pith. sign in

arxiv: 2606.04718 · v2 · pith:HOGVN6HCnew · submitted 2026-06-03 · 💻 cs.RO · cs.AI

CoRe-MoE: Contrastive Reweighted Mixture of Experts for Multi-Terrain Humanoid Locomotion with Gait Adaptation

Kailun Huang (1) , Zikang Xie (1) , Yanzhe Xie (1) , Panpan Liao (3) , Fanghai Zhang (1) , Yanheng Mai (1) , Wenhao Xu (2) , Yunheng Wang (1)
show 5 more authors
Renjing Xu (1) Haohui Huang (3) ((1) Hong Kong University of Science Technology (Guangzhou) (2) South China Agricultural University (3) Guangdong University of Technology)
This is my paper

Pith reviewed 2026-06-28 06:30 UTC · model grok-4.3

classification 💻 cs.RO cs.AI
keywords mixture of expertshumanoid locomotionreinforcement learninggait transitionterrain adaptationcontrastive learningmulti-terrain navigationzero-shot deployment
0
0 comments X

The pith

A two-stage contrastive MoE framework decouples gait generation from terrain adaptation to unify stable locomotion policies for humanoids.

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

The paper tries to establish that a two-stage reinforcement learning approach can overcome gradient interference between gait transitions and multi-terrain adaptation by first learning a base policy for natural walking and running, then adding a terrain-aware MoE branch whose gating network receives contrastive training. This matters to a sympathetic reader because direct joint training of such skills often fails due to task interference and terrain distribution shifts, while successful unification would let robots handle complex real environments with smooth gait changes. The central mechanism is the contrastive objective that promotes expert specialization and structured terrain representations, with final actions formed by weighted fusion of the base and terrain branches. If correct, the result is a single policy that preserves stable gaits yet adapts foothold placement and balance across varied surfaces.

Core claim

The authors claim that decoupling gait generation from terrain adaptation in a two-stage process, with contrastive training of the gating network in the second-stage terrain-aware MoE branch, produces clear expert specialization that avoids interference; the resulting weighted fusion of the base gait policy and terrain branch yields a policy that maintains natural locomotion while adapting to complex terrains.

What carries the argument

The contrastive objective on the gating network of the terrain-aware MoE branch, which learns structured terrain representations to drive expert specialization.

If this is right

  • The method outperforms baseline approaches in success rate, locomotion stability, and multi-terrain adaptability in simulation.
  • Zero-shot real-robot deployment produces robust walking and running with accurate foothold control and dynamic stability across stairs, slopes, steps, obstacles, and unstructured terrains.
  • Smooth transitions between walking and running gaits are preserved while terrain adaptation occurs.
  • The weighted fusion mechanism keeps the base policy's stable behaviors intact during terrain changes.

Where Pith is reading between the lines

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

  • Similar two-stage decoupling with contrastive gating could extend to other robot policies that combine locomotion with manipulation or navigation skills.
  • The learned terrain representations might support generalization to entirely novel surface types without additional fine-tuning.
  • If expert specialization holds, the approach could scale to higher-dimensional control problems where multiple skills compete during training.

Load-bearing premise

The contrastive objective alone on the gating network will produce clear expert specialization and prevent gradient interference between gait and terrain tasks without extra regularization or constraints.

What would settle it

A direct measurement showing that the MoE experts exhibit similar activation patterns across different terrains rather than distinct specialization, or that the full policy shows no gain over a single-policy baseline on held-out terrain sequences in simulation or real deployment.

Figures

Figures reproduced from arXiv: 2606.04718 by (2) South China Agricultural University, (3) Guangdong University of Technology), Fanghai Zhang (1), Haohui Huang (3) ((1) Hong Kong University of Science, Kailun Huang (1), Panpan Liao (3), Renjing Xu (1), Technology (Guangzhou), Wenhao Xu (2), Yanheng Mai (1), Yanzhe Xie (1), Yunheng Wang (1), Zikang Xie (1).

Figure 1
Figure 1. Figure 1: We propose CoRe-MoE enables smooth, natural, and continuous transitions between walking and running using a single unified policy, while maintaining stable adaptability to complex terrain. In contrast, Hiking in the Wild relies on separately trained walking and running policies that must be manually switched during execution, leading to abrupt transitions, reduced continuity, and noticeably unnatural motio… view at source ↗
Figure 2
Figure 2. Figure 2: Overview of our framework. In Stage 1, we learn a flat-terrain adaptive gait policy using MoE and AMP, enabling stable [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Sensitivity analysis of the action fusion coefficients [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Side-view gait sequence visualization results. CoRe-MoE [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Evolution of Mean Reward and Terrain Curriculum Level [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: T-SNE visualization of MoE gating features across different [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: The multi-terrain sequence demonstrates the robot’s continuous locomotion across six representative terrains, where A-F correspond [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Side-view gait sequence of the humanoid robot in real [PITH_FULL_IMAGE:figures/full_fig_p009_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: The robot is subjected to a sudden external impact while [PITH_FULL_IMAGE:figures/full_fig_p009_9.png] view at source ↗
read the original abstract

Humans primarily rely on walking and running to traverse complex terrains. Similarly, humanoid robots should be able to smoothly transition between walking and running while maintaining natural and stable locomotion. However, unifying gait transition and multi-terrain adaptation within a single policy remains challenging due to gradient interference between tasks and the distribution shift caused by terrain variations. Although Mixture-of-Experts (MoE) architectures can mitigate multi-skill interference, direct joint training often fails to achieve clear expert specialization. To address these challenges, we propose CoRe-MoE, a two-stage reinforcement learning framework that decouples gait generation from terrain adaptation. In the first stage, a stable locomotion policy is learned to produce natural walking and running behaviors with smooth transitions. In the second stage, a terrain-aware MoE branch is introduced, and the gating network is trained with a contrastive objective to learn structured terrain representations and promote expert specialization. The final action is obtained through weighted fusion of the base gait policy and the terrain-aware branch, enabling the policy to preserve stable locomotion while adapting to complex terrains. Extensive simulation results demonstrate that the proposed method outperforms baseline approaches in terms of success rate, locomotion stability, and multi-terrain adaptability. Furthermore, zero-shot deployment on a Unitree G1 humanoid robot validates the effectiveness of our framework, achieving robust walking and running across stairs, slopes, steps, obstacles, and unstructured outdoor terrains while maintaining accurate foothold control and dynamic stability.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 1 minor

Summary. The paper proposes CoRe-MoE, a two-stage reinforcement learning framework for multi-terrain humanoid locomotion that first learns a stable base policy for walking/running gait generation and transitions, then augments it with a terrain-aware Mixture-of-Experts branch whose gating network is trained via a contrastive objective to promote expert specialization and mitigate gradient interference. The final policy fuses the base gait output with the terrain branch. The central claims are that this yields higher success rates, stability, and adaptability than baselines in simulation, plus successful zero-shot transfer to a Unitree G1 robot across stairs, slopes, steps, obstacles, and outdoor terrain while preserving foothold control and dynamic stability.

Significance. If the empirical claims are substantiated with quantitative metrics, ablations, and statistical analysis, the two-stage contrastive-MoE design could offer a practical route to decoupling gait and terrain adaptation in humanoid control, addressing a recognized source of training instability in multi-skill policies. The zero-shot real-robot results, if reproducible, would strengthen the case for sim-to-real transfer in unstructured environments.

major comments (3)
  1. [Abstract] Abstract: the central claim of outperformance over baselines in success rate, locomotion stability, and multi-terrain adaptability is stated without any numerical values, baseline descriptions, ablation results, or statistical details; this absence prevents verification that the two-stage procedure plus contrastive gating objective actually produces the reported gains.
  2. [Method] Method description (two-stage procedure and contrastive objective): no equations, hyper-parameter values, or activation statistics are supplied to show that the contrastive loss on the gating network yields measurable expert specialization or reduces interference between gait and terrain tasks; without such evidence the weakest assumption (that the contrastive term suffices without further regularization) remains untested.
  3. [Experiments] Experiments section: the real-robot validation claim (robust walking/running on Unitree G1 across multiple terrain types) is presented without quantitative metrics, failure-mode analysis, or comparison to the simulation baselines, rendering the zero-shot transfer result impossible to assess for robustness or statistical significance.
minor comments (1)
  1. [Method] Notation for the weighted fusion of base gait policy and terrain-aware branch is introduced without an explicit equation or diagram clarifying the combination rule.

Simulated Author's Rebuttal

3 responses · 1 unresolved

We thank the referee for the constructive feedback, which identifies key areas for improving the clarity and empirical substantiation of our claims. We address each major comment below and outline the planned revisions.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim of outperformance over baselines in success rate, locomotion stability, and multi-terrain adaptability is stated without any numerical values, baseline descriptions, ablation results, or statistical details; this absence prevents verification that the two-stage procedure plus contrastive gating objective actually produces the reported gains.

    Authors: We agree that the abstract would be strengthened by including specific numerical results. In the revised manuscript, we will update the abstract to report key quantitative metrics, such as success rates and stability improvements relative to baselines, drawn from the simulation experiments. revision: yes

  2. Referee: [Method] Method description (two-stage procedure and contrastive objective): no equations, hyper-parameter values, or activation statistics are supplied to show that the contrastive loss on the gating network yields measurable expert specialization or reduces interference between gait and terrain tasks; without such evidence the weakest assumption (that the contrastive term suffices without further regularization) remains untested.

    Authors: We will revise the methods section to explicitly include the equations for the contrastive objective, the hyperparameter values used during training, and activation statistics or gating analysis demonstrating expert specialization and reduced interference. revision: yes

  3. Referee: [Experiments] Experiments section: the real-robot validation claim (robust walking/running on Unitree G1 across multiple terrain types) is presented without quantitative metrics, failure-mode analysis, or comparison to the simulation baselines, rendering the zero-shot transfer result impossible to assess for robustness or statistical significance.

    Authors: We acknowledge the value of quantitative metrics for the real-robot results. The current manuscript relies on qualitative observations and video evidence. We will add a failure-mode analysis based on observed behaviors and note limitations relative to simulation baselines, but quantitative metrics and statistical tests were not collected during hardware deployment. revision: partial

standing simulated objections not resolved
  • Providing quantitative metrics or statistical significance analysis for the real-robot zero-shot transfer experiments, as no such numerical data was recorded during the Unitree G1 deployments.

Circularity Check

0 steps flagged

Empirical RL framework with no derivational chain

full rationale

The paper describes a two-stage reinforcement learning procedure that trains a base gait policy followed by a terrain-aware MoE branch with an added contrastive loss on the gating network. All performance claims rest on simulation benchmarks and zero-shot hardware deployment rather than any closed-form derivation, uniqueness theorem, or parameter fit that is then re-labeled as a prediction. No equations appear in the provided text that would allow a reduction of outputs to inputs by construction, and no self-citations are invoked as load-bearing mathematical facts. The work is therefore self-contained as an empirical engineering contribution.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no explicit free parameters, axioms, or invented entities; the framework implicitly assumes standard RL assumptions (Markov property, reward design) and that contrastive loss induces specialization without further justification.

pith-pipeline@v0.9.1-grok · 5878 in / 1273 out tokens · 28727 ms · 2026-06-28T06:30:46.367007+00:00 · methodology

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

39 extracted references · 9 canonical work pages · 3 internal anchors

  1. [1]

    Pie: Parkour with implicit-explicit learning framework for legged robots.IEEE Robotics and Automation Letters, 2024

    Shixin Luo, Songbo Li, Ruiqi Yu, Zhicheng Wang, Jun Wu, and Qiuguo Zhu. Pie: Parkour with implicit-explicit learning framework for legged robots.IEEE Robotics and Automation Letters, 2024

    2024

  2. [2]

    Dreamwaq++: Obstacle-aware quadrupedal locomotion with resilient multimodal reinforcement learning.IEEE Transactions on Robotics, 42:819–836, 2026

    I Made Aswin Nahrendra, Byeongho Yu, Minho Oh, Dongkyu Lee, Seunghyun Lee, Hyeonwoo Lee, Hyungtae Lim, and Hyun Myung. Dreamwaq++: Obstacle-aware quadrupedal locomotion with resilient multimodal reinforcement learning.IEEE Transactions on Robotics, 42:819–836, 2026

    2026

  3. [3]

    Hiking in the wild: A scalable perceptive parkour framework for humanoids.arXiv preprint arXiv:2601.07718, 2026

    Shaoting Zhu, Ziwen Zhuang, Mengjie Zhao, Kun-Ying Lee, and Hang Zhao. Hiking in the wild: A scalable perceptive parkour framework for humanoids.arXiv preprint arXiv:2601.07718, 2026

    work page arXiv 2026

  4. [5]

    Adapting humanoid locomotion over challenging terrain via two-phase training

    Wenhao Cui, Shengtao Li, Huaxing Huang, Bangyu Qin, Tianchu Zhang, Liang Zheng, Ziyang Tang, Chenxu Hu, NING Yan, Jiahao Chen, et al. Adapting humanoid locomotion over challenging terrain via two-phase training. In8th Annual Conference on Robot Learning, 2024

    2024

  5. [6]

    Beamdojo: Learning agile humanoid locomotion on sparse footholds

    Huayi Wang, Zirui Wang, Junli Ren, Qingwei Ben, Tao Huang, Weinan Zhang, and Jiangmiao Pang. Beamdojo: Learning agile humanoid locomotion on sparse footholds. 2025

    2025

  6. [7]

    Learning humanoid locomotion with perceptive internal model, 2024

    Junfeng Long, Junli Ren, Moji Shi, Zirui Wang, Tao Huang, Ping Luo, and Jiangmiao Pang. Learning humanoid locomotion with perceptive internal model, 2024

    2024

  7. [8]

    Hybrid internal model: Learning agile legged loco- motion with simulated robot response

    Junfeng Long, ZiRui Wang, Quanyi Li, Liu Cao, Jiawei Gao, and Jiangmiao Pang. Hybrid internal model: Learning agile legged loco- motion with simulated robot response. InThe Twelfth International Conference on Learning Representations, 2024

    2024

  8. [9]

    A comprehensive survey on gait analysis: History, parameters, approaches, pose estima- tion, and future work.Artificial Intelligence in Medicine, 129:102314, 2022

    Dimple Sethi, Sourabh Bharti, and Chandra Prakash. A comprehensive survey on gait analysis: History, parameters, approaches, pose estima- tion, and future work.Artificial Intelligence in Medicine, 129:102314, 2022

    2022

  9. [10]

    E. Y . S. Chao.Biomechanics of the Human Gait, pages 225–244. Springer New York, New York, NY , 1986

    1986

  10. [11]

    More: Mixture of residual experts for humanoid lifelike gaits learning on complex terrains, 2025

    Dewei Wang, Xinmiao Wang, Xinzhe Liu, Jiyuan Shi, Yingnan Zhao, Chenjia Bai, and Xuelong Li. More: Mixture of residual experts for humanoid lifelike gaits learning on complex terrains, 2025

    2025

  11. [12]

    Humanoid-gym: Reinforcement learning for humanoid robot with zero-shot sim2real transfer

    Xinyang Gu, Yen-Jen Wang, and Jianyu Chen. Humanoid-gym: Reinforcement learning for humanoid robot with zero-shot sim2real transfer. 2024

    2024

  12. [13]

    Blind bipedal stair traversal via sim-to-real reinforcement learning

    Jonah Siekmann, Kevin Green, John Warila, Alan Fern, and Jonathan Hurst. Blind bipedal stair traversal via sim-to-real reinforcement learning. 2021

    2021

  13. [15]

    Learning humanoid locomotion over challenging terrain

    Ilija Radosavovic, Sarthak Kamat, Trevor Darrell, and Jitendra Malik. Learning humanoid locomotion over challenging terrain. 2024

    2024

  14. [16]

    Adaptive mixtures of local experts.Neural computation, 3(1):79–87, 1991

    Robert A Jacobs, Michael I Jordan, Steven J Nowlan, and Geoffrey E Hinton. Adaptive mixtures of local experts.Neural computation, 3(1):79–87, 1991

    1991

  15. [17]

    Moe-loco: Mixture of experts for multitask locomotion

    Runhan Huang, Shaoting Zhu, Yilun Du, and Hang Zhao. Moe-loco: Mixture of experts for multitask locomotion. 2025

    2025

  16. [18]

    Deep whole-body parkour, 2026

    Ziwen Zhuang, Shaoting Zhu, Mengjie Zhao, and Hang Zhao. Deep whole-body parkour, 2026

    2026

  17. [19]

    Swav: Unsupervised learning of visual features by contrasting cluster assignments

    Mathilde Caron, Ishan Misra, Julien Mairal, Piotr Goyal, Piotr Bo- janowski, and Armand Joulin. Swav: Unsupervised learning of visual features by contrasting cluster assignments. 2020

    2020

  18. [20]

    Now you see that: Learning end-to-end humanoid locomotion from raw pixels, 2026

    Wandong Sun, Yongbo Su, Leoric Huang, Alex Zhang, Dwyane Wei, Mu San, Daniel Tian, Ellie Cao, Finn Yan, Ethan Xie, and Zongwu Xie. Now you see that: Learning end-to-end humanoid locomotion from raw pixels, 2026

    2026

  19. [21]

    Real-world humanoid locomotion with reinforcement learning.Science Robotics, 9(89):eadi9579, 2024

    Ilija Radosavovic, Tete Xiao, Bike Zhang, Trevor Darrell, Jitendra Malik, and Koushil Sreenath. Real-world humanoid locomotion with reinforcement learning.Science Robotics, 9(89):eadi9579, 2024

    2024

  20. [22]

    Blind bipedal stair traversal via sim-to-real reinforcement learning.arXiv preprint arXiv:2105.08328, 2021

    Jonah Siekmann, Kevin Green, John Warila, Alan Fern, and Jonathan Hurst. Blind bipedal stair traversal via sim-to-real reinforcement learning.arXiv preprint arXiv:2105.08328, 2021

    work page arXiv 2021

  21. [23]

    Pvp: Data-efficient humanoid robot learning with proprioceptive-privileged contrastive representations.arXiv preprint arXiv:2512.13093, 2025

    Mingqi Yuan, Tao Yu, Haolin Song, Bo Li, Xin Jin, Hua Chen, and Wenjun Zeng. Pvp: Data-efficient humanoid robot learning with proprioceptive-privileged contrastive representations.arXiv preprint arXiv:2512.13093, 2025

    work page arXiv 2025

  22. [24]

    Whole-body humanoid mpc: Realtime physics-based procedural loco-manipulation planning and control

    Manuel Yves Galliker. Whole-body humanoid mpc: Realtime physics-based procedural loco-manipulation planning and control. https://github.com/1x-technologies/wb_humanoid_mpc, 2024

    2024

  23. [25]

    Mpc for humanoid gait generation: Stability and feasibility

    Nicola Scianca, Daniele De Simone, Leonardo Lanari, and Giuseppe Oriolo. Mpc for humanoid gait generation: Stability and feasibility. IEEE Transactions on Robotics, 36(4):1171–1188, 2020

    2020

  24. [26]

    Model predic- tive control of legged and humanoid robots: models and algorithms

    Sotaro Katayama, Masaki Murooka, and Yuichi Tazaki. Model predic- tive control of legged and humanoid robots: models and algorithms. Advanced Robotics, 37(5):298–315, 2023

    2023

  25. [27]

    Gait-adaptive perceptive humanoid locomotion with real time under-base terrain reconstruction

    Haolin Song, Hongbo Zhu, Tao Yu, Yan Liu, Mingqi Yuan, Wengang Zhou, Hua Chen, and Houqiang Li. Gait-adaptive perceptive humanoid locomotion with real time under-base terrain reconstruction. 2025

    2025

  26. [28]

    Extreme parkour with legged robots

    Xuxin Cheng, Kexin Shi, Ananye Agarwal, and Deepak Pathak. Extreme parkour with legged robots. In2024 IEEE International Conference on Robotics and Automation (ICRA), pages 11443–11450. IEEE, 2024

    2024

  27. [29]

    Rudin, J

    Nikita Rudin, Junzhe He, Joshua Aurand, and Marco Hutter. Parkour in the wild: Learning a general and extensible agile locomotion policy using multi-expert distillation and rl fine-tuning.arXiv preprint arXiv:2505.11164, 2025

    work page arXiv 2025

  28. [30]

    Attention-based map encoding for learning generalized legged locomotion.Science Robotics, 10(105):eadv3604, 2025

    Junzhe He, Chong Zhang, Fabian Jenelten, Ruben Grandia, Moritz Bächer, and Marco Hutter. Attention-based map encoding for learning generalized legged locomotion.Science Robotics, 10(105):eadv3604, 2025

    2025

  29. [31]

    Gallant: V oxel grid-based humanoid locomotion and local-navigation across 3d constrained terrains.arXiv preprint arXiv:2511.14625, 2025

    Qingwei Ben, Botian Xu, Kailin Li, Feiyu Jia, Wentao Zhang, Jingping Wang, Jingbo Wang, Dahua Lin, and Jiangmiao Pang. Gallant: V oxel grid-based humanoid locomotion and local-navigation across 3d constrained terrains.arXiv preprint arXiv:2511.14625, 2025

    work page arXiv 2025

  30. [32]

    Learning humanoid locomotion with perceptive internal model

    Junfeng Long, Junli Ren, Moji Shi, Zirui Wang, Tao Huang, Ping Luo, and Jiangmiao Pang. Learning humanoid locomotion with perceptive internal model. 2025

    2025

  31. [33]

    Zhuang, S

    Ziwen Zhuang, Shenzhe Yao, and Hang Zhao. Humanoid parkour learning.arXiv preprint arXiv:2406.10759, 2024

    work page arXiv 2024

  32. [34]

    Jordan and Robert A

    Michael I. Jordan and Robert A. Jacobs. Hierarchical mixtures of experts and the em algorithm.Neural Computation, 6(2):181–214, 1994

    1994

  33. [35]

    A Simple Framework for Contrastive Learning of Visual Representations

    Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. A simple framework for contrastive learning of visual representations.arXiv preprint arXiv:2002.05709, 2020

    work page internal anchor Pith review Pith/arXiv arXiv 2002

  34. [36]

    Proximal Policy Optimization Algorithms

    John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. Proximal policy optimization algorithms.arXiv preprint arXiv:1707.06347, 2017

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  35. [37]

    Amp: Adversarial motion priors for stylized physics-based character control.ACM Transactions on Graphics (ToG), 40(4):1–20, 2021

    Xue Bin Peng, Ze Ma, Pieter Abbeel, Sergey Levine, and Angjoo Kanazawa. Amp: Adversarial motion priors for stylized physics-based character control.ACM Transactions on Graphics (ToG), 40(4):1–20, 2021

    2021

  36. [38]

    Isaac Lab: A GPU-Accelerated Simulation Framework for Multi-Modal Robot Learning

    Mayank Mittal, Pascal Roth, James Tigue, et al. Isaac lab: A gpu- accelerated simulation framework for multi-modal robot learning. arXiv preprint arXiv:2511.04831, 2025

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  37. [39]

    Amass: Archive of motion capture as surface shapes

    Naureen Mahmood, Nima Ghorbani, Nikolaus F Troje, Gerard Pons- Moll, and Michael J Black. Amass: Archive of motion capture as surface shapes. InProceedings of the IEEE/CVF international conference on computer vision, pages 5442–5451, 2019

    2019

  38. [40]

    Harvey, Mike Yurick, Derek Nowrouzezahrai, and Christo- pher Pal

    Félix G. Harvey, Mike Yurick, Derek Nowrouzezahrai, and Christo- pher Pal. Robust motion in-betweening. 39(4), 2020

    2020

  39. [41]

    Cieslak, Ann M

    Matthew C. Cieslak, Ann M. Castelfranco, Vittoria Roncalli, Petra H. Lenz, and Daniel K. Hartline. t-distributed stochastic neighbor em- bedding (t-sne): A tool for eco-physiological transcriptomic analysis. Marine Genomics, 51:100723, 2020

    2020