pith. sign in

arxiv: 2505.12214 · v2 · submitted 2025-05-18 · 💻 cs.RO · cs.IT· math.IT

Behavior Synthesis via Contact-Aware Fisher Information Maximization

Hrishikesh Sathyanarayan , Ian Abraham This is my paper

Pith reviewed 2026-05-22 15:16 UTC · model grok-4.3

classification 💻 cs.RO cs.ITmath.IT
keywords contact dynamicsFisher informationbehavior synthesisparameter learningoptimal experimental designroboticscontact-rich interactionsinformation maximization
0
0 comments X

The pith

Maximizing a contact-aware Fisher information measure synthesizes robot behaviors that generate data for efficient object parameter learning.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper aims to establish that optimal experimental design with a derived contact-aware Fisher information measure can synthesize robot behaviors to produce rich contact data useful for learning object parameters. This matters because contact interactions tend to be sparse and non-smooth, so random or passive data collection often fails to yield enough information for accurate characterization. If correct, robots would actively seek out interactions that efficiently reveal parameters across multiple learning tasks. A sympathetic reader would care because it turns physical contacts into a deliberate source of knowledge rather than an occasional byproduct of motion.

Core claim

The central claim is that a contact-aware Fisher information measure characterizes information-rich contact behaviors, enabling the synthesis of emergent robot actions that excite contacts and thereby improve parameter learning efficiency across a range of examples, as shown in robotic experiments.

What carries the argument

The contact-aware Fisher information measure, which quantifies information gain about object parameters specifically from contact dynamics and is maximized to guide behavior synthesis in an optimal experimental design setting.

Load-bearing premise

The contact dynamics model is accurate enough that the computed Fisher information matrix remains reliable and that gains observed in simulation transfer to real-world parameter learning without large model mismatch.

What would settle it

If real-robot trials using the synthesized behaviors show no improvement or increased error in estimated object parameters relative to baseline behaviors that ignore contact awareness, the central claim would be falsified.

Figures

Figures reproduced from arXiv: 2505.12214 by Hrishikesh Sathyanarayan, Ian Abraham.

Figure 1
Figure 1. Figure 1: Emergent Contact-Based Learning Behaviors. Here, we show emergent tactile behaviors resulting from the proposed contact-aware Fisher information maximization method that results in human-like tactile behaviors for learning (a) mass and weight, (b) friction and textures, (c) stiffness, and (d) shape [20]. Additional media and code is provided in https://github.com/ialab-yale/contact aware active learning . … view at source ↗
Figure 2
Figure 2. Figure 2: Collapsing Parameter Uncertainty Through Contact. In this example, the robot seeks to estimate the friction coefficient, µ, of a nearby table. Given an initial prior of the parameter estimate, the robot seeks to update its belief of the friction coefficient by interacting with the environment through contact. In doing so, the robot collects contact sensor measurements that hold information, F (green curve)… view at source ↗
Figure 3
Figure 3. Figure 3: Experimental Setup. The experimental setup for (a) hefting experiments for mass estimation, (b) rubbing experiments for friction coefficient estimation, (c) pressing experiments for material stiffness and dampening estimation, and (d) contouring experiments for shape estimation. B. Closing the Loop Given the solution to Eq. (12), the following step is to exe￾cute the behaviors on the robot, record the meas… view at source ↗
Figure 5
Figure 5. Figure 5: Emergent Behaviors from Experimental Design. Time series of experiments conducted over all scenarios. By maximizing contact-aware Fisher information, the robot seeks information-rich contacts for efficient identification of (a) mass, (b) friction coefficient, (c) material properties, and (d) shape. Simulation results are visualized on MuJoCo [42]. The robots were given a planned trajectory to execute, wher… view at source ↗
Figure 6
Figure 6. Figure 6: Contact Excitations Improves Parameter Learning. We compare our experimental design approach with a baseline similar to works in [25] for all robot scenarios. The baseline implements a belief-space planning controller that reasons about contact na¨ıvely. We evaluate each robot scenario over a finite number of experiments. Here, we model the parameter error as %err = 100(ˆθ − θ ∗ )/θ∗ , where ˆθ is the curr… view at source ↗
Figure 7
Figure 7. Figure 7: Robustness to Initialization. We study robustness of our approach is to 7 initial parameter priors for all scenarios. Note here that we model the parameter error as a distance δ = | ˆθ − θ ∗ |. We observe that our contact-aware Fisher information maximization approach is resilient across a diverse initialization of priors over all experiments. scenarios. In Fig.7, we demonstrate the robustness of our appro… view at source ↗
Figure 8
Figure 8. Figure 8: Hefting Behavior Information Landscape. The robot excites penetration deformation ϕn by varying its hand positions and non-zero contact normal velocity vn in order to excite information-rich contact about mass, resulting in an up-and-down hefting behavior shown in [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Optimizing Over Contact-Seeking Information Landscapes. We show the contact-aware Fisher information landscape that our experimental design method optimizes over for the stiffness/dampening and friction estimation scenarios. Nonlinear Optimization Landscape. In general, the com￾plexity of the sensor model and dynamics of the robot create a challenging optimization landscape in the CA-MAP problem in Eq. 8 t… view at source ↗
read the original abstract

Contact dynamics hold immense amounts of information that can improve a robot's ability to characterize and learn about objects in their environment through interactions. However, collecting information-rich contact data is challenging due to its inherent sparsity and non-smooth nature, requiring an active approach to maximize the utility of contacts for learning. In this work, we investigate an optimal experimental design approach to synthesize robot behaviors that produce contact-rich data for learning. Our approach derives a contact-aware Fisher information measure that characterizes information-rich contact behaviors that improve parameter learning. We observe emergent robot behaviors that are able to excite contact interactions that efficiently learns object parameters across a range of parameter learning examples. Last, we demonstrate the utility of contact-awareness for learning parameters through contact-seeking behaviors on several robotic experiments.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper proposes an optimal experimental design method to synthesize robot behaviors that maximize a contact-aware Fisher information measure, thereby generating information-rich contact data for improved object parameter learning. It derives this specialized Fisher information to characterize useful contact behaviors, reports emergent contact-seeking robot motions across parameter learning examples, and validates the approach via several robotic experiments.

Significance. If the central claims hold, the work could advance active learning and perception in contact-rich robotics by supplying a principled information-theoretic objective for behavior synthesis. Credit is due for the derivation of the contact-aware Fisher measure and for the observation of emergent behaviors that improve parameter estimation without hand-engineered rewards. This has clear relevance to manipulation tasks where contacts are the primary source of information.

major comments (2)
  1. [Method] Method section (contact-aware Fisher derivation): the approach relies on the underlying rigid-body contact model being sufficiently accurate for the computed Fisher information matrix to rank trajectories meaningfully. No sensitivity analysis to contact parameters, complementarity approximations, or smoothing schemes is provided, which is load-bearing because model mismatch can bias the information measure and produce behaviors that are informative only in simulation.
  2. [Experiments] Experiments section: the robotic demonstrations show utility of contact-seeking behaviors, yet no sim-to-real transfer results or quantitative assessment of how contact-model error propagates into the learned parameters are reported. This weakens the claim that maximization in simulation yields improved real-world parameter learning.
minor comments (1)
  1. [Abstract] Abstract: the clause 'that efficiently learns object parameters' contains a subject-verb agreement error ('learns' should be 'learn' to match the plural 'behaviors').

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the positive summary and for highlighting the potential impact of the contact-aware Fisher information approach. We address each major comment below in detail and commit to revisions that strengthen the manuscript's claims regarding model robustness and real-world applicability.

read point-by-point responses

  1. Referee: [Method] Method section (contact-aware Fisher derivation): the approach relies on the underlying rigid-body contact model being sufficiently accurate for the computed Fisher information matrix to rank trajectories meaningfully. No sensitivity analysis to contact parameters, complementarity approximations, or smoothing schemes is provided, which is load-bearing because model mismatch can bias the information measure and produce behaviors that are informative only in simulation.

    Authors: We agree that the accuracy of the rigid-body contact model is critical for the Fisher information matrix to produce meaningful trajectory rankings. The derivation employs standard complementarity-based contact models with common smoothing approximations drawn from the contact-rich robotics literature. To address the concern directly, we will add a sensitivity analysis to the revised manuscript. This will include systematic variation of key parameters (friction coefficients, contact stiffness, and smoothing factors) over realistic ranges, with results showing that the emergent contact-seeking behaviors and relative information gains remain consistent. We believe this addition will demonstrate that the measure is not overly sensitive to moderate model variations. revision: yes

  2. Referee: [Experiments] Experiments section: the robotic demonstrations show utility of contact-seeking behaviors, yet no sim-to-real transfer results or quantitative assessment of how contact-model error propagates into the learned parameters are reported. This weakens the claim that maximization in simulation yields improved real-world parameter learning.

    Authors: The referee is correct that the current experiments do not include an explicit quantitative study of contact-model error propagation or dedicated sim-to-real transfer metrics. Our robotic results execute the simulation-optimized behaviors on hardware and demonstrate improved parameter estimation using real contact data. In the revision we will expand the experiments section with a new analysis that perturbs contact parameters in simulation, quantifies the resulting change in learned object parameters, and discusses observed differences between simulation predictions and physical trials. This will provide a clearer assessment of how model mismatch affects real-world learning performance. revision: yes

Circularity Check

0 steps flagged

Derivation of contact-aware Fisher information remains self-contained with no reduction to fitted inputs or self-citations

full rationale

The paper presents a derivation of a contact-aware Fisher information measure grounded in standard optimal experimental design principles applied to hybrid contact dynamics. This measure is then maximized to synthesize behaviors, with validation through robotic experiments demonstrating improved parameter learning. No equations or steps in the abstract or described method reduce the central result to a self-definition, a fitted parameter renamed as prediction, or a load-bearing self-citation chain. The approach relies on the mathematical construction of the information matrix from the underlying dynamics model and empirical testing, making the derivation independent rather than circular by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no free parameters, axioms, or invented entities can be extracted or audited.

pith-pipeline@v0.9.0 · 5653 in / 919 out tokens · 42430 ms · 2026-05-22T15:16:54.292167+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Stein Variational Uncertainty-Adaptive Model Predictive Control

    cs.RO 2026-04 unverdicted novelty 7.0

    A new Stein variational controller for nonlinear systems with parametric uncertainty achieves better performance-robustness tradeoffs than worst-case or ensemble baselines by shaping control around task-dependent unce...

  2. Distributionally Robust Control via Stein Variational Inference for Contact-Rich Manipulation

    cs.RO 2026-05 unverdicted novelty 6.0

    Introduces a Stein variational inference-based deterministic formulation for distributionally robust control in contact-rich robotic manipulation, reporting up to 3x improved robustness under parametric uncertainty.

Reference graph

Works this paper leans on

55 extracted references · 55 canonical work pages · cited by 2 Pith papers

  1. [1]

    Formulating dy- namic multi-rigid-body contact problems with friction as solvable linear complementarity problems.Nonlinear Dynamics, 14, 03 1997

    Mihai Anitescu and Florian Potra. Formulating dy- namic multi-rigid-body contact problems with friction as solvable linear complementarity problems.Nonlinear Dynamics, 14, 03 1997. doi: 10.1023/A:1008292328909

    work page doi:10.1023/a:1008292328909 1997

  2. [2]

    Nikolay Atanasov, Jerome Le Ny, Kostas Daniilidis, and George J. Pappas. Information acquisition with sensing robots: Algorithms and error bounds, 2013

    work page 2013

  3. [3]

    Varadarajan, A

    Alp Aydinoglu and Michael Posa. Real-time multi- contact model predictive control via admm. In2022 International Conference on Robotics and Automation (ICRA), page 3414–3421. IEEE Press, 2022. doi: 10. 1109/ICRA46639.2022.9811957. URL https://doi.org/ 10.1109/ICRA46639.2022.9811957

    work page doi:10.1109/icra46639.2022.9811957 2022

  4. [4]

    Bellman and K.J

    R. Bellman and K.J. ˚Astr¨om. On structural identifiability. Mathematical Biosciences, 7(3):329–339, 1970. ISSN 0025-5564. doi: https://doi.org/10.1016/0025-5564(70) 90132-X. URL https://www.sciencedirect.com/science/ article/pii/002555647090132X

    work page doi:10.1016/0025-5564(70 1970

  5. [5]

    Applied and Numerical Harmonic Analysis

    Gregory S Chirikjian.Stochastic Models, Information Theory, and Lie Groups, Volume 1: Classical Results and Geometric Methods. Applied and Numerical Harmonic Analysis. Birkh ¨auser Boston, Boston, 1 edition, 2009. ISBN 081764802X

    work page 2009

  6. [6]

    Bradley Efron and David V . Hinkley. Assessing the accuracy of the maximum likelihood estimator: Observed versus expected fisher information.Biometrika, 65(3): 457–482, 1978. ISSN 00063444, 14643510. URL http://www.jstor.org/stable/2335893

    work page arXiv 1978

  7. [7]

    Optimal exper- iment design.Measurement Science and Technology, 9 (6):864, jun 1998

    A F Emery and Aleksey V Nenarokomov. Optimal exper- iment design.Measurement Science and Technology, 9 (6):864, jun 1998. doi: 10.1088/0957-0233/9/6/003. URL https://dx.doi.org/10.1088/0957-0233/9/6/003

    work page doi:10.1088/0957-0233/9/6/003 1998

  8. [8]

    Parameter and contact force estimation of planar rigid-bodies undergoing frictional contact.The International Journal of Robotics Research, 36(13-14): 1437–1454, 2017

    Nima Fazeli, Roman Kolbert, Russ Tedrake, and Alberto Rodriguez. Parameter and contact force estimation of planar rigid-bodies undergoing frictional contact.The International Journal of Robotics Research, 36(13-14): 1437–1454, 2017. doi: 10.1177/0278364917698749. URL https://doi.org/10.1177/0278364917698749

    work page doi:10.1177/0278364917698749 2017

  9. [9]

    Optimal experimental design.Wiley Interdisciplinary Reviews: Computational Statistics, 2(5): 581–589, 2010

    Valerii Fedorov. Optimal experimental design.Wiley Interdisciplinary Reviews: Computational Statistics, 2(5): 581–589, 2010

    work page 2010

  10. [10]

    Fenner and G

    T.I. Fenner and G. Loizou. Matrix bounds on the spectral condition number.Linear Algebra and its Applications, 8(2):157–178, 1974. ISSN 0024-3795. doi: https://doi.org/10.1016/0024-3795(74) 90053-6. URL https://www.sciencedirect.com/science/ article/pii/0024379574900536

    work page doi:10.1016/0024-3795(74 1974

  11. [11]

    Gautier and W

    M. Gautier and W. Khalil. On the identification of the inertial parameters of robots. InProceedings of the 27th IEEE Conference on Decision and Control, pages 2264– 2269 vol.3, 1988. doi: 10.1109/CDC.1988.194738

    work page doi:10.1109/cdc.1988.194738 1988

  12. [12]

    Maximum likelihood estimation is all you need for well-specified covariate shift, 2023

    Jiawei Ge, Shange Tang, Jianqing Fan, Cong Ma, and Chi Jin. Maximum likelihood estimation is all you need for well-specified covariate shift, 2023. URL https://arxiv. org/abs/2311.15961

    work page arXiv 2023

  13. [13]

    Predictive sampling: Real-time behaviour synthesis with mujoco,

    Taylor Howell, Nimrod Gileadi, Saran Tunyasuvunakool, Kevin Zakka, Tom Erez, and Yuval Tassa. Predictive sampling: Real-time behaviour synthesis with mujoco,

    work page

  14. [14]

    URL https://arxiv.org/abs/2212.00541

    work page arXiv

  15. [15]

    Learning linear complementarity systems

    Wanxin Jin, Alp Aydinoglu, Mathew Halm, and Michael Posa. Learning linear complementarity systems. In Roya Firoozi, Negar Mehr, Esen Yel, Rika Antonova, Jeannette Bohg, Mac Schwager, and Mykel Kochenderfer, editors, Proceedings of The 4th Annual Learning for Dynamics and Control Conference, volume 168 ofProceedings of Machine Learning Research, pages 1137...

    work page 2022

  16. [16]

    Littman, and Anthony R

    Leslie Pack Kaelbling, Michael L. Littman, and Anthony R. Cassandra. Planning and acting in partially observable stochastic domains.Artificial Intelligence, 101(1):99–134, 1998. ISSN 0004-

    work page 1998

  17. [17]

    URL https://www.sciencedirect.com/science/ article/pii/S000437029800023X

    doi: https://doi.org/10.1016/S0004-3702(98) 00023-X. URL https://www.sciencedirect.com/science/ article/pii/S000437029800023X

    work page doi:10.1016/s0004-3702(98

  18. [18]

    Khosla and Takeo Kanade

    Pradeep K. Khosla and Takeo Kanade. Parameter identification of robot dynamics. In1985 24th IEEE Conference on Decision and Control, pages 1754–1760,

    work page

  19. [19]

    doi: 10.1109/CDC.1985.268838

    work page doi:10.1109/cdc.1985.268838 1985

  20. [20]

    J. Kiefer. Optimum experimental designs. Journal of the Royal Statistical Society: Series B (Methodological), 21(2):272–304, 1959. doi: https://doi.org/10.1111/j.2517-6161.1959.tb00338.x. URL https://rss.onlinelibrary.wiley.com/doi/abs/10.1111/ j.2517-6161.1959.tb00338.x

    work page doi:10.1111/j.2517-6161.1959.tb00338.x 1959

  21. [21]

    Nathan J. Kong, J. Joe Payne, James Zhu, and Aaron M. Johnson. Saltation matrices: The essential tool for linearizing hybrid dynamical systems.Proceedings of the IEEE, 112(6):585–608, June 2024. ISSN 1558-

    work page 2024

  22. [22]

    URL http: //dx.doi.org/10.1109/JPROC.2024.3440211

    doi: 10.1109/jproc.2024.3440211. URL http: //dx.doi.org/10.1109/JPROC.2024.3440211

    work page doi:10.1109/jproc.2024.3440211 2024

  23. [23]

    Planning for robotic exploration based on forward simulation.Robotics and Autonomous Systems, 83:15–31, September 2016

    Mikko Lauri and Risto Ritala. Planning for robotic exploration based on forward simulation.Robotics and Autonomous Systems, 83:15–31, September 2016. ISSN 0921-8890. doi: 10.1016/j.robot.2016.06.008. URL http://dx.doi.org/10.1016/j.robot.2016.06.008

    work page doi:10.1016/j.robot.2016.06.008 2016

  24. [24]

    Lederman and Roberta L

    Susan J. Lederman and Roberta L. Klatzky. Extracting object properties through haptic exploration.Acta Psychologica, 84(1):29–40, 1993. ISSN 0001-6918. doi: https://doi.org/10.1016/0001-6918(93)90070-8. URL https://www.sciencedirect.com/science/article/pii/ 0001691893900708. Tactile Pattern Recognition

    work page doi:10.1016/0001-6918(93)90070-8 1993

  25. [25]

    Wensing, and Frank C

    Taeyoon Lee, Jaewoon Kwon, Patrick M. Wensing, and Frank C. Park. Robot model identification and learning: A modern perspective.Annual Review of Control, Robotics, and Autonomous Systems, 2023. doi: https://doi.org/10.1146/annurev-control-061523-102310. URL https://www.annualreviews.org/content/journals/10. 1146/annurev-control-061523-102310

    work page doi:10.1146/annurev-control-061523-102310 2023

  26. [26]

    Identifiability of physical parame- ters in systems with limited sensors.IFAC Proceed- ings Volumes, 47(3):6454–6459, 2014

    Jonas Linder, Martin Enqvist, Fredrik Gustafsson, and Johan Sj ¨oberg. Identifiability of physical parame- ters in systems with limited sensors.IFAC Proceed- ings Volumes, 47(3):6454–6459, 2014. ISSN 1474-

    work page 2014

  27. [27]

    doi: https://doi.org/10.3182/20140824-6-ZA-1003. 02272. URL https://www.sciencedirect.com/science/ article/pii/S147466701642625X. 19th IFAC World Congress

    work page doi:10.3182/20140824-6-za-1003

  28. [28]

    Prentice Hall PTR, USA, 1999

    Lennart Ljung.System identification (2nd ed.): theory for the user. Prentice Hall PTR, USA, 1999. ISBN 0136566952

    work page 1999

  29. [29]

    Variational contact-implicit trajectory optimization

    Zac Manchester and Scott Kuindersma. Variational contact-implicit trajectory optimization. InProceedings of International Symposium on Robotics Research (ISRR ’17), pages 985 – 1000, December 2017

    work page 2017

  30. [30]

    Asid: Active exploration for system identification in robotic manipulation,

    Marius Memmel, Andrew Wagenmaker, Chuning Zhu, Patrick Yin, Dieter Fox, and Abhishek Gupta. Asid: Ac- tive exploration for system identification in robotic ma- nipulation, 2024. URL https://arxiv.org/abs/2404.12308

    work page arXiv 2024

  31. [31]

    Discovery of complex behaviors through contact- invariant optimization.ACM Trans

    Igor Mordatch, Emanuel Todorov, and Zoran Popovi ´c. Discovery of complex behaviors through contact- invariant optimization.ACM Trans. Graph., 31(4), July

    work page

  32. [32]

    doi: 10.1145/2185520.2185539

    ISSN 0730-0301. doi: 10.1145/2185520.2185539. URL https://doi.org/10.1145/2185520.2185539

    work page doi:10.1145/2185520.2185539

  33. [33]

    Dynamic parameters identification of a humanoid robot using joint torque sensors and/or contact forces

    Yusuke Ogawa, Gentiane Venture, and Christian Ott. Dynamic parameters identification of a humanoid robot using joint torque sensors and/or contact forces. 2015: 457–462, 02 2015. doi: 10.1109/HUMANOIDS.2014. 7041401

    work page doi:10.1109/humanoids.2014 2015

  34. [34]

    Fun- damental challenges in deep learning for stiff contact dynamics, 2021

    Mihir Parmar, Mathew Halm, and Michael Posa. Fun- damental challenges in deep learning for stiff contact dynamics, 2021

    work page 2021

  35. [35]

    Bayesian inference for parameters estima- tion using experimental data.Probabilistic Engi- neering Mechanics, 60:103025, 2020

    Chiara Pepi, Massimiliano Gioffr `e, and Mircea Grig- oriu. Bayesian inference for parameters estima- tion using experimental data.Probabilistic Engi- neering Mechanics, 60:103025, 2020. ISSN 0266-

    work page 2020

  36. [36]

    doi: https://doi.org/10.1016/j.probengmech.2020. 103025. URL https://www.sciencedirect.com/science/ article/pii/S0266892020300102

    work page doi:10.1016/j.probengmech.2020 2020

  37. [37]

    Contactnets: Learning of discontinuous contact dynamics with smooth, implicit representations

    Samuel Pfrommer, Mathew Halm, and Michael Posa. Contactnets: Learning of discontinuous contact dynamics with smooth, implicit representations. InConference on Robot Learning, 2020. URL https://api.semanticscholar. org/CorpusID:221857185

    work page 2020

  38. [38]

    Belief space planning assuming maximum likelihood observations

    Robert Platt, Russ Tedrake, Leslie Pack Kaelbling, and Tomas Lozano-Perez. Belief space planning assuming maximum likelihood observations. InRobotics: Science and Systems, 2010. URL https://api.semanticscholar.org/ CorpusID:2693863

    work page 2010

  39. [39]

    A direct method for trajectory optimization of rigid bodies through contact.The International Journal of Robotics Research, 33(1):69–81, 2014

    Michael Posa, Cecilia Cantu, and Russ Tedrake. A direct method for trajectory optimization of rigid bodies through contact.The International Journal of Robotics Research, 33(1):69–81, 2014. doi: 10. 1177/0278364913506757. URL https://doi.org/10.1177/ 0278364913506757

    work page 2014

  40. [40]

    Radhakrishna Rao

    C. Radhakrishna Rao. Minimum variance and the esti- mation of several parameters.Mathematical Proceedings of the Cambridge Philosophical Society, 43(2):280–283,

    work page

  41. [41]

    doi: 10.1017/S0305004100023471

    work page doi:10.1017/s0305004100023471

  42. [42]

    Bridging the sim-to-real gap with bayesian inference, 2024

    Jonas Rothfuss, Bhavya Sukhija, Lenart Treven, Florian D¨orfler, Stelian Coros, and Andreas Krause. Bridging the sim-to-real gap with bayesian inference, 2024. URL https://arxiv.org/abs/2403.16644

    work page arXiv 2024

  43. [43]

    Dart: Dense articulated real-time tracking

    Tanner Schmidt and Dieter Fox. Dart: Dense articulated real-time tracking

    work page

  44. [44]

    On the variance of the fisher information for deep learning, 2021

    Alexander Soen and Ke Sun. On the variance of the fisher information for deep learning, 2021. URL https: //arxiv.org/abs/2107.04205

    work page arXiv 2021

  45. [45]

    Stewart and Jeffrey C

    David E. Stewart and Jeffrey C. Trinkle. An im- plicit time-stepping scheme for rigid body dynamics with coulomb friction.Proceedings 2000 ICRA. Mil- lennium Conference. IEEE International Conference on Robotics and Automation. Symposia Proceedings (Cat. No.00CH37065), 1:162–169 vol.1, 2000. URL https: //api.semanticscholar.org/CorpusID:10920958

    work page 2000

  46. [46]

    Relative Fisher information and natural gradient for learning large modular models

    Ke Sun and Frank Nielsen. Relative Fisher information and natural gradient for learning large modular models. In Doina Precup and Yee Whye Teh, editors,Proceedings of the 34th International Conference on Machine Learn- ing, volume 70 ofProceedings of Machine Learning Research, pages 3289–3298. PMLR, 06–11 Aug 2017. URL https://proceedings.mlr.press/v70/su...

    work page 2017

  47. [47]

    In-hand object-dynamics inference using tactile fingertips, 2021

    Balakumar Sundaralingam and Tucker Hermans. In-hand object-dynamics inference using tactile fingertips, 2021. URL https://arxiv.org/abs/2003.13165

    work page arXiv 2021

  48. [48]

    Swevers, C

    J. Swevers, C. Ganseman, D.B. Tukel, J. de Schutter, and H. Van Brussel. Optimal robot excitation and identifica- tion.IEEE Transactions on Robotics and Automation, 13 (5):730–740, 1997. doi: 10.1109/70.631234

    work page doi:10.1109/70.631234 1997

  49. [49]

    Tassa, T

    Yuval Tassa, Tom Erez, and Emanuel Todorov. Synthesis and stabilization of complex behaviors through online trajectory optimization. In2012 IEEE/RSJ International Conference on Intelligent Robots and Systems, pages 4906–4913, 2012. doi: 10.1109/IROS.2012.6386025

    work page doi:10.1109/iros.2012.6386025 2012

  50. [50]

    Mujoco: A physics en- gine for model-based control, in: 2012 IEEE/RSJ International Con- ference on Intelligent Robots and Systems, IEEE

    Emanuel Todorov, Tom Erez, and Yuval Tassa. Mu- joco: A physics engine for model-based control. In 2012 IEEE/RSJ International Conference on Intelligent Robots and Systems, pages 5026–5033. IEEE, 2012. doi: 10.1109/IROS.2012.6386109

    work page doi:10.1109/iros.2012.6386109 2012

  51. [51]

    Learning active tactile perception through belief-space control, 2023

    Jean-Franc ¸ois Tremblay, David Meger, Francois Hogan, and Gregory Dudek. Learning active tactile perception through belief-space control, 2023. URL https://arxiv. org/abs/2312.00215

    work page arXiv 2023

  52. [52]

    Wilson, Jarvis A

    Andrew D. Wilson, Jarvis A. Schultz, and Todd D. Murphey. Trajectory synthesis for fisher information maximization.IEEE Transactions on Robotics, 30(6): 1358–1370, December 2014. ISSN 1941-0468. doi: 10.1109/tro.2014.2345918. URL http://dx.doi.org/10. 1109/TRO.2014.2345918

    work page doi:10.1109/tro.2014.2345918 2014

  53. [53]

    Li, 3d fully convolutional network for vehicle de- tection in point cloud, in: 2017 IEEE/RSJ Interna- tional Conference on Intelligent Robots and Systems (IROS), 2017, pp

    Andrew D. Wilson, Jarvis A. Schultz, Alex R. Ansari, and Todd D. Murphey. Real-time trajectory synthesis for information maximization using sequential action control and least-squares estimation. In2015 IEEE/RSJ Inter- national Conference on Intelligent Robots and Systems (IROS), pages 4935–4940, 2015. doi: 10.1109/IROS. 2015.7354071

    work page doi:10.1109/iros 2015

  54. [54]

    Impact-invariant con- trol: Maximizing control authority during impacts, 2023

    William Yang and Michael Posa. Impact-invariant con- trol: Maximizing control authority during impacts, 2023

    work page 2023

  55. [55]

    Belief space planning: A covariance steering approach, 2021

    Dongliang Zheng, Jack Ridderhof, Panagiotis Tsiotras, and Ali akbar Agha-mohammadi. Belief space planning: A covariance steering approach, 2021. URL https://arxiv. org/abs/2105.11092. APPENDIXA PROOFS Fisher Information Steepest Ascent Proof Proof.We first approximateL(D|τ, θ+δθ)in Eq.11 by a second order Taylor expansion, L(D|τ, θ+δθ)≈ L(D|τ, θ) +∇ θL(D|...

    work page arXiv 2021