Pith. sign in

REVIEW 5 major objections 6 minor 48 references

D4W: Dependable Data-Driven Dynamics for Wheeled Robots

T0 review · 5 major / 6 minor · reviewed 2026-08-12 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper claims that a learned hybrid model predicts long wheeled-robot trajectories more accurately than analytical physics simulation, with centimeter-level error at 32,768 steps.

desk verdict A useful hybrid dynamics framework with a real long-horizon gain, but the headline accuracy claim is oversold because the paper never runs its own parameter-fitting analytical baseline. read the letter →

arxiv 2411.09360 v1 pith:Z5SKO37E submitted 2024-11-14 cs.RO

Yunfeng Lin , Minghuan Liu , Yong Yu This is my paper · ORCID
classification cs.RO
keywords data-drivendynamicswheeledmobilerobotsphysicalsimulationegocentrictransformationdifferentiablesim-to-realtransfer
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper proposes D4W, a framework for learning wheeled-robot dynamics from real-world motion-capture recordings. It claims that a differentiable hybrid model—an analytical kinematic model whose inputs are corrected by a small neural network—predicts long trajectories more accurately than hand-formulated analytical models or physics-based simulators. In evaluation, the hybrid model trained on length-4096 trajectories reaches a step-average position error of about 72 mm over 32,768 simulated steps, compared with about 133 mm for the hand-formulated model and about 1,208 mm for the physics simulator. The point, if the claim holds, is that control and navigation algorithms can be developed in simulation with little or no real-world fine-tuning.

What carries the argument

The load-bearing mechanism is the egocentric transformation: before each prediction step, the model's input poses and commands are re-expressed relative to the robot's current pose and current time, so the learning target is a map that is invariant under translation, rotation, and time shift. This is combined with a differentiable hybrid model (an analytical kinematic model whose outputs feed a small MLP, or an MLP that predicts a correction to the analytical input), and progressive training in which the optimizer sees trajectories of exponentially growing length. The MLP in the best configuration is a three-layer network with hidden sizes (32, 16, 8) and batch normalization.

What would settle it

Evaluate the model on hold-out trajectories that contain abrupt speed-command steps (square waves) and compare per-step RMSE immediately after each step against steady segments; if the error spikes at the command edges, the memoryless input has failed to capture the actuator transient.

Watch

Extended reading notes

Core claim

The central discovery is that a dynamics model learned from recorded poses and command windows can outperform analytical physics simulation for long-horizon prediction of wheeled robots, provided the model is trained in an egocentric frame and on progressively longer trajectories. The paper reports that the best configuration, Formulated + MLP trained on length-4096 trajectories, achieves an RMSE of 72.102 mm per step at a test trajectory length of 32,768, against 133.37 mm for the hand-formulated analytical model and 1,207.9 mm for the physics-based simulator. The authors attribute the generalization to three components: the egocentric transformation that enforces translational, rotational, and time-translational symmetry; the differentiable hybrid architecture that keeps the analytical model in the loop while learning residual corrections; and progressive training on exponentially increasing trajectory lengths.

Load-bearing premise

The input to the model is only the current pose and a 200-millisecond window of commands, so the premise is that hidden internal states—wheel speeds, caster orientations, and actuator transients—are either irrelevant or recoverable from pose and command history.

Editorial extensions

If this is right

  • Long-horizon simulation becomes practical: the learned model predicts tens of thousands of steps with errors in the centimeter range, whereas non-learned baselines drift to meter-level errors.
  • Controllers tuned in D4W should transfer to the real robot with little or no fine-tuning, cutting the cost of physical experiments.
  • The egocentric transformation provides a template for enforcing physical symmetries in learned dynamics for other robots.
  • The framework can be integrated with existing simulators and controllers, letting teams swap in learned dynamics without rebuilding their pipeline.
  • Automated data collection (random commands, collision avoidance, and restart) makes it possible to gather training data without manual teleoperation.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The symmetry assumptions in the egocentric frame rule out position-dependent effects such as floor-friction patches or arena boundaries; a natural extension is to add positional or environmental features and test whether the model generalizes to non-uniform floors.
  • With history length H=1, the model sees no wheel-speed or actuator state, so it is likely to struggle during command transients; a testable extension is to include a short history of measured or commanded actuator signals and compare short-horizon error at command edges.
  • The reported gains come from a single robot instance and floor; an open question is whether a model trained on one unit transfers to another unit of the same model with different tire wear, which could be measured by training on one robot and evaluating on the other.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 6 minor

Summary. This paper presents D4W, a framework for learning the dynamics of unicycle-type wheeled robots from recorded motion-capture poses and speed commands. It describes an unattended data collection procedure, four model architectures (pure data-driven, dynamical hybrid, kinematic hybrid, and analytical parameter model), an egocentric coordinate transformation to enforce translation/rotation/time-translation symmetries, and progressive training on increasing trajectory lengths. The main evaluation compares learned models with a hand-formulated PyTorch analytical model and an Isaac Sim simulation on step-average RMSE at trajectory lengths from 1 to 32,768; the best configuration (Formulated+MLP trained on 4096-step trajectories) achieves 72.102 mm at 32,768 steps versus 133.37 mm for the hand-formulated model. The paper claims that this result supports best-in-class simulation accuracy and reduced real-world fine-tuning.

Significance. If the reported results are robust, the paper makes a useful contribution to data-driven robot simulation: it shows that a real-world recorded dataset, combined with a differentiable analytical prior and a symmetry-preserving input transformation, can produce a long-horizon trajectory predictor that improves on a fixed analytical baseline. The unattended data collection procedure and progressive training schedule are pragmatic. The main weakness is that the central comparison is against an uncalibrated, unspecified analytical baseline and omits the paper's own fitted analytical model, so the strength of the claim exceeds the current evidence. The within-distribution, single-arena evaluation further limits the 'no fine-tuning' conclusion.

major comments (5)
  1. [§4.2.1 and Eq. (11)] The comparison that supports the headline claim, Table 2, is not against a calibrated analytical baseline. The Hand-Formulated model is described only as a 'handmade formula written in PyTorch' (Sec. 4.2.1), with no equations or parameter values, and Isaac Sim uses manually chosen contact parameters (Appendix B). The 'Analytical Parameter Model' defined in Eq. (11), which would fit the analytical model's configuration parameters to data, is never implemented or reported. A fitted analytical model could close the 72.102 mm vs 133.37 mm gap at trajectory length 32,768. Without this comparison, the abstract's claim that D4W 'achieves the best simulation accuracy compared to traditional approaches' is not established.
  2. [Table 2] The central result is horizon-dependent. At trajectory length 64, the Hand-Formulated baseline (3.6603 mm) beats the best learned model (3.8150 mm for Formulated+MLP trained on length 4096); at lengths 1 and 8, the learned and hand-formulated models are essentially tied. The abstract and conclusion state the superiority claim without qualification. Since short-horizon prediction is often the relevant regime for controller evaluation, the paper should either qualify the claim by horizon or show that the crossover is systematic and explain why long-horizon error is the right criterion.
  3. [§4.1, Eq. (6), and §2] With H=1 and a 200 ms command window (Sec. 4.1), the model is effectively a memoryless map from the current pose and recent commands to the next pose. Section 2 explicitly omits wheel speeds, caster-wheel orientations, and actuator states. The evaluation does not isolate command transients, where the wheel-speed controller is still ramping, or contact transients such as a sliding caster. Figure 10 shows large discrepancies between commanded and measured speeds, so these unobserved states are not negligible in general. The claim that D4W requires 'less or no need for fine-tuning in reality' is therefore not supported outside the specific data collection distribution.
  4. [§4.4.1, Table 3, and Table 2] The ablation results in Table 3 appear inconsistent with Table 2. The Egocentric row of Table 3 should correspond to the same single linear-layer model used for the 'LR' rows in Table 2 with egocentric transformation, but the RMSE values differ substantially (for example, 1061.6 mm vs 2315.5 mm at trajectory length 32,768). The discrepancy is not explained. Since Table 3 is the quantitative evidence for the egocentric transformation, the paper must clarify the training setup or correct the numbers.
  5. [§6 and Abstract] The Conclusion states that 'Real-world validation experiments confirm the applicability and robustness of the learned dynamics models from D4W,' and the abstract claims verification 'through integration with existing simulators and controllers.' No such experiments or integration results are reported in the body. The only evaluation is open-loop trajectory prediction on a held-out portion of the same recordings. Either add the missing experiments or remove/soften these claims.
minor comments (6)
  1. [Eq. (13)] The offset update Δq_{t+1} = Δq_t + R·r_{t-H} is not defined for the first steps when H equals the history length, and the indexing appears inconsistent with the definition of Q_i in Eq. (6). Please clarify the intended indexing.
  2. [Eqs. (20)–(21)] The symbol α is overloaded: Eq. (20) uses α as a distance exponent, while Eq. (21) uses α as a weighting coefficient between the two Chamfer terms. Please state exactly which loss is implemented and how α is chosen in the experiments.
  3. [§4.3.1] The statement that 'recurrent networks have been reported to have poor performance [32]' is not well supported by reference [32], which concerns policy learning for legged locomotion rather than recurrent trajectory models. Please provide a more relevant citation or justification.
  4. [Table 2] The 'Trajectory lengths (Training)' column is ambiguous for the analytical baselines, which are listed with training length 0; it is not stated whether these baselines use the same command/pose windows and evaluation protocol as the learned models.
  5. [§3.3] The egocentric transformation assumes translational, rotational, and time-translational symmetry. The paper should explicitly state that position-dependent effects such as floor friction variation or arena-boundary contact are outside the model's scope, since the recorded data do not test those cases.
  6. [Abstract and §4] The paper uses the word 'dependable' in the title and abstract, but no confidence intervals, per-trajectory variances, or failure cases are reported. Reporting error bars across test trajectories would substantially strengthen the claim.

Circularity Check

0 steps flagged · score 2.0 of 10

No significant circularity; the central prediction is a held-out supervised learning evaluation, and the only self-citations are peripheral.

full rationale

D4W's derivation chain is a standard supervised learning pipeline: recorded command/pose data (Eq. 4) feed a model F (Eq. 6), predictions are scored against ground truth with the loss in Eq. 7, and Table 2 reports errors on a 30% held-out portion of the same sessions (Sec. 4.1). The data-driven models do not receive test labels as inputs, and the analytical baselines are fixed formulas/simulators rather than fits to the test set, so the accuracy numbers are genuine within-distribution predictions. The 'Analytical Parameter Model' of Eq. 11 is defined but never implemented or reported; that is an omitted comparison and a weakness of the headline claim, not a circular reduction. The symmetry argument cites Noether's theorems [19], an external mathematical result, and the egocentric transform (Eqs. 12-18) is a change of coordinates, not a definition of the target in terms of the model. The two self-citations ([20], [23]) appear only in related work and hyperparameter-tuning remarks and are not load-bearing for the central claim. Accordingly, no circular step can be quoted, and the paper merits a low circularity score.

Assumptions & free parameters 6 free parameters · 6 assumptions · 0 invented entities

The central claim rests on the dataset (one robot, one arena, motion capture ground truth), the Markovian/observability assumption in Eq. 6, and the symmetry assumptions of Sec. 3.3. The fitted model parameters are the main free parameters, plus several hand-chosen training choices (Chamfer alpha, progressive schedule, Isaac Sim contact parameters). No invented physical entities are postulated; the learned correction in the hybrid model is a residual, not a new physical quantity with an independent falsifiable handle.

free parameters (6)
  • Learned model parameters (LR/MLP/hybrid networks) = not reported (weights)
    The central artifact: all dynamics parameters are fit to 2000 s of this robot's motion capture data (Sec. 4.1), so the learned dynamics are robot-specific and arena-specific.
  • Chamfer loss weight alpha = 0.2, 0.5, 0.8 evaluated (Fig. 7)
    Hand-selected trade-off between trajectory-level and step-level loss terms (Sec. 3.4.1); the final choice used for Table 2 is not stated.
  • Progressive training schedule = lengths 1, 64, 4096 reported; 'increased exponentially'
    The sequence-length schedule (Sec. 3.4.2) strongly affects the results, and the exact schedule is not specified.
  • History and input window sizes = H=1, T=200 ms (Sec. 4.1)
    Chosen for constant input dimensions (Eq. 6); the observability of the dynamics depends on these values.
  • Isaac Sim contact parameters = static friction 0.5, dynamic friction 4.0, restitution 0.0, bounce 2.0 (Table 4)
    Hand-chosen 'to improve numerical stability and reduce prediction errors'; these tune the analytical baseline used for comparison.
  • Hand-formulated baseline constants = not stated
    The 'handmade formula' in PyTorch (Sec. 4.2.1) presumably uses design-schema mass, inertia, and geometry; whether any constant was tuned is not disclosed, and no parameter-identification baseline is run.
assumptions (6)
  • domain assumption Unicycle pure-rolling kinematics (Eqs. 2-3)
    Used as the analytical core of the hand-formulated baseline and hybrid models; the paper acknowledges sliding violates it (Sec. 2), which is the stated motivation for learned corrections.
  • domain assumption The next pose is determined by the last H poses and the command window (Eq. 6)
    With H=1, wheel speeds, accelerations, caster orientations, and actuator states are omitted (Sec. 2); the model must infer them from pose and command history.
  • domain assumption Dynamics are invariant under translation, rotation, and time translation (Sec. 3.3)
    The egocentric transformation assumes a homogeneous floor and position-independent behavior; arena boundaries, floor friction variation, or robot damage would break it.
  • domain assumption Command and pose timestamps are synchronized and run at equal speeds (Sec. 3.1)
    Mislabelled command-pose pairs would corrupt the learned input-output map; the motion capture and command logger are assumed aligned.
  • domain assumption Motion capture error is millimeter-level and ignorable as ground truth (Sec. 4.1)
    The paper notes 'significant noise in the data' (Fig. 6) yet treats mocap poses as targets; no noise model or uncertainty propagation is used.
  • standard math Standard chain rule and backpropagation through time (Eqs. 16-19)
    Gradient-based training assumes the derived gradient is correct; the derivation as printed is sloppy (duplicate terms, implicit R dependence).

how reviews work

0 comments
Cite this review

Pith. "Pith review of D4W: Dependable Data-Driven Dynamics for Wheeled Robots." pith.science (2026). https://pith.science/paper/Z5SKO37E

@misc{pith2026241109360,
  author       = {Pith},
  title        = {Pith review of: D4W: Dependable Data-Driven Dynamics for Wheeled Robots},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Z5SKO37E}},
  note         = {Machine review of arXiv:2411.09360}
}
read the original abstract

Wheeled robots have gained significant attention due to their wide range of applications in manufacturing, logistics, and service industries. However, due to the difficulty of building a highly accurate dynamics model for wheeled robots, developing and testing control algorithms for them remains challenging and time-consuming, requiring extensive physical experimentation. To address this problem, we propose D4W, i.e., Dependable Data-Driven Dynamics for Wheeled Robots, a simulation framework incorporating data-driven methods to accelerate the development and evaluation of algorithms for wheeled robots. The key contribution of D4W is a solution that utilizes real-world sensor data to learn accurate models of robot dynamics. The learned dynamics can capture complex robot behaviors and interactions with the environment throughout simulations, surpassing the limitations of analytical methods, which only work in simplified scenarios. Experimental results show that D4W achieves the best simulation accuracy compared to traditional approaches, allowing for rapid iteration of wheel robot algorithms with less or no need for fine-tuning in reality. We further verify the usability and practicality of the proposed framework through integration with existing simulators and controllers.

Figures

Figures reproduced from arXiv: 2411.09360 by the authors.

Figure 1
Figure 1. Diagram of a unicycle-type wheeled robot. It shows [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Overview of the data-driven simulation pipeline in D4W. Two framework phases are shown: 1) data gathering: [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Possible architectures of the data-driven model. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Training trajectories under various transformations. Egocentric transformation yields an easier training target. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Robot commands gathered in data collection. The [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Comparison between a ground truth trajectory and one predicted by the handmade formula (Blue: ground truth; [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 7
Figure 7. Figure 7: Training progress (MSE error) with various loss [PITH_FULL_IMAGE:figures/full_fig_p008_7.png]
Figure 8
Figure 8. Figure 8: Training progress under various transforms. The [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: Part of the raw pose data recorded from the robot [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]
Figure 10
Figure 10. Figure 10: Part of the robot command and proprioceptive sen [PITH_FULL_IMAGE:figures/full_fig_p010_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

48 extracted references · 36 canonical work pages

  1. [15]

    Isenberg

    Douglas R. Isenberg. 2023. System Identification of a Mobile Robot with Motion Capture Data. 2023 Intermountain Engineering, Technology and Computing (IETC) (2023), 132–137. https://api.semanticscholar.org/CorpusID:259215958

    work page 2023

  2. [1]

    Battaglia, Razvan Pascanu, Matthew Lai, Danilo Jimenez Rezende, and Koray Kavukcuoglu

    Peter W. Battaglia, Razvan Pascanu, Matthew Lai, Danilo Jimenez Rezende, and Koray Kavukcuoglu. 2016. Interaction Networks for Learning about Objects, Re- lations and Physics. In NIPS. https://api.semanticscholar.org/CorpusID:2200675

    work page 2016

  3. [2]

    James Bradbury, Roy Frostig, Peter Hawkins, Matthew James Johnson, Chris Leary, Dougal Maclaurin, George Necula, Adam Paszke, Jake VanderPlas, Skye Wanderman-Milne, and Qiao Zhang. 2018. JAX: composable transformations of Python+NumPy programs. http://github.com/google/jax

    2018

  4. [3]

    Michael B Chang, Tomer Ullman, Antonio Torralba, and Joshua B Tenenbaum

    work page

  5. [4]

    Siwei Chen, Xiao Ma, and Zhongwen Xu. 2022. Imitation Learning via Differen- tiable Physics. arXiv:2206.04873 [cs.LG]

    work page Pith review arXiv 2022

  6. [5]

    Siwei Chen, Yiqing Xu, Cunjun Yu, Linfeng Li, Xiao Ma, Zhongwen Xu, and David Hsu. 2023. DaxBench: Benchmarking Deformable Object Manipulation with Differentiable Physics. In The Eleventh International Conference on Learning Representations. https://openreview.net/forum?id=1NAzMofMnWl

    work page 2023

  7. [6]

    Erwin Coumans and Yunfei Bai. 2016. Pybullet, a python module for physics simulation for games, robotics and machine learning. (2016)

    work page 2016

  8. [7]

    Filipe de Avila Belbute-Peres, Kevin Smith, Kelsey Allen, Josh Tenenbaum, and J Zico Kolter. 2018. End-to-end differentiable physics for learning and control. Advances in neural information processing systems 31 (2018)

    work page 2018

Show all 48 references
  1. [8]

    Jonas Degrave, Michiel Hermans, Joni Dambre, et al . 2019. A differentiable physics engine for deep learning in robotics. Frontiers in neurorobotics (2019), 6

    2019

  2. [9]

    Tom Erez, Yuval Tassa, and Emanuel Todorov. 2015. Simulation tools for model- based robotics: Comparison of bullet, havok, mujoco, ode and physx. In2015 IEEE international conference on robotics and automation (ICRA) . IEEE, 4397–4404

    2015

  3. [10]

    Daniel Freeman, Erik Frey, Anton Raichuk, Sertan Girgin, Igor Mordatch, and Olivier Bachem

    C. Daniel Freeman, Erik Frey, Anton Raichuk, Sertan Girgin, Igor Mordatch, and Olivier Bachem. 2021. Brax - A Differentiable Physics Engine for Large Scale Rigid Body Simulation. ArXiv abs/2106.13281 (2021)

    2021 arXiv

  4. [11]

    Tim Head, MechCoder, Gilles Louppe, Iaroslav Shcherbatyi, fcharras, Zé Vinícius, cmmalone, Christopher Schröder, nel215, Nuno Campos, Todd Young, Stefano Cereda, Thomas Fan, rene rex, Kejia (KJ) Shi, Justus Schwabedal, carlosdanielc- santos, Hvass-Labs, Mikhail Pak, SoManyUser...

    2018

  5. [12]

    Narang, Dieter Fox, Animesh Garg, and Fabio Ramos

    Eric Heiden, Miles Macklin, Yashraj S. Narang, Dieter Fox, Animesh Garg, and Fabio Ramos. 2021. DiSECt: A Differentiable Simulation Engine for Autonomous Robotic Cutting. ArXiv abs/2105.12244 (2021). https://api.semanticscholar.org/ CorpusID:235195762

    2021 arXiv

  6. [13]

    Tenenbaum, William T

    Yuanming Hu, Jiancheng Liu, Andrew Everett Spielberg, Joshua B. Tenenbaum, William T. Freeman, Jiajun Wu, Daniela Rus, and Wojciech Matusik. 2018. Chain- Queen: A Real-Time Differentiable Physical Simulator for Soft Robotics. 2019 International Conference on Robotics and Autom...

    2018

  7. [14]

    Sergey Ioffe and Christian Szegedy. 2015. Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift. ArXiv abs/1502.03167 (2015)

    2015 arXiv

  8. [16]

    Karen Liu

    Yifeng Jiang and C. Karen Liu. 2018. Data-Augmented Contact Model for Rigid Body Simulation. In Conference on Learning for Dynamics & Control . https: //api.semanticscholar.org/CorpusID:3881096

    2018

  9. [17]

    Muhammad Aseer Khan, Dur-e-Zehra Baig, Bilal Ashraf, Husan Ali, Junaid Rashid, and Jungeun Kim. 2022. Dynamic modeling of a nonlinear two-wheeled robot using data-driven approach. Processes 10, 3 (2022), 524

    2022

  10. [18]

    Koenig and Andrew Howard

    Nathan P. Koenig and Andrew Howard. 2004. Design and use paradigms for Gazebo, an open-source multi-robot simulator. 2004 IEEE/RSJ International Con- ference on Intelligent Robots and Systems (IROS) (IEEE Cat. No.04CH37566) 3 (2004), 2149–2154. https://api.semanticscholar.org/...

    2004

  11. [19]

    Yvette Kosmann-Schwarzbach, Bertram E Schwarzbach, and Yvette Kosmann- Schwarzbach. 2011. The Noether Theorems. Springer

    2011

  12. [20]

    Hang Lai, Jian Shen, Weinan Zhang, and Yong Yu. 2020. Bidirectional Model-based Policy Optimization. ArXiv abs/2007.01995 (2020). https://api.semanticscholar. org/CorpusID:220364497

    2020 arXiv

  13. [21]

    Yifei Li, Tao Du, Kui Wu, Jie Xu, and Wojciech Matusik. 2021. DiffCloth: Differen- tiable Cloth Simulation with Dry Frictional Contact. ACM Transactions on Graph- ics (TOG) 42 (2021), 1 – 20. https://api.semanticscholar.org/CorpusID:235390650

    2021

  14. [22]

    Tenenbaum, and Antonio Torralba

    Yunzhu Li, Jiajun Wu, Russ Tedrake, Joshua B. Tenenbaum, and Antonio Torralba

  15. [23]

    Yunfeng Lin, Guilin Li, Xingzhi Zhang, Weinan Zhang, Bo Chen, Bo Chen, Ruim- ing Tang, Zhenguo Li, Jiashi Feng, and Yong Yu. 2021. ModularNAS: Towards Modularized and Reusable Neural Architecture Search. In Conference on Machine Learning and Systems

    2021

  16. [24]

    Rudin, Arthur Allshire, Ankur Handa, and Gavriel State

    Viktor Makoviychuk, Lukasz Wawrzyniak, Yunrong Guo, Michelle Lu, Kier Storey, Miles Macklin, David Hoeller, N. Rudin, Arthur Allshire, Ankur Handa, and Gavriel State. 2021. Isaac Gym: High Performance GPU-Based Physics Simulation For Robot Learning. ArXiv abs/2108.10470 (2021)

    2021 arXiv

  17. [25]

    Mayank Mittal, Calvin Yu, Qinxi Yu, Jingzhou Liu, Nikita Rudin, David Hoeller, Jia Lin Yuan, Ritvik Singh, Yunrong Guo, Hammad Mazhar, et al. 2023. Orbit: A unified simulation framework for interactive robot learning environments. IEEE Robotics and Automation Letters (2023)

    2023

  18. [26]

    Damian Mrowca, Chengxu Zhuang, Elias Wang, Nick Haber, Li F Fei-Fei, Josh Tenenbaum, and Daniel L Yamins. 2018. Flexible neural representation for physics prediction. Advances in neural information processing systems 31 (2018)

    2018

  19. [27]

    J Krishna Murthy, Miles Macklin, Florian Golemo, Vikram Voleti, Linda Petrini, Martin Weiss, Breandan Considine, Jérôme Parent-Lévesque, Kevin Xie, Kenny Erleben, et al. 2020. gradsim: Differentiable simulation for system identification and visuomotor control. In International...

    2020

  20. [28]

    Thulasi Mylvaganam and Mario Sassano. 2018. Autonomous collision avoidance for wheeled mobile robots using a differential game approach. European Journal of Control 40 (2018), 53–61

    2018

  21. [29]

    Nvidia. [n. d.]. Nvidia PhysX. https://developer.nvidia.com/physx-sdk

  22. [30]

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas Köpf, Edward Yang, Zach DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Steiner, Lu Fan...

    2019

  23. [31]

    Morgan Quigley, Ken Conley, Brian Gerkey, Josh Faust, Tully Foote, Jeremy Leibs, Rob Wheeler, Andrew Y Ng, et al. 2009. ROS: an open-source Robot Operating System. In ICRA workshop on open source software , Vol. 3. Kobe, Japan, 5

    2009

  24. [32]

    Rudin, David Hoeller, Philipp Reist, and Marco Hutter

    N. Rudin, David Hoeller, Philipp Reist, and Marco Hutter. 2021. Learning to Walk in Minutes Using Massively Parallel Deep Reinforcement Learning. ArXiv abs/2109.11978 (2021). https://api.semanticscholar.org/CorpusID:237635100

    2021 arXiv

  25. [33]

    Battaglia

    Alvaro Sanchez-Gonzalez, Jonathan Godwin, Tobias Pfaff, Rex Ying, Jure Leskovec, and Peter W. Battaglia. 2020. Learning to Simulate Complex Physics with Graph Networks. ArXiv abs/2002.09405 (2020). https://api.semanticscholar. org/CorpusID:211252550

    2020 arXiv

  26. [34]

    Alex Sherstinsky. 2020. Fundamentals of recurrent neural network (RNN) and long short-term memory (LSTM) network. Physica D: Nonlinear Phenomena 404 (2020), 132306

    2020

  27. [35]

    Xingjian Shi, Zhourong Chen, Hao Wang, D. Y. Yeung, Wai-Kin Wong, and Wang chun Woo. 2015. Convolutional LSTM Network: A Machine Learning Approach for Precipitation Nowcasting. In NIPS

    2015

  28. [36]

    Bruno Siciliano, Oussama Khatib, and Torsten Kröger. 2008. Springer handbook of robotics. Vol. 200. Springer

    2008

  29. [37]

    Russell Smith et al. 2005. Open dynamics engine. (2005)

    2005

  30. [38]

    Jürgen Sturm, Nikolas Engelhard, Felix Endres, Wolfram Burgard, and Daniel Cremers. 2012. A benchmark for the evaluation of RGB-D SLAM systems. 2012 IEEE/RSJ International Conference on Intelligent Robots and Systems (2012), 573–

    2012

  31. [39]

    Emanuel Todorov, Tom Erez, and Yuval Tassa. 2012. MuJoCo: A physics engine for model-based control. 2012 IEEE/RSJ International Conference on Intelligent Robots and Systems (2012), 5026–5033

    2012

  32. [40]

    Nur Uddin. 2020. System Identification of Two-Wheeled Robot Dynamics Using Neural Networks. Journal of Physics: Conference Series 1577 (2020). https: //api.semanticscholar.org/CorpusID:225567177

    2020

  33. [41]

    Wang, Haifeng Zhang, and Weinan Zhang

    Muning Wen, Runji Lin, Hanjing Wang, Yaodong Yang, Ying Wen, Luo Mai, J. Wang, Haifeng Zhang, and Weinan Zhang. 2023. Large sequence models for sequential decision-making: a survey. Frontiers of Computer Science 17 (2023). https://api.semanticscholar.org/CorpusID:259252117

    2023

  34. [42]

    Ramadge, and Hao Su

    Yueh-Hua Wu, Ting-Han Fan, Peter J. Ramadge, and Hao Su. 2019. Model Imitation for Model-Based Reinforcement Learning. ArXiv abs/1909.11821 (2019). https://api.semanticscholar.org/CorpusID:202889287

    2019 arXiv

  35. [43]

    Jie Xu, Viktor Makoviychuk, Yashraj Narang, Fabio Ramos, Wojciech Matusik, Animesh Garg, and Miles Macklin. 2021. Accelerated Policy Learning with Parallel Differentiable Simulation. InInternational Conference on Learning Representations

    2021

  36. [44]

    Vechev, and Stelian Coros

    Miguel Zamora, Momchil Peychev, Sehoon Ha, Martin T. Vechev, and Stelian Coros. 2021. PODS: Policy Optimization via Differentiable Simulation. In Interna- tional Conference on Machine Learning. https://api.semanticscholar.org/CorpusID: 235826467 DAI ’23, November 30-December 3...

    2021

  37. [48]

    measured angular acceleration. In Figure 9 and Figure 10, we provide extra visualizations of the robot pose data from the motion capture system, as well as the command and sensor data recorded within the robot itself. B SIMULATION DETAILS The analytical simulation baseline is ...

    2022

  38. [580]

    https://api.semanticscholar.org/CorpusID:206942855

  39. [2016]

    arXiv preprint arXiv:1612.00341 (2016)

    A compositional object-based approach to learning physical dynamics. arXiv preprint arXiv:1612.00341 (2016)

    2016 arXiv

  40. [2018]

    ArXiv abs/1810.01566 (2018)

    Learning Particle Dynamics for Manipulating Rigid Bodies, Deformable Objects, and Fluids. ArXiv abs/1810.01566 (2018). https://api.semanticscholar. org/CorpusID:52917627

    2018 arXiv

Pith tools

Reviewed August 12, 2026 · model on record in the stance chip above.