Pith. sign in

REVIEW 2 major objections 5 minor 34 references

Learning Context-Aware Motion Priors for Humanoid Control

T0 review · 2 major / 5 minor · reviewed 2026-08-05 · deepseek-v4-flash

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

Pith's one-line read A motion prior that reweights references by task context improves humanoid control on five tasks.

desk verdict A genuine, modest advance in making motion priors context-aware, with good ablations and honest limitations; the main soft spot is that the advantage-based relevance signal partly feeds on the adapter's own style reward, and the paper would benefit from a cleaner separation and code release. read the letter →

arxiv 2608.03234 v1 pith:YSHRYHL5 submitted 2026-08-04 cs.RO

Yunyang Mo , Yi Gu , Yangchen Zhou , Hanyang Cao , Renjing Xu This is my paper · ORCID
classification cs.RO
keywords motionpriorshumanoidcontrolreinforcementlearningcontext-awareadaptationcontrastiverelevancereferencereweightingadversarialscore-matching
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 establish that a general motion prior can be adapted online to the current task context without skill labels, dataset partitioning, or a separate skill-discovery stage. Existing priors treat every reference motion as equally relevant, so heterogeneous or imbalanced motion data can mislead the policy. The proposed framework, Context-Aware Motion Priors (CMP), learns context–motion compatibility from high-advantage policy rollouts and uses the resulting weights to reweight reference supervision for a lightweight adapter. Across five humanoid tasks and two motion-prior formulations (adversarial and score-matching), CMP reports consistent gains in task return and sample efficiency. If true, this turns motion-prior learning from a static, task-agnostic step into a continuously context-conditioned mechanism.

What carries the argument

The load-bearing object is the contrastive relevance function $R_\phi(c,x)$, the cosine similarity between a learned context encoder and a learned motion encoder. It is trained with two contrastive branches: an online branch that treats motion clips from rollouts with standardized advantage $>0$ as positives for the current context, weighted by the advantage magnitude, and a demonstration-positive branch that treats reference motions as positives to keep context embeddings near the reference support. The relevance scores become normalized importance weights $w_\phi(c,x)$, which reweight the reference distribution as $q_\phi(x\mid c) = w_\phi(c,x)\,p_E(x)$. A context-conditioned residual adap

What would settle it

On the Target Location task, deliberately bias the critic (for example, initialize the value function so that walking-rich states are overvalued) and compare CMP-AMP's learned weights against a well-calibrated critic. If the weights still shift toward running for far targets and the return gain persists, the advantage-positivity assumption is not decisive; if the weights track the bias instead, the critic error is the load-bearing channel. A second check: after training, hold the policy fixed and see whether reweighting references by the learned $R_\phi$ alone improves task return when the bas

Watch

Extended reading notes

Core claim

The central claim is that task relevance and motion plausibility are distinct, and that the reference distribution of a motion prior should therefore be conditioned on the task context. CMP learns a relevance function $R_\phi(c,x)$ between a task context (goal, command, upcoming trajectory, or object state) and a reference motion clip, trained with two contrastive branches: high-advantage Generalized Advantage Estimation rollouts as positive context–motion pairs, and a demonstration-positive branch that anchors the context embedding to the reference support. These scores define an effective context-conditioned reference distribution $q_\phi(x\mid c) = w_\phi(c,x)\,p_E(x)$, with weights norma

Load-bearing premise

The method assumes that a positive advantage signal from the current policy reliably marks the motion seen in that rollout as the right reference motion for that task context; if the value estimate is biased or the policy never tries the needed motion, the learned relevance will reinforce what the policy already does rather than discovering the right motions.

Editorial extensions

If this is right

  • On all five tasks (Target Location, Steering, Trajectory Following, Dodgeball, Dribbling), CMP reports higher test return than the corresponding base prior, and reaches 80% of the base final return with fewer environment interactions on four of five AMP tasks and on all five SMP tasks.
  • The learned relevance weights evolve from near-uniform early in training to context-dependent late in training—long walking for near targets, fast running for far targets, directional runs for side and back targets—showing interpretable context–motion alignment.
  • Removing the demonstration-positive branch degrades retrieval quality and slows learning; the branch anchors context queries to the reference support.
  • When walking motions are replicated 100x in the reference set, AMP's return drops 11.5% while CMP-AMP drops at most 2.8%, indicating reduced sensitivity to reference-frequency imbalance.
  • Because CMP is modular and leaves the base prior objective unchanged, the same context-conditioned reweighting can be layered onto other motion-prior formulations beyond AMP and SMP.

Reading between the lines

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

  • Editorial inference: the advantage-positivity signal is a generic 'relevance from experience' mechanism; the same weighting could steer which demonstrations to replay or emphasize in imitation learning and offline RL, where hand-designed priorities are common.
  • Editorial inference: the paper's stated sensitivity to critic error and insufficient exploration suggests a direct stress test—corrupt the value function or shrink the exploration noise and measure whether CMP's gains shrink or reverse; this would quantify how much of the benefit comes from the advantage signal versus the demonstration anchor.
  • Editorial inference: the smaller SMP gains suggest adapter capacity limits how much a frozen prior can be reshaped; increasing the residual scale or adapter size for SMP may narrow the gap with AMP, at the risk of overriding the pretrained denoiser.
  • Editorial inference: the reweighting view is a change of measure on the reference distribution, so the same mathematics could schedule a curriculum—using advantage-weighted reference probabilities to sample easier motions early and harder motions as the policy improves.
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

2 major / 5 minor

Summary. The paper introduces Context-Aware Motion Priors (CMP), a modular method that adapts a task-agnostic motion prior to the current task context. CMP learns a contrastive relevance function R_phi(c,x) from high-advantage policy rollouts (positive pairs) and a demonstration-positive objective that anchors context embeddings to the reference distribution. The learned relevance scores reweight the reference motion distribution and are used to train a lightweight context-conditioned adapter on top of the base prior, without modifying the base prior's own training objective. CMP is instantiated with Adversarial Motion Priors (AMP) and Score-Matching Motion Priors (SMP). Experiments on five simulated humanoid tasks report improved task return and sample efficiency for CMP over the base priors, interpretable relevance weights and motion retrieval, ablations separating adapter capacity from learned relevance, and robustness to reference-distribution imbalance. The paper also includes results on a second humanoid morphology (G1).

Significance. If the empirical gains hold, CMP is a worthwhile contribution: it addresses a recognized limitation of motion priors—the use of a fixed, context-independent reference distribution—without requiring skill labels, dataset partitioning, or a separate skill-discovery stage. The paper's strengths include evaluation across two distinct prior formulations, five tasks, a second humanoid morphology, and ablations that separate adapter capacity from the learned relevance signal. The reweighting equivalence in Appendix A is a useful formal framing. The main weaknesses are that validation is entirely simulated, only three seeds are used, and the advantage-based relevance labels are drawn from a total reward that includes the very style prior CMP adapts, creating a potential self-referential loop that is not quantified. Overall the idea is novel and the evidence is suggestive, but the core mechanism needs stronger validation before the central claim is fully supported.

major comments (2)
  1. [§3.2 and §3.3] The positive-pair construction in the online branch L_on uses the standardized GAE Â_i of the total reward. In the AMP instantiation, the total reward is 0.5*r_task + 0.5*r_AMP (Table 7), and r_AMP is computed from the adapter-augmented discriminator l_{θ,ψ} that CMP itself is training (§3.3). Thus high-advantage rollouts can be selected because they satisfy the current, adapting style prior rather than because they solve the task. The relevance model then reweights references to reinforce those style preferences, closing a self-referential loop. Section 5 acknowledges sensitivity to critic error and insufficient exploration, but the specific reward-contamination confound is neither analyzed nor ablated. A concrete test would be to train the relevance model using advantages computed only from the task reward (e.g., a critic for r_task) and compare final return and sample efficiency again
  2. [§4.2, Table 1] The claim of consistent gains with SMP is not fully supported by the reported statistics. For Steering, CMP-SMP return is 512±8 vs SMP 496±10; for Dribbling, 466±2 vs 458±5. These differences are within one standard deviation with three seeds. The text states CMP 'delivers consistent gains' across all five tasks and both motion-prior formulations, which overstates the evidence for these two tasks. Please report per-seed values, effect sizes, confidence intervals, or additional seeds, or qualify the claim for the SMP cases where gains are small relative to variance.
minor comments (5)
  1. [§3.2] The notation in the online batch B_on is ambiguous: x_i is described as the 'paired motion' but it is not stated clearly whether these are policy-generated motion clips or reference motions. Given that the contrastive candidates include both, please clarify explicitly.
  2. [§3.1] The minibatch weight formula uses B exp(a R)/sum_j exp(a R) followed by clipping; the text's statement that 'normalization gives the weights unit mean' is true only before clipping. Please state that clipping changes the mean and mention the implications for the reweighting equivalence in Appendix A.
  3. [§4.3, Figure 5] Many entries in the relevance-weight heatmaps are at the clip boundaries (0.50 and 2.00). The paper does not discuss the extent to which the learned weights are saturated. Adding a version without clipping or stating explicitly how saturation affects interpretability would help.
  4. [§4.4, Tables 4-5] The shuffle ablation is described as 'randomly permuting' weights, but the exact permutation procedure (within batch, per context, shared across references?) is not specified. Please define it precisely for reproducibility.
  5. [Appendix J] The samples-to-threshold metric is defined with respect to 80% of the base prior's final return. Because thresholds differ between base and CMP, a CMP method that improves early but has a lower final return could artificially appear better. Please discuss this potential bias, even if not present in the reported results.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central claim is evaluated against external task rewards, and no prediction reduces to a fitted input by construction.

full rationale

CMP's derivation chain is not circular in the sense defined here. The relevance function R_phi is learned from high-advantage rollouts, but the paper's central claim—improved task performance and sample efficiency—is measured on external task returns against the base AMP/SMP priors, not on quantities reconstructed from R_phi's own weights. The reweighting equivalence in Appendix A is a standard importance-weighting identity, not a circular argument. The paper does not rename a known result, does not import a uniqueness theorem, and does not rely on self-citation: its references to AMP, SMP, and MimicKit are external prior work. Section 5 explicitly acknowledges that reliance on policy advantages makes relevance learning sensitive to critic error and insufficient exploration; this is a real limitation and a potential feedback confound, since the adapted style reward contributes to the advantage signal used to label positives. However, the task reward remains an independent component of the total reward, and evaluation is based on task return, so the empirical gains do not reduce to a self-referential artifact by construction. The feedback loop is a training-dynamics concern, not a definitional equivalence between an output and its input.

Assumptions & free parameters 6 free parameters · 3 assumptions · 0 invented entities

CMP introduces a learned relevance function and a residual adapter, which are model components rather than new physical entities. The central empirical claim rests primarily on RL bootstrapping assumptions and a set of hand-tuned hyperparameters; no new object, force, or conserved quantity is postulated.

free parameters (6)
  • Advantage threshold and fallback fraction = A_i > 0; min positives 64, fallback fraction 0.35
    Section 3.2 and Table 8: defines which rollout motions count as positive context-motion pairs. Chosen by hand; no sensitivity analysis.
  • Reweighting sharpness alpha = 0.5 (AMP), 1.0 (SMP)
    Section 3.1, Tables 8 and 10: controls how strongly relevance scores distort the reference distribution; hand-set.
  • Weight clipping range [w_min, w_max] = [0.5, 2.0]
    Section 3.1, Tables 8 and 10: bounds the adapter supervision weights; no sensitivity analysis.
  • Demonstration branch weight lambda_demo = 0.5 (AMP), 1.0 (SMP)
    Section 3.2, Tables 8 and 10: balances online and demonstration objectives; hand-set.
  • Residual adapter strength lambda_res = 0.03 (AMP), 0.1 (SMP)
    Section 3.3, Tables 8 and 10: controls how far the adapter can move the base prior prediction; hand-set.
  • Contrastive temperature tau = 0.1
    Section 3.2: shapes the similarity distribution used in the contrastive objective; hand-set.
assumptions (3)
  • domain assumption GAE advantage from the current policy is a proxy for task-relevance of the paired reference motion.
    Section 3.2 selects samples with A_i > 0 as positive context-motion pairs; if the critic is biased, the learned relevance is biased. Acknowledged in Section 5.
  • domain assumption The reference dataset contains motions that are relevant to each task context encountered.
    Section 3.1 and Section 5: CMP cannot recover behaviors absent from D_E, and the effective distribution q_phi(x|c) is supported on D_E.
  • domain assumption The context observation c_t is a sufficient representation of task-relevant state for motion selection.
    Section 4.1 defines contexts as target position, velocity commands, future waypoints, or ball state; if these omit relevant factors, the relevance function cannot adapt.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Learning Context-Aware Motion Priors for Humanoid Control." pith.science (2026). https://pith.science/paper/YSHRYHL5

@misc{pith2026260803234,
  author       = {Pith},
  title        = {Pith review of: Learning Context-Aware Motion Priors for Humanoid Control},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YSHRYHL5}},
  note         = {Machine review of arXiv:2608.03234}
}
read the original abstract

Motion priors provide powerful guidance for learning naturalistic humanoid behaviors. However, existing methods typically learn a general, task-agnostic prior from the entire reference dataset and apply it uniformly throughout policy training. As a result, the prior cannot distinguish which reference motions are relevant to the current task context, potentially providing irrelevant or conflicting guidance. We present Context-Aware Motion Priors (CMP), a framework that adapts a general motion prior to the current task context without manual skill labels, dataset partitioning, or a separate skill discovery stage. Specifically, CMP learns context-motion compatibility using high-advantage policy rollouts, while a demonstration-based objective keeps the learned relevance grounded in the reference distribution. The resulting relevance scores reweight reference supervision for training a lightweight context-conditioned adapter. To evaluate the effectiveness and generality of CMP, we instantiate it with both Adversarial Motion Priors and Score-Matching Motion Priors. Across five humanoid control tasks, CMP consistently improves task performance and sample efficiency, learns meaningful context-motion alignment, and remains robust to imbalanced reference distributions. These results show that adapting motion priors to task contexts provides more relevant guidance for humanoid policy learning.

Figures

Figures reproduced from arXiv: 2608.03234 by the authors.

Figure 1
Figure 1. Reference motions for reaching a target location. Different reference motions, such as walking, dashing, and turning, may provide more suitable guidance depending on the humanoid’s distance and orientation relative to the target. Motion plausibility and task relevance are not equivalent. As illustrated in [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of CMP. CMP learns context–motion relevance and uses it to train a relevance-weighted context adapter over the base motion prior. 3 Method Context-Aware Motion Priors (CMP) is a modular frame￾work that augments task-agnostic motion priors with context￾dependent reference motion relevance, enabling adaptive use of motion references for different tasks. As illustrated in [PITH_FULL_IMAGE:figures/full_fig_p00… view at source ↗
Figure 3
Figure 3. Qualitative comparison of AMP and CMP-AMP. Temporally aligned snapshots compare CMP-AMP (blue) with AMP (red) on Target Location and Steering. CMP-AMP reaches the target three intervals earlier on Target Location and begins turning and accelerating earlier on Steering. Additional visual comparisons are provided in the supplementary material. 4.2 Main Results: Performance and Efficiency We first evaluate CMP with AMP… view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Learning curves across five humanoid control tasks. Top: CMP instantiated with AMP. Bottom: CMP extended to SMP. All methods use the same task settings and training budgets. Fast Run Forward Run Left Run Right Run Walk Long Walk Mirror Walk 0.50 0.50 0.50 0.50 0.50 0.5…
Figure 5
Figure 5. Figure 5: Evolution of learned relevance weights. Rows represent source motion groups and columns represent task contexts. Group names are used only for visualization and are unavailable to CMP during training. Early, Intermediate, Late, and Final correspond to approximately 1.3…
Figure 7
Figure 7. Figure 7: CMP-AMP ablations. Mean task return with stan￾dard deviation over three seeds on Dodgeball and Dribbling [PITH_FULL_IMAGE:figures/full_fig_p007_7.png]
Figure 6
Figure 6. Figure 6: Demonstration support improves motion re￾trieval. Top: PCA projections of reference and context em￾beddings without Ldemo and with the full CMP objective. Bottom: the highest-ranked reference clips for three repre￾sentative contexts, with two frames shown per clip. but…
Figure 8
Figure 8. Figure 8: G1 humanoid results. Learning curves of AMP and CMP-AMP on five humanoid control tasks. Each curve corresponds to one of three random seeds. 0 1 2 3 4 5 6 Samples (×10 8 ) 0 100 200 300 400 500 Task Return Dodgeball 0 1 2 3 4 5 6 Samples (×10 8 ) 0 100 200 300 400 500 …
Figure 9
Figure 9. Figure 9: CMP-SMP ablations. Learning curves on Dodge￾ball and Dribbling. Solid curves show the mean task return and shaded regions indicate the standard deviation over three seeds. H Learning Curves under Reference Dataset Imbalance [PITH_FULL_IMAGE:figures/full_fig_p012_9.png]
Figure 11
Figure 11. Figure 11: Network architectures used to model the com￾ponents of CMP. Left: the policy, value function, base AMP discriminator, contrastive relevance model, and residual logit adapter used in the primary CMP-AMP instantiation. These components are implemented using fully connec…
Figure 12
Figure 12. Figure 12: Visual overview of the five task implementations. [PITH_FULL_IMAGE:figures/full_fig_p014_12.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

34 extracted references · 23 canonical work pages

  1. [1]

    Proceedings of the 28th Annual Conference on Computer Graphics and Interactive Techniques , pages=

    Composable Controllers for Physics-Based Character Animation , author=. Proceedings of the 28th Annual Conference on Computer Graphics and Interactive Techniques , pages=

    work page

  2. [2]

    Advances in Neural Information Processing Systems , volume=

    Generative Adversarial Imitation Learning , author=. Advances in Neural Information Processing Systems , volume=

    work page

  3. [3]

    2021 , publisher=

    Peng, Xue Bin and Ma, Ze and Abbeel, Pieter and Levine, Sergey and Kanazawa, Angjoo , journal=. 2021 , publisher=

    work page 2021

  4. [4]

    2512.03028 , archivePrefix=

    Mu, Yuxuan and Zhang, Ziyu and Shi, Yi and Yang, Dun and Matsumoto, Minami and Imamura, Kotaro and Tevet, Guy and Guo, Chuan and Taylor, Michael and Shu, Chang and Xi, Pengcheng and Peng, Xue Bin , year=. 2512.03028 , archivePrefix=

    arXiv

  5. [5]

    Proceedings of the Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers , pages=

    Skillmimic-v2: Learning robust and generalizable interaction skills from sparse and noisy demonstrations , author=. Proceedings of the Special Interest Group on Computer Graphics and Interactive Techniques Conference Conference Papers , pages=

    work page

  6. [6]

    2025 , publisher=

    Xu, Michael and Shi, Yi and Yin, KangKang and Peng, Xue Bin , booktitle=. 2025 , publisher=

    work page 2025

  7. [7]

    2018 , publisher=

    Peng, Xue Bin and Kanazawa, Angjoo and Malik, Jitendra and Abbeel, Pieter and Levine, Sergey , journal=. 2018 , publisher=

    2018

  8. [8]

    Proceedings of the Conference on Robot Learning , pages=

    Learning to Walk in Minutes Using Massively Parallel Deep Reinforcement Learning , author=. Proceedings of the Conference on Robot Learning , pages=. 2022 , publisher=

    work page 2022

Show all 34 references
  1. [9]

    2022 , publisher=

    Peng, Xue Bin and Guo, Yunrong and Halper, Lina and Levine, Sergey and Fidler, Sanja , journal=. 2022 , publisher=

    2022

  2. [10]

    Tessler, Chen and Kasten, Yoni and Guo, Yunrong and Mannor, Shie and Chechik, Gal and Peng, Xue Bin , booktitle=

  3. [11]

    Dou, Zhiyang and Chen, Xuelin and Fan, Qingnan and Komura, Taku and Wang, Wenping , booktitle=. C

  4. [12]

    ACM Transactions On Graphics (TOG) , volume=

    Deepmimic: Example-guided deep reinforcement learning of physics-based character skills , author=. ACM Transactions On Graphics (TOG) , volume=. 2018 , publisher=

    2018

  5. [13]

    Proceedings of the 26th annual international conference on machine learning , pages=

    Curriculum learning , author=. Proceedings of the 26th annual international conference on machine learning , pages=

  6. [14]

    International Conference on Learning Representations , year=

    Prioritized Experience Replay , author=. International Conference on Learning Representations , year=

  7. [15]

    Parameter-Efficient Transfer Learning for

    Houlsby, Neil and Giurgiu, Andrei and Jastrzebski, Stanislaw and Morrone, Bruna and de Laroussilhe, Quentin and Gesmundo, Andrea and Attariyan, Mona and Gelly, Sylvain , booktitle=. Parameter-Efficient Transfer Learning for. 2019 , publisher=

    2019

  8. [16]

    Proceedings of the IEEE/CVF international conference on computer vision , pages=

    Adding conditional control to text-to-image diffusion models , author=. Proceedings of the IEEE/CVF international conference on computer vision , pages=

  9. [17]

    Proceedings of the 38th International Conference on Machine Learning , pages=

    Learning Transferable Visual Models from Natural Language Supervision , author=. Proceedings of the 38th International Conference on Machine Learning , pages=. 2021 , publisher=

    2021

  10. [18]

    2018 , eprint=

    Representation Learning with Contrastive Predictive Coding , author=. 2018 , eprint=

    2018

  11. [19]

    Proceedings of the 37th International Conference on Machine Learning , pages=

    A Simple Framework for Contrastive Learning of Visual Representations , author=. Proceedings of the 37th International Conference on Machine Learning , pages=. 2020 , publisher=

    2020

  12. [20]

    Eysenbach, Benjamin and Salakhutdinov, Ruslan and Levine, Sergey , booktitle =

  13. [21]

    Advances in Neural Information Processing Systems , volume=

    Contrastive Learning as Goal-Conditioned Reinforcement Learning , author=. Advances in Neural Information Processing Systems , volume=

  14. [22]

    Stabilizing Contrastive

    Zheng, Chongyi and Eysenbach, Benjamin and Walke, Homer and Yin, Patrick and Fang, Kuan and Salakhutdinov, Ruslan and Levine, Sergey , booktitle=. Stabilizing Contrastive

  15. [23]

    Liao, Qiayuan and Truong, Takara E and Huang, Xiaoyu and Gao, Yuman and Tevet, Guy and Sreenath, Koushil and Liu, C Karen , journal=

  16. [24]

    arXiv preprint arXiv:2511.17373 , year=

    Agility meets stability: Versatile humanoid control with heterogeneous data , author=. arXiv preprint arXiv:2511.17373 , year=

  17. [25]

    Advances in Neural Information Processing Systems , volume=

    Adversarial locomotion and motion imitation for humanoid policy learning , author=. Advances in Neural Information Processing Systems , volume=

  18. [26]

    arXiv preprint arXiv:2506.14770 , year=

    Gmt: General motion tracking for humanoid whole-body control , author=. arXiv preprint arXiv:2506.14770 , year=

    arXiv

  19. [27]

    ACM Transactions On Graphics (TOG) , volume=

    DReCon: data-driven responsive control of physics-based characters , author=. ACM Transactions On Graphics (TOG) , volume=. 2019 , publisher=

    2019

  20. [28]

    ACM Transactions on Graphics (TOG) , volume=

    Environment-aware motion matching , author=. ACM Transactions on Graphics (TOG) , volume=. 2025 , publisher=

    2025

  21. [29]

    Game Developers Conference , year=

    Motion Matching and The Road to Next-Gen Animation , author=. Game Developers Conference , year=

  22. [30]

    International Conference on Learning Representations , year =

    Universal Humanoid Motion Representations for Physics-Based Control , author =. International Conference on Learning Representations , year =

  23. [31]

    2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages=

    Learning a single policy for diverse behaviors on a quadrupedal robot using scalable motion imitation , author=. 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , pages=. 2023 , organization=

    2023

  24. [32]

    arXiv preprint arXiv:2510.13794 , year=

    MimicKit: A Reinforcement Learning Framework for Motion Imitation and Control , author=. arXiv preprint arXiv:2510.13794 , year=

  25. [33]

    ACM Transactions on Graphics (ToG) , volume=

    Learned motion matching , author=. ACM Transactions on Graphics (ToG) , volume=. 2020 , publisher=

    2020

  26. [34]

    Laskin, Michael and Srinivas, Aravind and Abbeel, Pieter , booktitle =

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.