Pith. sign in

REVIEW 4 major objections 4 minor 53 references

NeuralActuator jointly learns a torque surrogate and sensorless force perception for low-cost servo robots, trained from pose trajectories alone.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-02 06:47 UTC pith:VTSKPW34

load-bearing objection Solid, transparent empirical paper; the force head is a telemetry-to-force regressor in the deployed mode, which the paper discloses but which tempers the headline claim. the 4 major comments →

arxiv 2607.11734 v2 pith:VTSKPW34 submitted 2026-07-13 cs.RO cs.CVcs.GRcs.LG

NeuralActuator: Neural Actuation Modeling for Robot Dynamics and External Force Perception

Zhiyang Dou , John U. Onyemelukwe , Hangxing Zhang , Heng Zhang , Minghao Guo , Yunsheng Tian , Michal Piotr Lipiec , Joshua Jacob
show 4 more authors
Chao Liu Peter Yichen Chen Yuri Ivanov Wojciech Matusik
This is my paper
classification cs.RO cs.CVcs.GRcs.LG
keywords neural actuator modeltorque surrogatedifferentiable simulationsensorless force estimationcontact probability gatemotor condition estimationbehavior cloninglow-cost servo robots
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper tries to establish that low-cost servo-driven robots, whose actuators violate the linear current-to-torque assumption, can still be modeled accurately enough to serve as both dynamics simulators and force sensors. The proposed method, NeuralActuator, is a Transformer that reads a short history of commands, states, and actuator telemetry and outputs a torque surrogate, a contact-gated external force, and a motor-condition score. The torque surrogate is trained through differentiable simulation using only pose trajectories, removing the need for joint-torque sensors or calibrated current-to-torque maps. The paper reports payload force errors as low as 0.08–0.12 N on a 5-DoF arm and shows that feeding the predicted force to a behavior-cloning controller raises pick-and-place success from 80% to 92.5%. If correct, this would make sensorless force perception and force-aware control accessible to budget robot platforms.

Core claim

The paper's central claim is that on low-cost servo robots the standard linear current-to-torque law is unreliable, and a single Transformer-based model — fed a nine-frame history of commanded targets, proprioception, and actuator telemetry — can jointly predict (i) a torque surrogate that, when injected into a differentiable rigid-body simulator, reproduces real pose trajectories; (ii) a contact-gated external end-effector force; and (iii) a per-motor condition score. The torque surrogate is trained without any ground-truth joint-torque label: the only supervision is measured pose, compared with simulated pose after forward integration, and gradients flow back through the differentiable dyn

What carries the argument

The load-bearing object is the torque surrogate τ_pred = g_θ(X_t), produced by a Transformer encoder over a nine-token history, which replaces the linear τ = K_t I assumption as the generalized control input to a differentiable rigid-body simulator. Training uses backpropagation through time on a pose Huber loss, so no torque sensing is needed. Alongside it, a force head outputs a raw 3D force multiplied by a contact-probability gate (f̂ = g·f̂_raw), and a condition head outputs a per-motor score. The design uses 'implicit coupling': the torque surrogate alone drives the simulator, while the force head is supervised in parallel but not applied as a generalized load — a choice the paper tests

Load-bearing premise

The force estimation only stays physically meaningful if the torque surrogate alone can explain the observed motion; the paper acknowledges that under implicit coupling the surrogate 'may absorb interaction effects needed to reproduce the observed motion,' which would make the force head a telemetry-to-force regressor rather than a physically grounded estimator.

What would settle it

Mount a joint-torque sensor on the same arm and compare measured physical torque with the learned torque surrogate under matched commands and loads. If the surrogate differs from the measured torque by much more than the force-head error budget while pose prediction remains accurate, the surrogate is a kinematic fit; equivalently, freeze the shared features and retrain only the force head on a 600 g payload not present in training — a large error jump would show the force head has no independent physical grounding.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Low-cost servo arms can get sensorless end-effector force estimation (order 0.1 N on payloads), enabling force-aware manipulation without force/torque sensors.
  • Differentiable simulation with pose supervision can substitute for torque sensing when calibrating actuator models.
  • Force feedback improves behavior-cloning success rates (80→92.5% pick-and-place, 85→95% lift-and-hold) and reduces over-current trips.
  • A few-trajectory fine-tune (12 trajectories, about 6 minutes) adapts the model to new hardware instances.
  • Motor-condition monitoring emerges from the same telemetry, achieving 91% accuracy and 0.95 AUC under controlled mechanical restriction.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The force head may be largely mediated by the torque surrogate's ability to absorb external-load effects implicitly; dynamic or compliant contacts could expose this dependency, since the surrogate is only trained to match pose.
  • The pose-supervised surrogate scheme could transfer to other low-cost actuator families beyond the two servo types and one industrial arm tested, for example legged robots or dexterous hands with similar non-ideal actuators.
  • The released dataset and protocol could serve as a benchmark for actuator modeling, with the 'no torque labels' trick making large-scale teleoperation data collection feasible without instrumented joints.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 4 minor

Summary. The paper introduces NeuralActuator, a Transformer-based multi-task model that predicts (i) a torque surrogate used to advance a differentiable rigid-body simulator, (ii) a contact-gated external force estimate, and (iii) per-motor condition scores. The torque-surrogate head is trained from pose trajectories through differentiable simulation, avoiding direct joint-torque labels. The authors collect a synchronized teleoperation dataset (NAD) on OpenManipulator-X, extend to SO-101 and Franka, and report low rollout errors, force MAEs of 0.08–0.12 N on payload benchmarks and 0.47–0.73 N on SO-101, improved behavior-cloning success rates, and sub-millisecond inference. The paper is transparent about several limitations, including the implicit force-coupling setup, recorded-telemetry-conditioned rollouts, and the future-state-conditioned Franka benchmark.

Significance. If the force-estimation results hold under causal deployment and consistent evaluation, NeuralActuator would be a valuable contribution to low-cost robot actuation modeling and sensorless force perception. The released dataset, code, and hardware configurations, together with cross-platform validation, careful ablations, and real-time performance analysis, are notable strengths. The central ‘sensorless force perception’ claim, however, currently rests on implicitly coupled, telemetry-conditioned evaluations with a training/evaluation label-alignment mismatch and nominal payload labels during motion. Additional evidence is needed to establish that the force head is physically grounded and generalizes beyond the labeled task set. The work is not circular—the force head is supervised by independent force labels—but the identifiability and generalization concerns are load-bearing for the headline claim.

major comments (4)
  1. [Sec. III-B, Eq. (11); Sec. IV-L] The deployed implicit coupling lets the torque surrogate alone drive the simulator while the force head is supervised but never applied to the dynamics. As the paper states, “The torque surrogate may therefore absorb interaction effects needed to reproduce the observed motion.” Consequently, the force head is not constrained by the residual τ_ID − τ_act of Eq. (4); it is a telemetry-to-force regressor whose strong MAEs in Tabs. IV–VI may reflect correlations between motor current/task identity and the constant or near-constant labels of the protocol, rather than a dynamics-based estimate. The two “unseen contact geometry” checks are stationary holds with constant loads, and the Franka benchmark is explicitly future-state-conditioned, so neither establishes causal generalization outside the labeled task set. Please provide a causal evaluation of the force head under explicit coupling, or
  2. [Appendix B] There is a one-sample label-alignment inconsistency between training and evaluation. Training uses features at time t to advance the state toward q_{t+1}, with force, gate, and condition labels aligned at t+1; the offline evaluators instead compare the time-t feature output with the force label time-stamped at t. At the platform sampling rate this is roughly a 17 ms shift, which can materially affect force metrics during contact onsets and releases. Please report force results under a consistent alignment (either train and evaluate at the same offset, or justify that the one-sample shift is negligible for the reported tasks).
  3. [Sec. III-C, Sec. IV-A] For the payload benchmarks, the ground-truth force label is the nominal gravity vector [0,0,−mg]^T, assigned even while the payload is moving. The paper acknowledges this neglects object inertia and gripping friction. The very low force MAEs (0.08–0.12 N) may therefore reflect the model learning the task-conditioned constant label from telemetry and task identity rather than estimating the actual instantaneous external wrench. Please validate on the force-sensor dataset with the same metric as the payload benchmark, and/or add dynamic payload motions with measured force labels, to show that the payload numbers are not an artifact of a constant reference.
  4. [Sec. IV-F and Appendix B] The Franka benchmark uses a smooth commanded-pose proxy q˜_t^cmd = 1.03 q_rec_{t+5} constructed from a future recorded state. As the paper notes, these results are a future-state-conditioned offline benchmark, not an online evaluation. The conclusion and abstract nevertheless list Franka among the validation platforms for sensorless force estimation. Please either add a causal Franka evaluation (using the controller's actual commanded pose at time t) or explicitly scope the Franka claim in the abstract and conclusion to the offline, future-state-conditioned setting.
minor comments (4)
  1. [Fig. 1 caption] The caption contains incomplete sentences: “The shows the predicted magnitude” and “the shows the force-gauge ground truth” are missing legend identifiers. Please fix the caption.
  2. [Sec. IV-G and Tab. XI] The motor-condition comparison uses 32 training trajectories for NeuralActuator versus 16 for the handcrafted baselines, and evaluates at different decision granularities (framewise vs. windowed). This is disclosed but should be stated more prominently in the table or text to avoid an apples-to-oranges comparison.
  3. [Abstract and Sec. IV-F] The abstract reports “force MAE 0.47–0.73 N on SO-101” without stating that this is at a 500-step recorded-telemetry-conditioned rollout. Please include the horizon and evaluation condition in the abstract or refer to the table.
  4. [Sec. III-D] Training is described as running “up to 100000 epochs” with best checkpoints within 30000 epochs. This is likely iterations or training steps rather than epochs; please clarify the terminology.

Circularity Check

0 steps flagged

No circular derivation found: force head is label-supervised, torque surrogate is pose-supervised, and the main limiations are identifiability/scope caveats, not definitional reductions.

full rationale

NeuralActuator's outputs do not reduce to their own inputs by construction. The torque-surrogate head is trained through differentiable rollout against measured poses (Eqs. 11-15), with no ground-truth torque label that could later be renamed as a prediction. The force head is directly supervised by independent F/T measurements or known-payload gravitational labels (Eq. 16; Sec. III-C), and the contact gate is trained with BCE against a threshold derived from those same force labels (Eq. 17) - supervised classification, not a tautology. Held-out trajectory splits (8:1:1) and cross-platform tests (Tabs. IV-VI, IX-X) provide external evidence. The most plausible circularity concern is the implicit-coupling mode. The paper states: 'Under implicit coupling, τ_pred alone drives the simulated body, while f_hat_ext is directly supervised but is not applied to the differentiable simulator. The torque surrogate may therefore absorb interaction effects needed to reproduce the observed motion.' This is an identifiability and physical-grounding limitation: the force head is free to behave as a telemetry-to-force regressor because it never enters the dynamics. But this is not circular reasoning - the force prediction is still supervised by independent labels, and the paper does not claim it is derived from the residual relation in Eq. 4. Similarly, the Franka experiment is explicitly disclosed as a 'future-state-conditioned offline benchmark rather than an online evaluation' (Sec. IV-F, App. B), and the limitations section states the model 'cannot perform counterfactual rollouts over candidate future command sequences without an additional effort-response model.' These are honest scope restrictions, not hidden reuse of the target. No load-bearing self-citation or uniqueness theorem is invoked. The only overlapping-author citation ([7]) appears in related work and is not used to justify the architecture, training scheme, or results. No equation equates a predicted quantity with a fitted parameter, and no 'prediction' is constructed from the data it is supposed to predict. Therefore the paper exhibits no significant circularity.

Axiom & Free-Parameter Ledger

7 free parameters · 8 axioms · 0 invented entities

The central claim rests on a supervised learning pipeline whose inputs are real robot telemetry and whose labels are either measured forces or known weights; no new physical entity is postulated. The listed parameters are hand-tuned training choices, and the listed axioms are the key domain assumptions that make the pose-supervised torque surrogate and the nominal force labels valid. The identifiability assumption—that the torque surrogate does not silently absorb external-load effects—is the most important unverified premise.

free parameters (7)
  • Transformer architecture hyperparameters = L=9 history, d=192, 4 layers, 4 heads, d_ff=384
    Chosen by hand; no sensitivity analysis reported for these, and they define capacity for all heads.
  • Loss weights w_pos/w_force/w_gate/w_cond/w_vel = 100, 30, 1, 100 (when used), 5/10
    Tuned on validation; the central objective Eq. 19/21 depends on them, with no systematic ablation except architecture/coupling/parameterization.
  • Huber transition beta_f = 0.15 or 1 depending on stage
    Chosen per training stage for the force loss; affects gradient behavior and reported force errors.
  • Contact gate threshold epsilon = 0.01 N
    Set "solely to distinguish zero from nonzero supervision labels"; the gate and hence the gated force output depend on this choice.
  • Franka command-proxy scale and offset = 1.03 q_rec_{t+5}
    Chosen to replace the sparse setpoint channel; the 1.03 factor is hand-set and uses a future recorded state, making the benchmark offline and not directly causal.
  • Rollout horizon curriculum = 128 -> 256 -> 320 (OMX/SO-101); 128 fixed for Franka
    Selected based on gradient norms in Tab. II; affects training dynamics and final accuracy.
  • Gripper loss scale kappa_g = 1000
    Converts single-finger slide-coordinate residual from meters to millimeters; hand-set.
axioms (8)
  • domain assumption Rigid-body inverse dynamics identity (Eq. 4): M(q)q_ddot + C(q,q_dot)q_dot + g(q) = tau_act + J^T f_ext
    The paper assumes M, C, g are correctly encoded by the simulator and that residual generalized forces decompose into actuator torque plus Jacobian-transposed external force (Sec. III-B).
  • domain assumption 3D end-effector force with negligible residual moment
    "using a three-dimensional force rather than a six-dimensional wrench assumes that the residual moment about that point is negligible" (Sec. III-B).
  • domain assumption Known-weight labels are exact during static/quasi-static holds
    "This reference is exact for static or quasi-static holding; during motion, it provides nominal supervision that does not account for object inertia or gripping friction" (Sec. III-C).
  • ad hoc to paper Contact-gate threshold epsilon = 0.01 N defines ground-truth contact
    The gate target g_gt = I[||f_gt|| > epsilon] uses this threshold, and the final force output f_hat = g * f_raw depends on it; no principled calibration is given.
  • ad hoc to paper Franka smooth command proxy approximates the real command channel
    q_cmd_tilde = 1.03 q_rec_{t+5} replaces sparse setpoints; future-state conditioning makes the benchmark offline and depends on this proxy being a valid stand-in for commanded pose.
  • domain assumption Direct torque-surrogate parameterization is a valid generalized effort input
    The model treats tau_pred as the simulator's control input without assuming a linear current-torque prior; this is the core modeling choice, tested only indirectly through pose-matching (Sec. III-D, Eq. 5).
  • domain assumption MuJoCo/MJX and Warp provide accurate differentiable rigid-body dynamics
    The training and evaluation use MJX as the differentiable simulator and Warp as an additional backend; errors or non-physical parameters in these engines would be absorbed into the learned surrogate.
  • ad hoc to paper Nine-frame Transformer context captures the relevant actuator history
    History length H_hist = 8 plus current frame is chosen without an ablation on context length; temporal dependencies in friction/thermal/backlash are assumed to fit in this window.

pith-pipeline@v1.3.0-alltime-deepseek · 32179 in / 17289 out tokens · 168903 ms · 2026-08-02T06:47:31.115054+00:00 · methodology

0 comments
read the original abstract

Differentiable simulators have advanced policy learning and model-based control across robotic tasks. Yet actuator dynamics remain underexplored and can be a major source of sim-to-real error, particularly on low-cost platforms, where the linear current-to-joint-torque approximation $\tau = K_t I$ becomes unreliable because of friction, hysteresis, backlash, and thermal effects. Accurate actuator models can also support force perception and integrated force/position control. We present NeuralActuator, which jointly predicts (i) a torque surrogate for trajectory propagation on low-cost servo platforms, (ii) external forces with a contact-probability gate for sensorless force perception, and (iii) a motor-condition score for a supervised joint, distinguishing normal from mechanically restricted operation. A twin-arm teleoperation system records robot states and actuator telemetry alongside external-force labels, yielding the Neural Actuation Dataset (NAD). The torque-surrogate head is trained through differentiable simulation from pose trajectories without ground-truth joint-torque measurements. A Transformer captures temporal dependencies while enabling real-time inference. We validate NeuralActuator on a 5-DoF OpenManipulator-X, a 6-DoF SO-101 from LeRobot, and a 7-DoF Franka Emika Panda, spanning three actuator families and costs from approximately \$500 to more than \$30{,}000. The low-cost platforms support physically plausible dynamics and force evaluation, while the offline Franka experiment provides a payload-force-estimation benchmark. We also demonstrate motor-condition estimation and improved behavior-cloning performance using NeuralActuator as a pretrained module. We release the dataset, code, and hardware configurations on the project page: https://frank-zy-dou.github.io/projects/NeuralActuator/index.html.

Figures

Figures reproduced from arXiv: 2607.11734 by Chao Liu, Hangxing Zhang, Heng Zhang, John U. Onyemelukwe, Joshua Jacob, Michal Piotr Lipiec, Minghao Guo, Peter Yichen Chen, Wojciech Matusik, Yunsheng Tian, Yuri Ivanov, Zhiyang Dou.

Figure 1
Figure 1. Figure 1: (a) Push–force gauge validation at three end-effector heights. We show NeuralActuator pushing a force gauge at low, middle, and high end-effector heights (top to bottom). Left: real-robot execution; middle: corresponding simulated rollout; right: contact-axis force-magnitude profiles. The trace￾level mean absolute errors (MAEs) are 0.037 N, 0.015 N, and 0.028 N, respectively. These values correspond to the… view at source ↗
Figure 2
Figure 2. Figure 2: Neural Actuation Dataset (NAD) data collection. (a) Overview of the twin-arm leader–follower system with an external camera. (b) An operator kinesthetically drives the leader to generate diverse motions; the follower mirrors these motions while manipulating a payload and recording synchronized actuator currents, robot states, and end-effector forces. out tactile skins) [26], whole-robot touch sensing witho… view at source ↗
Figure 3
Figure 3. Figure 3: NeuralActuator pipeline. At each time step t, NeuralActuator takes the commanded pose q cmd(t), an effort-related actuator signal u(t), actuator-side telemetry, and tracking feedback e(t), together with a history buffer Ht that summarizes recent commands, states, and telemetry (e.g., joint poses and velocities). A Transformer encoder (×N blocks) maps the sequence to four task heads: (i) a simulator￾equival… view at source ↗
Figure 4
Figure 4. Figure 4: Data verification in NAD. We time-synchronize the real-robot video (left) with the logged trajectory and verify alignment in a GUI (right), where the red arrow visualizes the estimated external force. 20.49 s), including (a) clockwise and counterclockwise circular end-effector motions, (b) individual joint sweeps covering approximately 90% of feasible ranges, and (c) whole-arm primitives such as lean and e… view at source ↗
Figure 5
Figure 5. Figure 5: Robot arm simulation with external force estimation on the test set. (a) NeuralActuator estimates end-effector external forces along six directions (±X, ±Y, ±Z); red arrows visualize the estimated force vectors in simulation (paired with corresponding real-robot executions). (b) Force estimation during force-aware manipulation: lifting and holding 200 g and 300 g payloads (top) and pick-and-place with 200 … view at source ↗
Figure 5
Figure 5. Figure 5: Robot arm simulation with external force estimation on the test set. (a) NeuralActuator estimates end-effector external forces along six directions (±X, ±Y, ±Z); red arrows visualize the estimated force vectors in simulation (paired with corresponding real-robot executions). (b) Force estimation during force-aware manipulation: lifting and holding 200 g and 300 g payloads (top) and pick-and-place with 200 … view at source ↗
Figure 6
Figure 6. Figure 6: Kinematics and force transmission of the symmetric gripper. Top (a) and side (b) views show the rotary motor driving two symmetric prismatic jaws. Each link transmits an axial force T, whose component along the closing direction gives the per-jaw force F = T cos β. where mt,3 = 1 denotes normal operation of Joint 3 and mt,3 = 0 denotes the mechanically restricted condition. For selected SO-101 and Franka r… view at source ↗
Figure 7
Figure 7. Figure 7: Force-aware pick-and-place rollouts. From top to bottom: 200 g, 300 g, 400 g, and 500 g payloads. Each row pairs the real execution with the simulated rollout and overlays the predicted end-effector external force. TABLE VI: Simulation and force prediction accuracy on gauge pushing. The end-effector contacts the gauge from the front or top at three positions. J1–J4: joint-angle MAE (deg). Grip: single-fing… view at source ↗
Figure 8
Figure 8. Figure 8: Force-aware lift-and-hold rollouts. From top to bottom: 200 g, 300 g, and 400 g payloads. Each row pairs the real execution with the simulated rollout and overlays the predicted end-effector external force [PITH_FULL_IMAGE:figures/full_fig_p011_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Force prediction for front-direction gauge pushing. For high, middle, and low contact positions, we show the physical setup and simulated scene together with the measured and predicted signed contact-axis force. The bottom panels show the corresponding binary contact interval. TABLE VII: Force MAE (N) against model-based sensorless baselines on the payload benchmark at a 100-step horizon (lower is better).… view at source ↗
Figure 11
Figure 11. Figure 11: Online adaptation progress. The curve-level aggre￾gate shows approximately 44% average tracking-error reduc￾tion after 70 epochs with 12 trajectories (7,810 frames). Fine￾tuning takes 356 seconds in total (approximately 29.7 seconds per trajectory and 5.1 seconds per epoch) [PITH_FULL_IMAGE:figures/full_fig_p012_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Cross-platform validation. We train NeuralActuator from scratch on a 7-DoF Franka Emika Panda and a 6-DoF SO-101 low-cost arm with STS3215 servos. Franka provides an offline payload-force-estimation benchmark, whereas SO￾101 provides both rollout and force evaluation. Together with OpenManipulator-X, these robots span three actuator families and costs from approximately $500 to more than $30,000. on exter… view at source ↗
Figure 13
Figure 13. Figure 13: Motor-condition experiment. (a) Rubber bands mechanically restrict Joint 3. (b) Under matched commands, the unrestricted and restricted trials follow similar position trajectories (top), but the restricted joint draws more current (middle), producing a measurable difference (bottom). and-place task with a 200 g object (Tab. XI), with higher reported values than the threshold, SVM, and Random Forest baseli… view at source ↗
Figure 14
Figure 14. Figure 14: Image-space refinement. The evaluated pipeline refines the hand–eye transform and NeuralActuator parameters through the simulated robot state; the visualization reports silhouette alignment. J. Visual Supervision We evaluate an image-space refinement pipeline that com￾poses the neural actuator, differentiable dynamics, forward kinematics, and a differentiable renderer. A geometric three￾point initializati… view at source ↗
Figure 15
Figure 15. Figure 15: NeuralActuator with NVIDIA Warp. Representative articulated-arm rollouts under different initial configurations and external push directions (red arrows), simulated with wp.sim.FeatherstoneIntegrator. TABLE XIV: Comparison of neural architectures for actua￾tion modeling. We report rollout and force prediction accuracy at a 500-step prediction horizon. Bold marks the best result per column; ties are bolded… view at source ↗
Figure 16
Figure 16. Figure 16: NeuralActuator with NVIDIA Warp. Representative articulated-arm rollouts under different initial configurations and external push directions (red arrows), simulated with wp.sim.FeatherstoneIntegrator. predict the effort response to unexecuted commands and there￾fore cannot perform counterfactual rollouts over candidate fu￾ture command sequences without an additional effort-response model. The force head e… view at source ↗

Review history (2 revisions) →

discussion (0)

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

Reference graph

Works this paper leans on

53 extracted references · 4 linked inside Pith

  1. [1]

    Augmenting physical simulators with stochastic neural networks: Case study of planar pushing and bouncing

    Anurag Ajay, Jiajun Wu, Nima Fazeli, Maria Bauza, Leslie P Kaelbling, Joshua B Tenenbaum, and Al- berto Rodriguez. Augmenting physical simulators with stochastic neural networks: Case study of planar pushing and bouncing. In2018 IEEE/RSJ International Confer- ence on Intelligent Robots and Systems (IROS), pages 3066–3073. IEEE, 2018

    2018

  2. [2]

    Observer-extended direct method for collision monitoring in robot manipulators using proprioception and IMU sensing.IEEE Robotics and Automation Letters, 5(2):954–961, 2020

    Seyed Ali Baradaran Birjandi, Johannes K ¨uhn, and Sami Haddadin. Observer-extended direct method for collision monitoring in robot manipulators using proprioception and IMU sensing.IEEE Robotics and Automation Letters, 5(2):954–961, 2020

    2020

  3. [3]

    JAX: composable transformations of Python+NumPy programs, 2018

    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. JAX: composable transformations of Python+NumPy programs, 2018. URL http://github.com/jax-ml/jax

    2018

  4. [4]

    LeRobot: State-of-the-art machine learning for real-world robotics in PyTorch

    Remi Cadene, Simon Alibert, Alexander Soare, Quentin Gallouedec, Adil Zouitine, Steven Palma, Pepijn Kooijmans, Michel Aractingi, Mustafa Shukor, Dana Aubakirova, Martino Russi, Francesco Capuano, Caro- line Pascal, Jade Choghari, Jess Moss, and Thomas Wolf. LeRobot: State-of-the-art machine learning for real-world robotics in PyTorch. https://github.com/...

    2024

  5. [5]

    SAM 3: Segment anything with concepts, 2025

    Nicolas Carion, Laura Gustafson, Yuan-Ting Hu, Shoub- hik Debnath, Ronghang Hu, Didac Suris, Chaitanya Ryali, Kalyan Vasudev Alwala, Haitham Khedr, An- drew Huang, Jie Lei, Tengyu Ma, Baishan Guo, Arpit Kalla, Markus Marks, Joseph Greer, Meng Wang, Peize Sun, Roman R ¨adle, Triantafyllos Afouras, Effrosyni Mavroudi, Katherine Xu, Tsung-Han Wu, Yu Zhou, Li...

    Pith/arXiv arXiv 2025

  6. [6]

    EasyHeC: Accurate and automatic hand-eye cal- ibration via differentiable rendering and space explo- ration.IEEE Robotics and Automation Letters, 8(11): 7234–7241, November 2023

    Linghao Chen, Yuzhe Qin, Xiaowei Zhou, and Hao Su. EasyHeC: Accurate and automatic hand-eye cal- ibration via differentiable rendering and space explo- ration.IEEE Robotics and Automation Letters, 8(11): 7234–7241, November 2023. ISSN 2377-3774. doi: 10. 1109/lra.2023.3315551. URL http://dx.doi.org/10.1109/ LRA.2023.3315551

    arXiv 2023

  7. [7]

    Learning object properties using robot proprioception via differentiable robot-object interaction

    Peter Yichen Chen, Chao Liu, Pingchuan Ma, John Eastman, Daniela Rus, Dylan Randle, Yuri Ivanov, and Wojciech Matusik. Learning object properties using robot proprioception via differentiable robot-object interaction. In2025 IEEE International Conference on Robotics and Automation (ICRA), pages 5997–6004. IEEE, 2025

    2025

  8. [8]

    Empirical evaluation of gated recurrent neural networks on sequence modeling.arXiv preprint arXiv:1412.3555, 2014

    Junyoung Chung, Caglar Gulcehre, KyungHyun Cho, and Yoshua Bengio. Empirical evaluation of gated recurrent neural networks on sequence modeling.arXiv preprint arXiv:1412.3555, 2014

    Pith/arXiv arXiv 2014

  9. [9]

    Visual contact pressure estimation for grippers in the wild

    Jeremy A Collins, Cody Houff, Patrick Grady, and Charles C Kemp. Visual contact pressure estimation for grippers in the wild. In2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 10947–10954. IEEE, 2023

    2023

  10. [10]

    Sensorless robot collision detection and hybrid force/motion control

    Alessandro De Luca and Raffaella Mattone. Sensorless robot collision detection and hybrid force/motion control. InProceedings of the 2005 IEEE international confer- ence on robotics and automation, pages 999–1004. IEEE, 2005

    2005

  11. [11]

    Collision detection and safe reaction with the DLR-III lightweight manipulator arm

    Alessandro De Luca, Alin Albu-Schaffer, Sami Had- dadin, and Gerd Hirzinger. Collision detection and safe reaction with the DLR-III lightweight manipulator arm. In2006 IEEE/RSJ international conference on intelligent robots and systems, pages 1623–1630. IEEE, 2006

    2006

  12. [12]

    Bridging the sim-to-real gap for athletic loco-manipulation.arXiv preprint arXiv:2502.10894, 2025

    Nolan Fey, Gabriel B Margolis, Martin Peticco, and Pulkit Agrawal. Bridging the sim-to-real gap for athletic loco-manipulation.arXiv preprint arXiv:2502.10894, 2025

    Pith/arXiv arXiv 2025

  13. [13]

    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. Brax - a differentiable physics engine for large scale rigid body simulation, 2021. URL http://github.com/google/brax

    2021

  14. [14]

    UniTac: Whole-robot touch sensing without tactile sensors

    Wanjia Fu, Hongyu Li, Ivy X He, Stefanie Tellex, and Srinath Sridhar. UniTac: Whole-robot touch sensing without tactile sensors. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, pages 3976–3985, June 2026

    2026

  15. [15]

    Genesis: A generative and universal physics engine for robotics and beyond, December 2024

    Genesis Authors. Genesis: A generative and universal physics engine for robotics and beyond, December 2024. URL https://github.com/Genesis-Embodied-AI/Genesis

    2024

  16. [16]

    MuJoCo XLA (MJX)

    Google DeepMind. MuJoCo XLA (MJX). https:// mujoco.readthedocs.io/en/stable/mjx.html, 2026. Official software documentation, accessed July 13, 2026

    2026

  17. [17]

    Collision detection and reaction: A contribution to safe physical human-robot interaction

    Sami Haddadin, Alin Albu-Schaffer, Alessandro De Luca, and Gerd Hirzinger. Collision detection and reaction: A contribution to safe physical human-robot interaction. In2008 IEEE/RSJ International Conference on Intelligent Robots and Systems, pages 3356–3363. IEEE, 2008

    2008

  18. [18]

    Toward sensorless interaction force estimation for industrial robots using high-order finite-time ob- servers.IEEE Transactions on Industrial Electronics, 69 (7):7275–7284, 2022

    Linyan Han, Jianliang Mao, Pengfei Cao, Yahui Gan, and Shihua Li. Toward sensorless interaction force estimation for industrial robots using high-order finite-time ob- servers.IEEE Transactions on Industrial Electronics, 69 (7):7275–7284, 2022. doi: 10.1109/TIE.2021.3095820

    arXiv 2022

  19. [19]

    Force map: Learning to predict contact force distribution from vision

    Ryo Hanai, Yukiyasu Domae, Ixchel G Ramirez-Alpizar, Bruno Leme, and Tetsuya Ogata. Force map: Learning to predict contact force distribution from vision. In2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 3129–3136. IEEE, 2023

    2023

  20. [20]

    NeuralSim: Augmenting differentiable simulators with neural networks

    Eric Heiden, David Millard, Erwin Coumans, Yizhou Sheng, and Gaurav S Sukhatme. NeuralSim: Augmenting differentiable simulators with neural networks. In2021 IEEE International Conference on Robotics and Automa- tion (ICRA), pages 9474–9481. IEEE, 2021

    2021

  21. [21]

    Long short- term memory.Neural computation, 9(8):1735–1780, 1997

    Sepp Hochreiter and J ¨urgen Schmidhuber. Long short- term memory.Neural computation, 9(8):1735–1780, 1997

    1997

  22. [22]

    Dojo: A differentiable physics engine for robotics.arXiv preprint arXiv:2203.00806, 2022

    Taylor A Howell, Simon Le Cleac’h, Jan Br ¨udigam, J Zico Kolter, Mac Schwager, and Zachary Manchester. Dojo: A differentiable physics engine for robotics.arXiv preprint arXiv:2203.00806, 2022

    Pith/arXiv arXiv 2022

  23. [23]

    Contact force estimation for robot manipulator using semiparametric model and dis- turbance Kalman filter.IEEE Transactions on Industrial Electronics, 65(4):3365–3375, 2018

    Jin Hu and Rong Xiong. Contact force estimation for robot manipulator using semiparametric model and dis- turbance Kalman filter.IEEE Transactions on Industrial Electronics, 65(4):3365–3375, 2018. doi: 10.1109/TIE. 2017.2748056

    arXiv 2018

  24. [24]

    DiffTaichi: Differentiable programming for physical sim- ulation.ICLR, 2020

    Yuanming Hu, Luke Anderson, Tzu-Mao Li, Qi Sun, Nathan Carr, Jonathan Ragan-Kelley, and Fr ´edo Durand. DiffTaichi: Differentiable programming for physical sim- ulation.ICLR, 2020

    2020

  25. [25]

    Learning agile and dynamic motor skills for legged robots.Science Robotics, 4(26):eaau5872, 2019

    Jemin Hwangbo, Joonho Lee, Alexey Dosovitskiy, Dario Bellicoso, Vassilios Tsounis, Vladlen Koltun, and Marco Hutter. Learning agile and dynamic motor skills for legged robots.Science Robotics, 4(26):eaau5872, 2019

    2019

  26. [26]

    Intrinsic sense of touch for intuitive physi- cal human-robot interaction.Science Robotics, 9(93): eadn4008, 2024

    Maged Iskandar, Alin Albu-Sch ¨affer, and Alexander Di- etrich. Intrinsic sense of touch for intuitive physi- cal human-robot interaction.Science Robotics, 9(93): eadn4008, 2024

    2024

  27. [27]

    Data- augmented contact model for rigid body simulation

    Yifeng Jiang, Jiazheng Sun, and C Karen Liu. Data- augmented contact model for rigid body simulation. In Learning for dynamics and control conference, pages 378–390. PMLR, 2022

    2022

  28. [28]

    Force control of a hydraulic actuator with a neural network inverse model.IEEE Robotics and Automation Letters, 6(2): 2814–2821, 2021

    Sung-Woo Kim, Buyoun Cho, Seunghoon Shin, Jun-Ho Oh, Jemin Hwangbo, and Hae-Won Park. Force control of a hydraulic actuator with a neural network inverse model.IEEE Robotics and Automation Letters, 6(2): 2814–2821, 2021

    2021

  29. [29]

    ILBiT: Imitation learning for robot using position and torque information based on bilateral control with transformer.IEEJ Journal of Industry Applications, 14(2):161–168, 2025

    Masato Kobayashi, Thanpimon Buamanee, Yuki Uran- ishi, and Haruo Takemura. ILBiT: Imitation learning for robot using position and torque information based on bilateral control with transformer.IEEJ Journal of Industry Applications, 14(2):161–168, 2025

    2025

  30. [30]

    Learning-based force control of twisted string actuators using a neural network-based inverse model.IEEE Robotics and Automation Letters, 9(9):8170–8177, 2024

    Hyeokjun Kwon, Sung-Woo Kim, and Hyun-Min Joe. Learning-based force control of twisted string actuators using a neural network-based inverse model.IEEE Robotics and Automation Letters, 9(9):8170–8177, 2024. doi: 10.1109/LRA.2024.3428909

    arXiv 2024

  31. [31]

    Robotic force estimation using motor torques and modeling of low velocity friction dis- turbances

    Magnus Linderoth, Andreas Stolt, Anders Robertsson, and Rolf Johansson. Robotic force estimation using motor torques and modeling of low velocity friction dis- turbances. In2013 IEEE/RSJ International Conference on Intelligent Robots and Systems, pages 3550–3556. IEEE, 2013

    2013

  32. [32]

    Direct torque control for PMSM based on the RBFNN surrogate model of electromagnetic torque and stator flux linkage

    Hongda Liu, Wentie Niu, and Yonghao Guo. Direct torque control for PMSM based on the RBFNN surrogate model of electromagnetic torque and stator flux linkage. Control Engineering Practice, 148:105943, 2024. doi: 10.1016/j.conengprac.2024.105943

    arXiv 2024

  33. [33]

    Sichao Liu, Lihui Wang, and Xi Vincent Wang. Sensor- less force estimation for industrial robots using distur- bance observer and neural learning of friction approxima- tion.Robotics and Computer-Integrated Manufacturing, 71:102168, 2021

    2021

  34. [34]

    Decoupled weight decay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. InInternational Conference on Learning Representations, 2019

    2019

  35. [35]

    PincherX 150 5DOF robot arm

    Tribotix Pty Ltd. PincherX 150 5DOF robot arm. https://tribotix.com/product/pincherx-150-robot-arm/,

  36. [36]

    Warp: A high-performance Python framework for GPU simulation and graphics

    Miles Macklin. Warp: A high-performance Python framework for GPU simulation and graphics. https: //github.com/nvidia/warp, March 2022. NVIDIA GPU Technology Conference (GTC)

    2022

  37. [37]

    Estimation of contact forces using a virtual force sensor

    Emanuele Magrini, Fabrizio Flacco, and Alessandro De Luca. Estimation of contact forces using a virtual force sensor. In2014 IEEE/RSJ International Conference on Intelligent Robots and Systems, pages 2126–2133. IEEE, 2014

    2014

  38. [38]

    Localizing external contact using proprioceptive sensors: The contact particle filter

    Lucas Manuelli and Russ Tedrake. Localizing external contact using proprioceptive sensors: The contact particle filter. In2016 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 5062–5069. IEEE, 2016

    2016

  39. [39]

    Using deep neural networks to improve contact wrench estimation of serial robotic manipulators in static tasks.Frontiers in Robotics and AI, 9:892916, 2022

    Jonas Osburg, Ivo Kuhlemann, Jannis Hagenah, and Floris Ernst. Using deep neural networks to improve contact wrench estimation of serial robotic manipulators in static tasks.Frontiers in Robotics and AI, 9:892916, 2022

    2022

  40. [40]

    PhantomX reactor robot arm (AX-12A)

    Interbotix (Trossen Robotics). PhantomX reactor robot arm (AX-12A). https://www.interbotix.com/p/ phantomx-ax-12-reactor-robot-arm.aspx, 2025. Ac- cessed: 2025-08-12

    2025

  41. [41]

    OpenMANIPULATOR-X: Overview (RM-X52-TNM)

    ROBOTIS Co., Ltd. OpenMANIPULATOR-X: Overview (RM-X52-TNM). https://emanual.robotis.com/docs/en/ platform/openmanipulator x/overview/, 2017. Accessed: 2025-07-12

    2017

  42. [42]

    Improving domain transfer of robot dynamics models with geometric system identifi- cation and learned friction compensation

    Laura Schwendeman, Andrew SaLoutos, Elijah Stanger- Jones, and Sangbae Kim. Improving domain transfer of robot dynamics models with geometric system identifi- cation and learned friction compensation. In2023 IEEE- RAS 22nd International Conference on Humanoid Robots (Humanoids), pages 1–8. IEEE, 2023

    2023

  43. [43]

    Transformer-based neural augmentation of robot simu- lation representations.IEEE Robotics and Automation Letters, 8(6):3748–3755, 2023

    Agon Serifi, Espen Knoop, Christian Schumacher, Naveen Kumar, Markus Gross, and Moritz B ¨acher. Transformer-based neural augmentation of robot simu- lation representations.IEEE Robotics and Automation Letters, 8(6):3748–3755, 2023

    2023

  44. [44]

    Surrogate modeling of electrical machine torque using artificial neural networks.IEEE Access, 8:220027–220045, 2020

    Mikko Tahkola, Janne Ker ¨anen, Denis Sedov, Mehrnaz Farzam Far, and Juha Kortelainen. Surrogate modeling of electrical machine torque using artificial neural networks.IEEE Access, 8:220027–220045, 2020. doi: 10.1109/ACCESS.2020.3042834

    arXiv 2020

  45. [45]

    Mu- JoCo: A physics engine for model-based control

    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

    arXiv 2012

  46. [46]

    VGGT: Visual geometry grounded transformer

    Jianyuan Wang, Minghao Chen, Nikita Karaev, An- drea Vedaldi, Christian Rupprecht, and David Novotny. VGGT: Visual geometry grounded transformer. InPro- ceedings of the Computer Vision and Pattern Recognition Conference, pages 5294–5306, 2025

    2025

  47. [47]

    A PINN-based non- linear PMSM electromagnetic model using differential inductance theory.Applied Sciences, 15(13):7162, 2025

    Songyi Wang and Xinjian Wang. A PINN-based non- linear PMSM electromagnetic model using differential inductance theory.Applied Sciences, 15(13):7162, 2025. doi: 10.3390/app15137162

    doi:10.3390/app15137162 2025

  48. [48]

    Visual haptic reasoning: Estimating contact forces by observing deformable object interactions.IEEE Robotics and Automation Letters, 7(4):11426–11433, 2022

    Yufei Wang, David Held, and Zackory Erickson. Visual haptic reasoning: Estimating contact forces by observing deformable object interactions.IEEE Robotics and Automation Letters, 7(4):11426–11433, 2022

    2022

  49. [49]

    Neural robot dynamics

    Jie Xu, Eric Heiden, Iretiayo Akinola, Dieter Fox, Miles Macklin, and Yashraj Narang. Neural robot dynamics. InProceedings of the 9th Conference on Robot Learn- ing, volume 305 ofProceedings of Machine Learning Research, pages 3915–3935. PMLR, 2025. URL https: //proceedings.mlr.press/v305/xu25g.html

    2025

  50. [50]

    Torque estimation and control of PMSM based on deep learning

    Yu-Bai Yan, Jia-Ning Liang, Tian-Fu Sun, Jian-Ping Geng, Gang Xie, and Dong-Jia Pan. Torque estimation and control of PMSM based on deep learning. In2019 22nd International Conference on Electrical Machines and Systems (ICEMS). IEEE, 2019. doi: 10.1109/ICEMS. 2019.8921886

    arXiv 2019

  51. [51]

    TossingBot: Learning to throw arbitrary objects with residual physics.IEEE Transactions on Robotics, 36(4):1307–1319, 2020

    Andy Zeng, Shuran Song, Johnny Lee, Alberto Ro- driguez, and Thomas Funkhouser. TossingBot: Learning to throw arbitrary objects with residual physics.IEEE Transactions on Robotics, 36(4):1307–1319, 2020

    2020

  52. [52]

    Learning a unified policy for position and force control in legged loco-manipulation

    Peiyuan Zhi, Peiyang Li, Jianqin Yin, Baoxiong Jia, and Siyuan Huang. Learning a unified policy for position and force control in legged loco-manipulation. InProceed- ings of the 9th Conference on Robot Learning, volume 305 ofProceedings of Machine Learning Research, pages 652–669. PMLR, 2025. APPENDIX This appendix provides dataset details (Sec. A), impl...

    2025

  53. [2020]

    Accessed: 2025-08-12

    2025