision robotic assembly task from simulated depth images (* equal contribution)},
author = {Litvak, Y and Biess, A* and Bar-Hillel, A*},
url = {https://arxiv.org/abs/1809.10699
https://armin-biess.net/wp-content/uploads/2018/10/Litvak-et-al-2018-Learning-a-High-Precision-Robotic-Assembly-Task-Using-Pose-Estimation-from-Simulated-Depth-Images.pdf
https://www.youtube.com/watch?v=uMvq2-Tg-9g},
year = {2018},
date = {2018-09-22},
abstract = {Most of industrial robotic assembly tasks today require fixed initial conditions for successful assembly. These constraints induce high production costs and low adaptability to new tasks. In this work we aim towards flexible and adaptable robotic assembly by using 3D CAD models for all parts to be assembled. We focus on a generic assembly task - the Siemens Innovation Challenge - in which a robot needs to assemble a gear-like mechanism with high precision into an operating system. To obtain the millimeter-accuracy required for this task and industrial settings alike, we use a depth camera mounted near the robot’s end-effector. We present a high-accuracy three-stage pose estimation pipeline based on deep convolutional neural networks, which includes detection, pose estimation, refinement, and handling of near- and full symmetries of parts. The networks are trained on simulated depth images by means to ensure successful transfer to the real robot. We obtain an average pose estimation error of 2.14 millimeters and 1.09 degree leading to 88.6% success rate for robotic assembly of randomly distributed parts. To the best of our knowledge, this is the first time that the Siemens Innovation Challenge is fully solved, opening up new possibilities for automated industrial assembly.},
note = {Proceedings ICRA 2019},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Most of industrial robotic assembly tasks today require fixed initial conditions for successful assembly. These constraints induce high production costs and low adaptability to new tasks. In this work we aim towards flexible and adaptable robotic assembly by using 3D CAD models for all parts to be assembled. We focus on a generic assembly task - the Siemens Innovation Challenge - in which a robot needs to assemble a gear-like mechanism with high precision into an operating system. To obtain the millimeter-accuracy required for this task and industrial settings alike, we use a depth camera mounted near the robot’s end-effector. We present a high-accuracy three-stage pose estimation pipeline based on deep convolutional neural networks, which includes detection, pose estimation, refinement, and handling of near- and full symmetries of parts. The networks are trained on simulated depth images by means to ensure successful transfer to the real robot. We obtain an average pose estimation error of 2.14 millimeters and 1.09 degree leading to 88.6% success rate for robotic assembly of randomly distributed parts. To the best of our knowledge, this is the first time that the Siemens Innovation Challenge is fully solved, opening up new possibilities for automated industrial assembly.
2013
Biess, Armin
In: 2013.
@article{,
title = {Shaping of arm configuration space by prescription of non-Euclidean metrics with applications to human motor control. Physical Review E 87, 012729},
author = {Armin Biess},
url = {https://armin-biess.net/wp-content/uploads/2018/10/Biess-et-al-2013-Shaping-of-Arm-Configuration-Space-by-Prescription-of-non-Euclidean-Metrics.pdf},
year = {2013},
date = {2013-08-02},
abstract = {The study of the kinematic and dynamic features of human arm movements provides insights into the computational strategies underlying human motor control. In this paper a differential geometric (coordinate-independent) approach to movement control is taken by endowing arm configuration space with different non-Euclidean metrics to study the predictions of the generalized minimum-jerk (MJ) model in the resulting Riemannian manifold for different types of human arm movements. For each metric space the solution of the generalized MJ model is given by re-parametrized geodesic paths. This geodesic model is applied to a variety of motor tasks ranging from three-dimensional unconstrained movements of a four degree of freedom arm between point-like targets to constrained movements where the hand location is confined to a surface (e.g.,a sphere) or a curve (e.g., an ellipse). For the latter speed-curvature relations are derived depending on the boundary conditions imposed (periodic or non-periodic) and the compatibility with the empirical one-third power law is shown. Based on these theoretical studies and recent experimental findings I argue that the sensorimotor system may shape arm configuration space by learning metric structures through sensorimotor feed-back and that geodesics may be an emergent property of the motor system.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The study of the kinematic and dynamic features of human arm movements provides insights into the computational strategies underlying human motor control. In this paper a differential geometric (coordinate-independent) approach to movement control is taken by endowing arm configuration space with different non-Euclidean metrics to study the predictions of the generalized minimum-jerk (MJ) model in the resulting Riemannian manifold for different types of human arm movements. For each metric space the solution of the generalized MJ model is given by re-parametrized geodesic paths. This geodesic model is applied to a variety of motor tasks ranging from three-dimensional unconstrained movements of a four degree of freedom arm between point-like targets to constrained movements where the hand location is confined to a surface (e.g.,a sphere) or a curve (e.g., an ellipse). For the latter speed-curvature relations are derived depending on the boundary conditions imposed (periodic or non-periodic) and the compatibility with the empirical one-third power law is shown. Based on these theoretical studies and recent experimental findings I argue that the sensorimotor system may shape arm configuration space by learning metric structures through sensorimotor feed-back and that geodesics may be an emergent property of the motor system.
2011
Biess, A; Korkotian, E; Holcman, D
Barriers to diffusion in dendrites and estimation of calcium spread following synaptic input. PLoS Comput Biol 7(10): e1002182 Journal Article
In: 2011.
@article{A2011b,
title = {Barriers to diffusion in dendrites and estimation of calcium spread following synaptic input. PLoS Comput Biol 7(10): e1002182},
author = {Biess, A and Korkotian, E and Holcman, D},
url = {https://armin-biess.net/wp-content/uploads/2018/10/Biess-et-al-2011-Diffusion-in-Dendritic-Spines-The-Role-of-Geometry-1.pdf},
year = {2011},
date = {2011-08-12},
abstract = {The motion of ions, molecules or proteins in dendrites is restricted by cytoplasmic obstacles such as organelles, microtubule and actin network. We are interested in calcium spread, which can be very restricted in aspiny dendrites due to speci c calcium channels. To study how molecular crowding can restrict calcium dynamics in dendrites, in parallel to uncaging experiments with inert dyes, we build a biophysical model to evaluate the separate contributions of cytoplasmic crowding, calcium buffers, pumps and dendritic spines, which cannot be dissociated experimentally. By deriving a one dimensional diffusion equation, we show that depending on the crowding concentration and organization, impenetrable organelles slows dendritic diffusion. We estimate the mean time a Brownian object (which can be a transcription factor or a macromolecule) takes to travel inside a nonbranching dendrite. By comparing di usion of an inert dye in a spiny dendrite and in a thin glass tube of similar size, we found that crowding can decrease the apparent di usion by a factor 20. Finally, using numerical simulations, we estimate calcium spreads in a crowded dendrite. We find that for moderate crowding, calcium dynamics is mainly regulated by bu er concentration, but not by the cytoplasmic obstacles, dendritic spines or synaptic inputs. We conclude that calcium spread in dendrites is limited to small microdomains of the order of a few microns.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
The motion of ions, molecules or proteins in dendrites is restricted by cytoplasmic obstacles such as organelles, microtubule and actin network. We are interested in calcium spread, which can be very restricted in aspiny dendrites due to speci c calcium channels. To study how molecular crowding can restrict calcium dynamics in dendrites, in parallel to uncaging experiments with inert dyes, we build a biophysical model to evaluate the separate contributions of cytoplasmic crowding, calcium buffers, pumps and dendritic spines, which cannot be dissociated experimentally. By deriving a one dimensional diffusion equation, we show that depending on the crowding concentration and organization, impenetrable organelles slows dendritic diffusion. We estimate the mean time a Brownian object (which can be a transcription factor or a macromolecule) takes to travel inside a nonbranching dendrite. By comparing di usion of an inert dye in a spiny dendrite and in a thin glass tube of similar size, we found that crowding can decrease the apparent di usion by a factor 20. Finally, using numerical simulations, we estimate calcium spreads in a crowded dendrite. We find that for moderate crowding, calcium dynamics is mainly regulated by bu er concentration, but not by the cytoplasmic obstacles, dendritic spines or synaptic inputs. We conclude that calcium spread in dendrites is limited to small microdomains of the order of a few microns.
Biess, A; Flash, T; Liebermann, DG
Riemannian geometric approach to human arm dynamics, movement optimization and invariance. Physical Review E 83, 031927 Journal Article
In: 2011.
@article{A2011,
title = {Riemannian geometric approach to human arm dynamics, movement optimization and invariance. Physical Review E 83, 031927},
author = {Biess, A and Flash, T and Liebermann, DG},
url = {https://armin-biess.net/wp-content/uploads/2018/10/Biess-et-al-2011-Riemannian-Geometric-Approach-To-Human-Arm-Dynamics-Movement-Optimization-and-Invariance.pdf},
year = {2011},
date = {2011-08-04},
abstract = {We present a generally covariant formulation of human arm dynamics and optimization principles in Riemannian configuration space. We extend the one-parameter family of mean-squared-derivative (MSD) cost functionals from Euclidean to Riemannian space, and we show that they are mathematically identical to the corresponding dynamic costs when formulated in a Riemannian space equipped with the kinetic energy metric. In particular, we derive the equivalence of the minimum-jerk and minimum-torque change models in this metric space. Solutions of the one-parameter family of MSD variational problems in Riemannian space are given by (reparametrized) geodesic paths, which correspond to movements with least muscular effort. Finally, movement invariants are derived from symmetries of the Riemannian manifold. We argue that the geometrical structure imposed on the arm's configuration space may provide insights into the emerging properties of the movements generated by the motor system.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
We present a generally covariant formulation of human arm dynamics and optimization principles in Riemannian configuration space. We extend the one-parameter family of mean-squared-derivative (MSD) cost functionals from Euclidean to Riemannian space, and we show that they are mathematically identical to the corresponding dynamic costs when formulated in a Riemannian space equipped with the kinetic energy metric. In particular, we derive the equivalence of the minimum-jerk and minimum-torque change models in this metric space. Solutions of the one-parameter family of MSD variational problems in Riemannian space are given by (reparametrized) geodesic paths, which correspond to movements with least muscular effort. Finally, movement invariants are derived from symmetries of the Riemannian manifold. We argue that the geometrical structure imposed on the arm's configuration space may provide insights into the emerging properties of the movements generated by the motor system.
2008
Knutsen, P*; Biess, A*; Ahissar, E
Vibrissal whisking in 3D: Tight coupling of azimuth, elevation and torsion during natural whisking. Neuron 59(1):35-42 (*: equal contribution). Journal Article
In: 2008.
@article{P*2008,
title = {Vibrissal whisking in 3D: Tight coupling of azimuth, elevation and torsion during natural whisking. Neuron 59(1):35-42 (*: equal contribution).},
author = {Knutsen, P* and Biess, A* and Ahissar, E},
url = {https://armin-biess.net/wp-content/uploads/2018/08/VibrissalKinematicsIn3DThightCouplingOfAzimuthElevationAndTorsionAcrossDifferentWhsikingModes_Knutsen_et_al.pdf},
year = {2008},
date = {2008-08-01},
abstract = {Perception is usually an active process by which action selects and affects sensory information. During rodent active touch, whisker kinematics influences how objects activate sensory receptors. In order to fully characterize whisker motion, we reconstructed whisker position in 3D and decomposed whisker motion to all its degrees of freedom. We found that, across behavioral modes, in both head-fixed and freely moving rats, whisker motion is characterized by translational movements and three rotary components: azimuth, elevation, and torsion. Whisker torsion, which has not previously been described, was large (up to 100°), and torsional angles were highly correlated with whisker azimuths. The coupling of azimuth and torsion was consistent across whisking epochs and rats and was similar along rows but systematically varied across rows such that rows A and E counterrotated. Torsional rotation of the whiskers enables contact information to be mapped onto the circumference of the whisker follicles in a predictable manner across protraction-retraction cycles.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Perception is usually an active process by which action selects and affects sensory information. During rodent active touch, whisker kinematics influences how objects activate sensory receptors. In order to fully characterize whisker motion, we reconstructed whisker position in 3D and decomposed whisker motion to all its degrees of freedom. We found that, across behavioral modes, in both head-fixed and freely moving rats, whisker motion is characterized by translational movements and three rotary components: azimuth, elevation, and torsion. Whisker torsion, which has not previously been described, was large (up to 100°), and torsional angles were highly correlated with whisker azimuths. The coupling of azimuth and torsion was consistent across whisking epochs and rats and was similar along rows but systematically varied across rows such that rows A and E counterrotated. Torsional rotation of the whiskers enables contact information to be mapped onto the circumference of the whisker follicles in a predictable manner across protraction-retraction cycles.
2007
Biess, A; Liebermann, DG; T, Flash
In: 2007.
@article{A2007,
title = {A computational model for redundant human 3D pointing movements: Integration of independent spatial and temporal motor plans simplifies movement dynamics. J Neurosci 27(48):13045-13064.},
author = {Biess, A and Liebermann, DG and Flash T},
url = {https://armin-biess.net/wp-content/uploads/2018/10/Biess-et-al-2007A-Computational-Model-For-Redundant-Human-3D-Pointing-Movements.pdf},
year = {2007},
date = {2007-08-03},
abstract = {Few computational models have addressed the spatiotemporal features of unconstrained three-dimensional (3D) arm motion. Empirical observations made on hand paths, speed profiles, and arm postures during point-to-point movements led to the assumption that hand path and arm posture are independent of movement speed, suggesting that the geometric and temporal properties of movements are decoupled. In this study, we present a computational model of 3D movements for an arm with four degrees of freedom based on the assumption that optimization principles are separately applied at the geometric and temporal levels of control. Geometric properties (path and posture) are defined in terms of geodesic paths with respect to the kinetic energy metric in the Riemannian configuration space. Accordingly, a geodesic path can be generated with less muscular effort than on any other, nongeodesic path, because the sum of all configuration-speed-dependent torques vanishes. The temporal properties of the movement (speed) are determined in task space by minimizing the squared jerk along the selected end-effector path. The integration of both planning levels into a single spatiotemporal representation simplifies the control of arm dynamics along geodesic paths and results in movements with near minimal torque change and minimal peak value of kinetic energy. Thus, the application of Riemannian geometry allows for a reconciliation of computational models previously proposed for the description of arm movements. We suggest that geodesics are an emergent property of the motor system through the exploration of dynamical space. Our data validated the predictions for joint trajectories, hand paths, final postures, speed profiles, and driving torques.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Few computational models have addressed the spatiotemporal features of unconstrained three-dimensional (3D) arm motion. Empirical observations made on hand paths, speed profiles, and arm postures during point-to-point movements led to the assumption that hand path and arm posture are independent of movement speed, suggesting that the geometric and temporal properties of movements are decoupled. In this study, we present a computational model of 3D movements for an arm with four degrees of freedom based on the assumption that optimization principles are separately applied at the geometric and temporal levels of control. Geometric properties (path and posture) are defined in terms of geodesic paths with respect to the kinetic energy metric in the Riemannian configuration space. Accordingly, a geodesic path can be generated with less muscular effort than on any other, nongeodesic path, because the sum of all configuration-speed-dependent torques vanishes. The temporal properties of the movement (speed) are determined in task space by minimizing the squared jerk along the selected end-effector path. The integration of both planning levels into a single spatiotemporal representation simplifies the control of arm dynamics along geodesic paths and results in movements with near minimal torque change and minimal peak value of kinetic energy. Thus, the application of Riemannian geometry allows for a reconciliation of computational models previously proposed for the description of arm movements. We suggest that geodesics are an emergent property of the motor system through the exploration of dynamical space. Our data validated the predictions for joint trajectories, hand paths, final postures, speed profiles, and driving torques.
Biess, A; Korkotian, E; D, Holcman
Diffusion in a dendritic spine, the role of geometry. Phys Rev E 76: 0219222. Journal Article
In: 2007.
@article{A2007b,
title = {Diffusion in a dendritic spine, the role of geometry. Phys Rev E 76: 0219222.},
author = {Biess, A and Korkotian, E and Holcman D},
url = {https://armin-biess.net/wp-content/uploads/2018/10/Biess-et-al-2007-Barriers-to-Diffusion-2.pdf},
year = {2007},
date = {2007-08-03},
abstract = {Dendritic spines, the sites where excitatory synapses are made in most neurons, can dynamically regulate diffusing molecules by changing their shape. We present here a combination of theory, simulations, and experiments to quantify the diffusion time course in dendritic spines. We derive analytical formulas and compared them to Brownian simulations for the mean sojourn time a diffusing molecule stays inside a dendritic spine when either the molecule can reenter the spine head or not, once it is located in the spine neck. We show that the spine length is the fundamental regulatory geometrical parameter for the diffusion decay rate in the neck only. By changing the spine length, dendritic spines can be dynamically coupled or uncoupled to their parent dendrites, which regulates diffusion, and this property makes them unique structures, different from static dendrites.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Dendritic spines, the sites where excitatory synapses are made in most neurons, can dynamically regulate diffusing molecules by changing their shape. We present here a combination of theory, simulations, and experiments to quantify the diffusion time course in dendritic spines. We derive analytical formulas and compared them to Brownian simulations for the mean sojourn time a diffusing molecule stays inside a dendritic spine when either the molecule can reenter the spine head or not, once it is located in the spine neck. We show that the spine length is the fundamental regulatory geometrical parameter for the diffusion decay rate in the neck only. By changing the spine length, dendritic spines can be dynamically coupled or uncoupled to their parent dendrites, which regulates diffusion, and this property makes them unique structures, different from static dendrites.
2006
Biess, A; Nagurka, M; Flash, T
Simulating discrete and rhythmic multi-joint human arm movements by optimization of nonlinear performance indices. Biol Cybern 95: 31-53. Journal Article
In: 2006.
@article{A2006,
title = {Simulating discrete and rhythmic multi-joint human arm movements by optimization of nonlinear performance indices. Biol Cybern 95: 31-53.},
author = {Biess, A and Nagurka, M and Flash, T},
url = {https://armin-biess.net/wp-content/uploads/2018/08/SimulatingDiscreteAndRhytmicMultiJointHumanArmMovementsByOptimizationOfNonlinearPerformancesIndices_Biess_et_al_2006.pdf},
year = {2006},
date = {2006-08-04},
abstract = {An optimization approach applied to mechanical linkage models is used to simulate human arm movements. Predicted arm trajectories are the result of minimizing a nonlinear performance index that depends on kinematic or dynamic variables of the movement. A robust optimization algorithm is presented that computes trajectories which satisfy the necessary conditions with high accuracy. It is especially adapted to the analysis of discrete and rhythmic movements. The optimization problem is solved by parameterizing each generalized coordinate (e.g., joint angular displacement) in terms of Jacobi polynomials and Fourier series, depending on whether discrete or rhythmic movements are considered, combined with amultiple shooting algorithm. The parameterization of coordinates has two advantages. First, it provides an initial guess for the multiple shooting algorithmwhich solves the optimization problem with high accuracy. Second, it leads to a low dimensional representation of discrete and rhythmic movements in terms of expansion coefficients. The selection of a suitable feature space is an important prerequisite for comparison, recognition and classification ofmovements. In addition, the separate computational analysis of discrete and rhythmic movements is motivated by their distinct neurophysiological realizations in the cortex. By investigating different performance indices subject to different boundary conditions, the approach can be used to examine possible strategies that humans adopt in selecting specific arm motions for the performance of different tasks in a plane and in three dimensional space.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
An optimization approach applied to mechanical linkage models is used to simulate human arm movements. Predicted arm trajectories are the result of minimizing a nonlinear performance index that depends on kinematic or dynamic variables of the movement. A robust optimization algorithm is presented that computes trajectories which satisfy the necessary conditions with high accuracy. It is especially adapted to the analysis of discrete and rhythmic movements. The optimization problem is solved by parameterizing each generalized coordinate (e.g., joint angular displacement) in terms of Jacobi polynomials and Fourier series, depending on whether discrete or rhythmic movements are considered, combined with amultiple shooting algorithm. The parameterization of coordinates has two advantages. First, it provides an initial guess for the multiple shooting algorithmwhich solves the optimization problem with high accuracy. Second, it leads to a low dimensional representation of discrete and rhythmic movements in terms of expansion coefficients. The selection of a suitable feature space is an important prerequisite for comparison, recognition and classification ofmovements. In addition, the separate computational analysis of discrete and rhythmic movements is motivated by their distinct neurophysiological realizations in the cortex. By investigating different performance indices subject to different boundary conditions, the approach can be used to examine possible strategies that humans adopt in selecting specific arm motions for the performance of different tasks in a plane and in three dimensional space.
Liebermann, DG; Biess, A; Gielen, CC; Flash, T
Intrinsic joint kinematic planning - II: Hand-path predictions based on a Listing's plane constraint. Exp Brain Res 171:155-73. Journal Article
In: 2006.
@article{DG2006,
title = {Intrinsic joint kinematic planning - II: Hand-path predictions based on a Listing's plane constraint. Exp Brain Res 171:155-73.},
author = {Liebermann, DG and Biess, A and Gielen, CC and Flash, T},
url = {https://armin-biess.net/wp-content/uploads/2018/10/Liebermann-et-al-2006-Intrinsic-Joint-Kinematic-Planning-II.pdf},
year = {2006},
date = {2006-08-04},
abstract = {This study was aimed at examining the assumption that three-dimensional (3D) hand movements follow specific paths that are dictated by the operation of a Listing’s law constraint at the intrinsic joint level of the arm. A kinematic model was used to simulate hand paths during 3D point-to-point movements. The model was based on the assumption that the shoulder obeys a 2D Listing’s constraint and that rotations are about fixed single-axes. The elbow rotations were assumed to relate linearly to those of the shoulder. Both joints were assumed to rotate without reversals, and to start and end rotating simultaneously with zero initial and final velocities. Model predictions were compared to experimental observations made on four right-handed individuals that moved toward virtual objects in ‘‘extended arm’’, ‘‘radial’’, and ‘‘frontal plane’’ movement types. The results showed that the model was partially successful in accounting for the observed behavior. Best hand-path predictions were obtained for extended arm movements followed by radial ones. Frontal plane movements resulted in the largest discrepancies between the predicted and the observed paths. During such movements, the upper arm rotation vectors did not obey Listing’s law and this may explain the observed discrepancies. For other movement types, small deviations from the predicted paths were observed which could be explained by the fact that single-axis rotations were not followed even though the rotation vectors remained within Listing’s plane. Dynamic factors associated with movement execution, which were not taken into account in our purely kinematic approach, could also explain some of these small discrepancies. In conclusion, a kinematic model based on Listing’s law can describe an intrinsic joint strategy for the control of arm orientation during pointing and reaching movements, but only in conditions in which the movements closely obey the Listing’s plane assumption.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
This study was aimed at examining the assumption that three-dimensional (3D) hand movements follow specific paths that are dictated by the operation of a Listing’s law constraint at the intrinsic joint level of the arm. A kinematic model was used to simulate hand paths during 3D point-to-point movements. The model was based on the assumption that the shoulder obeys a 2D Listing’s constraint and that rotations are about fixed single-axes. The elbow rotations were assumed to relate linearly to those of the shoulder. Both joints were assumed to rotate without reversals, and to start and end rotating simultaneously with zero initial and final velocities. Model predictions were compared to experimental observations made on four right-handed individuals that moved toward virtual objects in ‘‘extended arm’’, ‘‘radial’’, and ‘‘frontal plane’’ movement types. The results showed that the model was partially successful in accounting for the observed behavior. Best hand-path predictions were obtained for extended arm movements followed by radial ones. Frontal plane movements resulted in the largest discrepancies between the predicted and the observed paths. During such movements, the upper arm rotation vectors did not obey Listing’s law and this may explain the observed discrepancies. For other movement types, small deviations from the predicted paths were observed which could be explained by the fact that single-axis rotations were not followed even though the rotation vectors remained within Listing’s plane. Dynamic factors associated with movement execution, which were not taken into account in our purely kinematic approach, could also explain some of these small discrepancies. In conclusion, a kinematic model based on Listing’s law can describe an intrinsic joint strategy for the control of arm orientation during pointing and reaching movements, but only in conditions in which the movements closely obey the Listing’s plane assumption.
Liebermann, DG; Biess, A; Friedman, J; Gielen, CC; Flash, T
In: 2006.
@article{DG2006b,
title = {Intrinsic joint kinematic planning - I: Reassessing the Listing's Law constraint in the control of three-dimensional arm movements. Exp Brain Res 171:139-54. },
author = {Liebermann, DG and Biess, A and Friedman, J and Gielen, CC and Flash, T},
url = {https://armin-biess.net/wp-content/uploads/2018/08/IntrinsicJointKinematicPlanning1_Liebermann_et_al_2006.pdf},
year = {2006},
date = {2006-08-04},
abstract = {This study tested the validity of the assumption that intrinsic kinematic constraints, such as Listing's law, can account for the geometric features of threedimensional arm movements. In principle, if the arm joints follow a Listing's constraint, the hand paths may be predicted. Four individuals performed 'extended arm', 'radial', 'frontal plane', and 'random mixed' movements to visual targets to test Listing's law assumption. Three-dimensional rotation vectors of the upper arm and forearm were calculated from three dimensional marker data. Data fitting techniques were used to test Donders' and Listing's laws. The coefficient values obtained from fitting rotation vectors to the surfaces described by a second-order equation were analyzed. The results showed that the coefficients that represent curvature and twist of the surfaces were often not significantly different from zero, particularly not during randomly mixed and extended arm movements. These coefficients for forearm rotations were larger compared to those for the upper arm segment rotations. The mean thickness of the rotation surfaces ranged between 1.7° and 4.7° for the rotation vectors of the upper arm segment and 2.6° and 7.5° for those of the forearm. During frontal plane movements, forearm rotations showed large twist scores while upper arm segment rotations showed large curvatures, although the thickness of the surfaces remained low. The curvatures, but not the thicknesses of the surfaces, were larger for large versus small amplitude radial movements. In conclusion, when examining the surfaces obtained for the different movement types, the rotation vectors may lie within manifolds that are anywhere between curved or twisted manifolds. However, a two-dimensional thick surface may roughly represent a global arm constraint. Our findings suggest that Listing's law is implemented for some types of arm movement, such as pointing to targets with the extended arm and during radial reaching movements.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
This study tested the validity of the assumption that intrinsic kinematic constraints, such as Listing's law, can account for the geometric features of threedimensional arm movements. In principle, if the arm joints follow a Listing's constraint, the hand paths may be predicted. Four individuals performed 'extended arm', 'radial', 'frontal plane', and 'random mixed' movements to visual targets to test Listing's law assumption. Three-dimensional rotation vectors of the upper arm and forearm were calculated from three dimensional marker data. Data fitting techniques were used to test Donders' and Listing's laws. The coefficient values obtained from fitting rotation vectors to the surfaces described by a second-order equation were analyzed. The results showed that the coefficients that represent curvature and twist of the surfaces were often not significantly different from zero, particularly not during randomly mixed and extended arm movements. These coefficients for forearm rotations were larger compared to those for the upper arm segment rotations. The mean thickness of the rotation surfaces ranged between 1.7° and 4.7° for the rotation vectors of the upper arm segment and 2.6° and 7.5° for those of the forearm. During frontal plane movements, forearm rotations showed large twist scores while upper arm segment rotations showed large curvatures, although the thickness of the surfaces remained low. The curvatures, but not the thicknesses of the surfaces, were larger for large versus small amplitude radial movements. In conclusion, when examining the surfaces obtained for the different movement types, the rotation vectors may lie within manifolds that are anywhere between curved or twisted manifolds. However, a two-dimensional thick surface may roughly represent a global arm constraint. Our findings suggest that Listing's law is implemented for some types of arm movement, such as pointing to targets with the extended arm and during radial reaching movements.
2001
Biess, A; Flash, T; Liebermann, DG
Multijoint point-to-point arm movements of humans in 3D-space: Minimum kinetic energy paths. In: Proceed Tenth Biennal Conference Int Graphonomics Soc, University of Nijmegen, 142-146. Journal Article
In: 2001.
@article{A2001,
title = {Multijoint point-to-point arm movements of humans in 3D-space: Minimum kinetic energy paths. In: Proceed Tenth Biennal Conference Int Graphonomics Soc, University of Nijmegen, 142-146.},
author = {Biess, A and Flash, T and Liebermann, DG},
year = {2001},
date = {2001-08-02},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Patents
2016
Suissa, Avshalom; Biess, Armin
Sensorless position control of active-material actuators, US 9829304 Patent
2016.
@patent{Suissa2016,
title = {Sensorless position control of active-material actuators, US 9829304},
author = {Avshalom Suissa and Armin Biess},
year = {2016},
date = {2016-08-05},
keywords = {},
pubstate = {published},
tppubtype = {patent}
}
Biess, Armin; Zelman, Ido; Jordokovsky, Mario
Dynamic safety shields for situation assessment and decision making in collision avoidance tasks, US 9280899 Patent
2016.
@patent{Biess2016,
title = {Dynamic safety shields for situation assessment and decision making in collision avoidance tasks, US 9280899},
author = {Armin Biess and Ido Zelman and Mario Jordokovsky},
year = {2016},
date = {2016-08-04},
keywords = {},
pubstate = {published},
tppubtype = {patent}
}