Pith. sign in

REVIEW 4 major objections 4 minor 22 references

Task-specific Self-body Controller Acquisition by Musculoskeletal Humanoids: Application to Pedal Control in Autonomous Driving

T0 review · 4 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read This paper claims a musculoskeletal humanoid can learn a pedal controller directly from 60 seconds of random driving data, reaching target speeds faster than PID.

desk verdict A credible but under-evidenced application of learned-dynamics MPC to musculoskeletal pedal control; the central generalization claim is plausible but not yet demonstrated. read the letter →

arxiv 2412.08270 v1 pith:MSVCOZ65 submitted 2024-12-11 cs.RO

Kento Kawaharazuka , Kei Tsuzuki , Shogo Makino , Moritaka Onitsuka , Koki Shinjo , Yuki Asano , Kei Okada , Koji Kawasaki
show 1 more author
Masayuki Inaba
This is my paper · ORCID
classification cs.RO
keywords musculoskeletalhumanoidDDC-Nettask-specificcontrolneuralnetworkdynamicsmodelautonomousdrivingpedalreal-timebackpropagationtoinput
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

This paper tries to show that a musculoskeletal humanoid can acquire a task-specific controller without building an accurate model of its own flexible body. The proposed method, DDC-Net, learns a direct time-series map from ankle-pitch pedal commands to future car velocity from 300 random driving samples collected over 60 seconds, then inverts that map in real time by backpropagating the target-velocity error into the command sequence. In accelerator-pedal experiments the learned controller reaches 5 km/h within 20% in 0.9 seconds and 10 km/h within 10% in 5.3 seconds, beating two PID baselines at those thresholds. The point is that the robot does not need an accurate joint-muscle model; it needs only a task-level input-output model and a way to optimize through it.

What carries the argument

The central object is DDC-Net, a five-layer fully connected neural network implementing Eq. (1): $f$ maps the current state $i_t$ together with $N$ future control inputs $u_{t+1},\dots,u_{t+N}$ to $N$ predicted future task states $s_{t+1},\dots,s_{t+N}$. Each hidden layer is batch-normalized with sigmoid activations, with hidden units $\{80,50,20\}$, input dimension $N_i + N \times N_u = 4 + 30$, and output dimension $N \times N_s = 30$. It carries the argument because it replaces the unmodelable chain of muscle wires, soft tissue, and vehicle dynamics with a differentiable surrogate that can be trained from random motion data and then inverted by gradient descent on the control input. The control loop's loss $L = \mathrm{MSE}(s_{\text{pred}}, s_{\text{target}}) + \alpha \,\mathrm{AdjacentError}(u_{\text{init}})$ makes the inversion produce smooth command sequences, and the two-stage batched initialization ($N_{\text{batch}}=10$ noisy copies, $N_1=10$ then $N_2=20$ gradient steps) keeps the real-time optimization stable at 5 Hz.

What would settle it

Run the learned DDC-Net controller on ten fresh trials from different initial velocities (for example, starting at 2 km/h and 7 km/h instead of 0) and compare the predicted velocity trajectory against the measured one over the six-second horizon; if predicted and actual velocities diverge beyond the reported 10-20% bands, or if the commanded pedal sequence fails to converge faster than PID2 on those starts, the claim that 60 seconds of random data suffice for real-time control is refuted.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that for under-modeled musculoskeletal robots the right object to learn is not the joint-muscle mapping but the dynamic relationship between the control input sequence and the task state sequence. DDC-Net is defined by Eq. (1): a function $f$ that takes the current state $i_t$ (car velocity, car acceleration, ankle-pitch angle, ankle-pitch velocity) and a length-$N=30$ sequence of future ankle-pitch angles $u_{t+1},\dots,u_{t+N}$ and outputs the predicted velocities $s_{t+1},\dots,s_{t+N}$. After training on random pedal inputs at 5 Hz for 60 seconds, the network is used as a differentiable forward model: the control phase initializes a command sequence from the previous solution, adds noise to form a batch, minimizes the MSE between predicted and target velocity plus a smoothness penalty on adjacent commands, and backpropagates the gradient into the input sequence. The first optimized command is sent to the robot; repeated at 5 Hz, this realizes target speeds in roughly one to five seconds. The paper reports that this converges faster than PID controls and, unlike them, requires no gain tuning for the pedal task.

Load-bearing premise

The load-bearing premise is that 300 random pedal samples taken over one 60-second run capture the robot-vehicle dynamics accurately enough that gradient descent through the learned network produces correct commands on later runs, including speeds and initial conditions not explicitly in the training set; the paper's reported experiments are single runs with no repetition or off-distribution test.

Editorial extensions

If this is right

  • Applied to the same robot, the method should control the brake pedal, steering, and body-tilt compensation without retuning the network, using the same DDC-Net structure.
  • The control input need not be joint angle; the same loss-and-backprop scheme should work with joint velocity, joint torque, or muscle tension commands.
  • Because the target state sequence enters the loss, the method should track time-varying velocity profiles, such as smooth stopping trajectories chosen by jerk minimization.
  • The method removes manual PID gain tuning for pedal control; the main remaining tuning parameter is $\alpha$, the smoothness weight.
  • Scaling to multi-DOF tasks requires reducing the control input space, for example by muscle synergy or autoencoder compression, because the input space grows exponentially with the number of control dimensions.

Reading between the lines

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

  • The algorithm is, in effect, a learned model-predictive controller with a six-second horizon and a differentiable forward model; any plant whose dynamics are hard to model analytically but easy to excite with random inputs could use the same recipe, not just musculoskeletal humanoids.
  • Because training data come from one random controller run near the target region, the learned model is only guaranteed where that run explored; an active data-collection scheme that re-runs random inputs after seeing large prediction errors could widen the reliable operating envelope.
  • The batched noisy initialization and normalized gradient descent resemble a tiny population-based optimizer; treating it that way suggests testable variations, such as covariance-matrix-style updates or longer horizons, that might improve convergence further.
  • A direct comparison with a learned dynamics model plus standard model-predictive control would isolate whether the contribution is the network structure or the backprop-through-input optimizer.
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

4 major / 4 minor

Summary. The paper proposes DDC-Net, a fully-connected neural network f that maps the current task state i_t and a time-series of N future control inputs u_{t+1},...,u_{t+N} to a time-series of predicted task states s_{t+1},...,s_{t+N}. The network is trained offline on 300 samples (60 s at 5 Hz) of random ankle-pitch pedal control data collected on the musculoskeletal humanoid Musashi driving a COMS electric vehicle on rollers. In the control phase, the method backpropagates a loss between the predicted and target car-velocity sequences through f to iteratively optimize the control input sequence, then executes the first element at 5 Hz. The paper reports convergence times of 0.9 s (within 20% of 5 km/h) and 5.3 s (within 10% of 10 km/h), claiming faster convergence than two PID-based baselines. The contribution is framed as task-specific self-body controller acquisition that bypasses explicit modeling of the musculoskeletal vehicle-driver dynamics.

Significance. If the central claim were substantiated, the method would be a practically valuable way to acquire controllers for musculoskeletal humanoids in tasks where analytic modeling is intractable: it learns a direct input-to-task-state map from a single short random episode and requires no manual PID tuning. The paper also contains a useful conceptual discussion of control-input/state spaces for musculoskeletal versus axis-driven humanoids. However, the current evidence is limited to single trajectories without repeated trials, and no held-out prediction accuracy of the learned dynamics model is reported. The strength of the claimed improvement over the PID baselines therefore rests on unquantified generalization and run-to-run variability, which the paper does not address.

major comments (4)
  1. [III-B and IV-C] The training set is a single 60-second Random-control episode (300 samples), and the only validation mentioned is 'we use the 1/5 of the data as test data' with no numerical result reported. No train/test prediction error for DDC-Net is given anywhere, so there is no evidence that f generalizes beyond the single training trajectory. Since N=30 gives an input dimension of 34 and output dimension of 30, and the roughly 270 training windows are heavily overlapping, a 5-layer network could easily overfit this data. Please report the held-out prediction error of DDC-Net and, ideally, evaluate the predicted state trajectories on a separately collected random-control episode or on a portion of the control range not used in training.
  2. [IV-C and Fig. 10] The headline convergence times (0.9 s for 5 km/h within 20%, 5.3 s for 10 km/h within 10%) are single observations from what appears to be a single experimental run for each target. The PID1 and PID2 results in Figs. 7 and 8 are also single trajectories. Without repeated trials, error bars, or another measure of variability, the claimed improvement over PID1/PID2 cannot be distinguished from run-to-run variation. At minimum, please run several repetitions of each condition (Proposed, PID1, PID2) and report the mean and spread of T_conv for each criterion.
  3. [IV-B] The convergence criterion is asymmetric and chosen post hoc: A=20% for v_target=5 km/h but A=10% for v_target=10 km/h, with the stated justification that tracking 5 km/h is difficult. This makes the comparison between Proposed and PID1/PID2 not apples-to-apples: PID2's 10.7 s to the 20% threshold at 5 km/h is compared with Proposed's 0.9 s to the same threshold, while at 10 km/h the threshold changes to 10%. Please use a common threshold (or report both thresholds for every method) and fix the evaluation criterion before comparing methods.
  4. [III-C] The optimization procedure can produce control inputs far outside the distribution of the random-control training data. The random controller perturbs u by only -1 to +2 degrees per step (Section IV-C), while the first-stage optimization uses beta1=3.0 degrees for N1=10 iterations with a batch of noisy initializations, and the only constraint on u is clamping to [0,50] degrees. Since the accuracy of f is never evaluated outside the observed input range, the optimized u_seq may exploit extrapolation error rather than true dynamics. Please provide an analysis or experiment showing that the optimized control inputs remain within (or are robust beyond) the range spanned by the training data, for example by plotting the distribution of optimized u_t during the control phase against the training distribution.
minor comments (4)
  1. [III-C, Eq. (2)] The adjacent-error term in Eq. (2) is written with u_init^seq, although the quantity being optimized is the target control input sequence; using the same symbol for the initial guess and the optimized variable is confusing.
  2. [Fig. 9] The caption of Fig. 9 appears truncated ('Target joint angle of ankle-p'), and the target velocity used in the Random controller of Eq. (11)-(12) is not specified; please state whether the data-collection episode targeted 5 km/h, 10 km/h, or a time-varying target.
  3. [IV-C] The random controller equations are referenced in the text as Eq. (11) and Eq. (12) but are not numbered in the manuscript; please number them.
  4. [V] The statement 'we only had to tune alpha after setting all parameters' is inconsistent with the many hand-chosen hyperparameters (N, hidden-layer sizes, beta1, beta2, Nbatch, N1, N2, delta_ubatch). A sensitivity study of at least alpha and beta would support the claim that the method is easy to deploy.

Circularity Check

0 steps flagged · score 1.0 of 10

No substantive circularity: DDC-Net is trained on random driving data and target velocities enter only through the control-phase loss, so the reported convergence is an evaluated outcome rather than a construction.

full rationale

The derivation chain is: Eq. (1) defines DDC-Net as a time-series map from initial state and control sequence to predicted task states; Section III-B fits it by MSE on 300 samples of random-control driving data; Section III-C optimizes the control sequence by backpropagating the error between the network output and the intended target state, then executes the first command on the real robot. The target velocities 5/10 km/h appear only in the control-phase loss, not in the training labels, so the convergence times in Fig. 10 are measurements of a closed-loop experiment, not quantities baked into the fit. Hand-tuned hyperparameters (alpha, beta1, beta2) and the fact that training and evaluation use the same robot and task are weaknesses in generalization evidence, but they do not make the prediction equivalent to an input. The only same-author citations used in the execution pipeline are [7], [8], and [12] for joint-muscle mapping; [8] is used concretely to convert joint angle to muscle length in Section III-C, but the central DDC-Net claim does not reduce to that mapping, and [8] was itself tested in its own experimental setting. No equation defines a predicted quantity in terms of itself, no fitted parameter is renamed as a prediction, and no uniqueness claim is imported from the authors' prior work. Accordingly no circular step can be exhibited, and the appropriate finding is a minor self-citation at most.

Assumptions & free parameters 10 free parameters · 6 assumptions · 0 invented entities

The method introduces no invented physical entities. Its central dependency is a learned neural model trained on 300 samples, plus a set of hand-chosen hyperparameters that affect smoothness, convergence speed, and optimization behavior. The prior joint-to-muscle conversion and CAN-based velocity measurement are adopted from previous work without revalidation here.

free parameters (10)
  • N (time-series horizon) = 30
    Hand-chosen horizon for the sequence of control inputs and predicted states; affects planning depth and computational cost.
  • Hidden layer sizes = [80, 50, 20]
    Hand-chosen architecture for the 5-layer fully connected network; no architecture search is reported.
  • alpha (adjacent error weight) = 30.0
    Hand-tuned weight for the smoothness penalty on control inputs; the paper states alpha was the only parameter tuned after setting the others.
  • beta1 = 3.0 [deg]
    Hand-chosen learning rate for the first optimization stage in the control phase.
  • beta2 = 0.5 [deg]
    Hand-chosen smaller learning rate for the second optimization stage in the control phase.
  • Nbatch = 10
    Hand-chosen number of random-noise perturbed control sequences used in the first optimization stage.
  • N1 and N2 = 10 and 20
    Hand-chosen iteration counts for the two optimization stages; larger values would increase computation.
  • delta_ubatch = 5 [deg]
    Hand-chosen amplitude of random noise added to initial control inputs during the first optimization stage.
  • umin and umax = 0 and 50 [deg]
    Hand-chosen bounds for the ankle joint angle control input, used to clip optimized inputs.
  • Convergence thresholds A = 20% at 5 km/h, 10% at 10 km/h
    Hand-chosen evaluation thresholds used to define convergence time; the paper admits slow target tracking is difficult and sets a looser threshold at 5 km/h.
assumptions (6)
  • domain assumption The dynamics from ankle angle control input to car velocity are stationary between the 60-second random training phase and later control experiments.
    DDC-Net is trained once and then used without online adaptation; any change in vehicle or robot response invalidates the optimized inputs.
  • domain assumption 300 samples at 5 Hz from one random-driving bout sufficiently cover the dynamics of the 0-50 deg input range.
    The network is trained on this single dataset (Section III.B) and selected by a 1/5 held-out split from the same bout; no external validation is described.
  • domain assumption The joint-to-muscle conversion from prior work [8] maps commanded ankle angles to muscle lengths accurately enough for real-time execution.
    The paper sends muscle length commands without joint angle sensors (end of Section III.C) and does not revalidate this mapping.
  • domain assumption Car velocity measured from CAN-USB is an accurate representation of the task state.
    The control experiments use this measurement for both training and evaluation (Section IV.A).
  • domain assumption A 5-layer MLP with batch normalization and sigmoid activations can represent the required input-state dynamics f.
    No expressivity or convergence analysis is given; the network is trained with MSE and Adam (Section III.B).
  • domain assumption Backpropagation through the learned network provides useful gradients for optimizing the control input sequence in real time.
    The control phase (Section III.C) assumes gradients of the loss with respect to u_init_seq guide the search to valid inputs; no stability proof is provided, and stability analysis is deferred to future work.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Task-specific Self-body Controller Acquisition by Musculoskeletal Humanoids: Application to Pedal Control in Autonomous Driving." pith.science (2026). https://pith.science/paper/MSVCOZ65

@misc{pith2026241208270,
  author       = {Pith},
  title        = {Pith review of: Task-specific Self-body Controller Acquisition by Musculoskeletal Humanoids: Application to Pedal Control in Autonomous Driving},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/MSVCOZ65}},
  note         = {Machine review of arXiv:2412.08270}
}
read the original abstract

The musculoskeletal humanoid has many benefits that human beings have, but the modeling of its complex flexible body is difficult. Although we have developed an online acquisition method of the nonlinear relationship between joints and muscles, we could not completely match the actual robot and its self-body image. When realizing a certain task, the direct relationship between the control input and task state needs to be learned. So, we construct a neural network representing the time-series relationship between the control input and task state, and realize the intended task state by applying the network to a real-time control. In this research, we conduct accelerator pedal control experiments as one application, and verify the effectiveness of this study.

Figures

Figures reproduced from arXiv: 2412.08270 by the authors.

Figure 1
Figure 1. Autonomous driving by the musculoskeletal humanoid, Musashi. [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 3
Figure 3. In this comparison, we define control input and state [PITH_FULL_IMAGE:figures/full_fig_p001_3.png] view at source ↗
Figure 3
Figure 3. Comparison of robot controls between musculoskeletal humanoids [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗
Figures from the paper (7 more)
Figure 4
Figure 4. Figure 4: Comparison of autonomous driving between using musculoskeletal [PITH_FULL_IMAGE:figures/full_fig_p002_4.png]
Figure 5
Figure 5. Figure 5: Overview of the task-specific self-body controller acquisition method. [PITH_FULL_IMAGE:figures/full_fig_p003_5.png]
Figure 6
Figure 6. Figure 6: Experimental setup of COMS. [deg], α = 30.0, β1 = 3.0 [deg], and β2 = 0.5 [deg]. In this study, we use a joint angle as the control input. Therefore, after we convert the joint angle to the muscle length by [8], we send it to the actual robot. There is no need of joint…
Figure 7
Figure 7. Figure 7: The result of the basic PID control (PID1). 0 2 4 6 8 10 12 0 5 10 15 20 25 30 35 Car velocity [km/h] Time [sec] 0 2 4 6 8 10 12 0 5 10 15 20 25 30 35 Car velocity [km/h] Time [sec] Current state Target state Transition of car velocity (𝑣𝑡𝑎𝑟𝑔𝑒𝑡 = 5 km/h) Transition of …
Figure 8
Figure 8. Figure 8: The result of the PID control with acceleration estimation ( [PITH_FULL_IMAGE:figures/full_fig_p004_8.png]
Figure 9
Figure 9. Figure 9: The result of the random controller ( Tiif li( [PITH_FULL_IMAGE:figures/full_fig_p005_9.png]
Figure 10
Figure 10. Figure 10: The result of the task-specific self-body controller acquisition [PITH_FULL_IMAGE:figures/full_fig_p005_10.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

22 extracted references · 18 canonical work pages

  1. [15]

    Dynamic Manipulation of Flexible Objects with Torque Sequence Using a Deep Neural Network,

    K. Kawaharazuka, T. Ogawa, J. Tamura, and C. Nabeshima, “Dynamic Manipulation of Flexible Objects with Torque Sequence Using a Deep Neural Network,” in Proceedings of the 2019 IEEE International Conference on Robotics and Automation , 2019, pp. 2139–2145

    work page 2019

  2. [1]

    Toward Anthropomimetic Robotics: Development, Simulation, and Control of a Musculoskeletal Torso,

    S. Wittmeier, C. Alessandro, N. Bascarevic, K. Dalamagkidis, D. De- vereux, A. Diamond, M. J ¨antsch, K. Jovanovic, R. Knight, H. G. Marques, P. Milosavljevic, B. Mitra, B. Svetozarevic, V . Potkonjak, R. Pfeifer, A. Knoll, and O. Holland, “Toward Anthropomimetic Robotics: Development, Simulation, and Control of a Musculoskeletal Torso,” Artificial Life, ...

    work page 2013

  3. [2]

    Design Approach of Biologically-Inspired Musculoskeletal Humanoids,

    Y . Nakanishi, S. Ohta, T. Shirai, Y . Asano, T. Kozuki, Y . Kake- hashi, H. Mizoguchi, T. Kurotobi, Y . Motegi, K. Sasabuchi, J. Urata, K. Okada, I. Mizuuchi, and M. Inaba, “Design Approach of Biologically-Inspired Musculoskeletal Humanoids,” International Journal of Advanced Robotic Systems , vol. 10, no. 4, pp. 216–228, 2013

    work page 2013

  4. [3]

    Human Mimetic Musculoskele- tal Humanoid Kengoro toward Real World Physically Interactive Ac- tions,

    Y . Asano, T. Kozuki, S. Ookubo, M. Kawamura, S. Nakashima, T. Katayama, Y . Iori, H. Toshinori, K. Kawaharazuka, S. Makino, Y . Kakiuchi, K. Okada, and M. Inaba, “Human Mimetic Musculoskele- tal Humanoid Kengoro toward Real World Physically Interactive Ac- tions,” in Proceedings of the 2016 IEEE-RAS International Conference on Humanoid Robots , 2016, pp. 876–883

    work page 2016

  5. [4]

    Anthrob - A Printed Anthropomimetic Robot,

    M. J ¨antsch, S. Wittmeier, K. Dalamagkidis, A. Panos, F. V olkart, and A. Knoll, “Anthrob - A Printed Anthropomimetic Robot,” in Proceedings of the 2013 IEEE-RAS International Conference on Humanoid Robots, 2013, pp. 342–347

    work page 2013

  6. [5]

    Motion Generation of Redundant Musculoskeletal Humanoid Based on Robot-Model Error Compensation by Muscle Load Sharing and Interactive Control Device,

    Y . Asano, T. Shirai, T. Kozuki, Y . Motegi, Y . Nakanishi, K. Okada, and M. Inaba, “Motion Generation of Redundant Musculoskeletal Humanoid Based on Robot-Model Error Compensation by Muscle Load Sharing and Interactive Control Device,” in Proceedings of the 2013 IEEE-RAS International Conference on Humanoid Robots, 2013, pp. 336–341

    work page 2013

  7. [6]

    Antagonist Inhibition Control in Redundant Tendon- driven Structures Based on Human Reciprocal Innervation for Wide Range Limb Motion of Musculoskeletal Humanoids,

    K. Kawaharazuka, M. Kawamura, S. Makino, Y . Asano, K. Okada, and M. Inaba, “Antagonist Inhibition Control in Redundant Tendon- driven Structures Based on Human Reciprocal Innervation for Wide Range Limb Motion of Musculoskeletal Humanoids,” IEEE Robotics and Automation Letters , vol. 2, no. 4, pp. 2119–2126, 2017

    work page 2017

  8. [7]

    Online Learning of Joint-Muscle Mapping using Vision in Tendon-driven Musculoskeletal Humanoids,

    K. Kawaharazuka, S. Makino, M. Kawamura, Y . Asano, K. Okada, and M. Inaba, “Online Learning of Joint-Muscle Mapping using Vision in Tendon-driven Musculoskeletal Humanoids,” IEEE Robotics and Automation Letters, vol. 3, no. 2, pp. 772–779, 2018

    2018

Show all 22 references
  1. [8]

    Online Self-body Image Acquisition Con- sidering Changes in Muscle Routes Caused by Softness of Body Tissue for Tendon-driven Musculoskeletal Humanoids,

    K. Kawaharazuka, S. Makino, M. Kawamura, A. Fujii, Y . Asano, K. Okada, and M. Inaba, “Online Self-body Image Acquisition Con- sidering Changes in Muscle Routes Caused by Softness of Body Tissue for Tendon-driven Musculoskeletal Humanoids,” in Proceedings of the 2018 IEEE /RSJ...

    2018

  2. [9]

    Component Modularized Design of Musculoskeletal Hu- manoid Platform Musashi to Investigate Learning Control Systems,

    K. Kawaharazuka, S. Makino, K. Tsuzuki, M. Onitsuka, Y . Nagamatsu, K. Shinjo, T. Makabe, Y . Asano, K. Okada, K. Kawasaki, and M. Inaba, “Component Modularized Design of Musculoskeletal Hu- manoid Platform Musashi to Investigate Learning Control Systems,” in Proceedings of th...

    2019

  3. [10]

    The Development of Honda Humanoid Robot,

    K. Hirai, M. Hirose, Y . Haikawa, and T. Takenaka, “The Development of Honda Humanoid Robot,” in Proceedings of the 1998 IEEE Interna- tional Conference on Robotics and Automation , 1998, pp. 1321–1326

    1998

  4. [11]

    DARPA Robotics Challenge,

    “DARPA Robotics Challenge,” http: //archive.darpa.mil/ roboticschallenge/, Accessed: 2021-06-30

    2021

  5. [12]

    Learning Nonlinear Muscle-Joint State Mapping Toward Geometric Model-Free Tendon Driven Musculoskeletal Robots,

    S. Ookubo, Y . Asano, T. Kozuki, T. Shirai, K. Okada, and M. Inaba, “Learning Nonlinear Muscle-Joint State Mapping Toward Geometric Model-Free Tendon Driven Musculoskeletal Robots,” in Proceedings of the 2015 IEEE-RAS International Conference on Humanoid Robots, 2015, pp. 765–770

    2015

  6. [13]

    EMD Net: An Encode- Manipulate-Decode Network for Cloth Manipulation,

    D. Tanaka, S. Arnold, and K. Yamazaki, “EMD Net: An Encode- Manipulate-Decode Network for Cloth Manipulation,” IEEE Robotics and Automation Letters , vol. 3, no. 3, pp. 1771–1778, 2018

    2018

  7. [14]

    SE3-Pose-Nets: Structured Deep Dynamics Models for Visuomotor Planning and Control,

    A. Byravan, F. Leeb, F. Meier, and D. Fox, “SE3-Pose-Nets: Structured Deep Dynamics Models for Visuomotor Planning and Control,” in Proceedings of the 2018 IEEE International Conference on Robotics and Automation, 2018, pp. 3339–3346

    2018

  8. [16]

    Development of life-sized high- power humanoid robot JAXON for real-world use,

    K. Kojima, T. Karasawa, T. Kozuki, E. Kuroiwa, S. Yukizaki, S. Iwaishi, T. Ishikawa, R. Koyama, S. Noda, F. Sugai, S. Nozawa, Y . Kakiuchi, K. Okada, and M. Inaba, “Development of life-sized high- power humanoid robot JAXON for real-world use,” in Proceedings of the 2015 IEEE-...

    2015

  9. [17]

    Steer-by-wire: a study into the bandwidth and force requirements,

    T. van der Sande, I. Besselink, and H. Nijmeijer, “Steer-by-wire: a study into the bandwidth and force requirements,” in Proceedings of the 11th International Symposium on Advanced Vechicle Control , 2012, pp. 1–6

    2012

  10. [18]

    Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift,

    S. Io ffe and C. Szegedy, “Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift,” in Proceed- ings of the 32nd International Conference on Machine Learning, 2015, pp. 448–456

    2015

  11. [19]

    Adam: A Method for Stochastic Optimiza- tion,

    D. P. Kingma and J. Ba, “Adam: A Method for Stochastic Optimiza- tion,” in Proceedings of the 3rd International Conference on Learning Representations, 2015, pp. 1–15

    2015

  12. [20]

    Muscle synergies in neuroscience and robotics: from input-space to task-space perspectives,

    C. Alessandro, I. Delis, F. Nori, S. Panzeri, and B. Berret, “Muscle synergies in neuroscience and robotics: from input-space to task-space perspectives,” Frontiers in Computational Neuroscience, vol. 7, no. 43, pp. 1–16, 2013

    2013

  13. [21]

    Reducing the Dimensionality of Data with Neural Networks,

    G. E. Hinton and R. R. Salakhutdinov, “Reducing the Dimensionality of Data with Neural Networks,” Science, vol. 313, no. 5786, pp. 504– 507, 2006

    2006

  14. [22]

    Long short-term memory,

    S. Hochreiter and J. Schmidhuber, “Long short-term memory,” Neural computation, vol. 9, no. 8, pp. 1735–1780, 1997

    1997

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.