pith. sign in

arxiv: 2605.27724 · v1 · pith:QP3JWNGVnew · submitted 2026-05-26 · 💻 cs.RO · cs.AI

HumanoidMimicGen: Data Generation for Loco-Manipulation via Whole-Body Planning

Kevin Lin , Ajay Mandlekar , Caelan Reed Garrett , Nikita Chernyadev , Yu Fang , Runyu Ding , Yuqi Xie , Justin Tran
show 2 more authors
Linxi Fan Yuke Zhu
This is my paper

Pith reviewed 2026-06-29 16:33 UTC · model grok-4.3

classification 💻 cs.RO cs.AI
keywords humanoid robotsloco-manipulationdata generationimitation learningwhole-body planningvisuomotor policiessimulation benchmark
0
0 comments X

The pith

HumanoidMimicGen generates large sets of stable whole-body loco-manipulation demonstrations from a small number of source examples.

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

Imitation learning for humanoid robots that must walk and manipulate objects requires many demonstrations, yet collecting them through teleoperation is slow and difficult. Existing automatic data generators work for fixed-base arms but fail on humanoids whose high-dimensional actions couple legs, torso, and arms. HumanoidMimicGen adapts contact-rich arm skills from a few demonstrations to new object poses, then interleaves those skills with whole-body locomotion and manipulation planning to produce collision-free, stable trajectories across varied scenes. On a new nine-task simulated benchmark the method yields enough data that whole-body visuomotor policies trained on the mixture of real and generated demonstrations outperform policies trained on real data alone by 20 percent.

Core claim

HumanoidMimicGen adapts contact-rich whole-body skills from a handful of source demonstrations to new states, generalizing across changes in object pose. By interleaving single- and dual-arm skills with whole-body locomotion and manipulation planning, the method generates stable, collision-free data across diverse scenes and layouts. Whole-body visuomotor policies co-trained with this generated data outperform those trained only on real-world data by 20 percent on a new simulated loco-manipulation benchmark containing nine tasks.

What carries the argument

HumanoidMimicGen, a whole-body planning procedure that adapts contact-rich arm skills to new states and interleaves them with locomotion planning to synthesize stable loco-manipulation trajectories.

If this is right

  • Large numbers of loco-manipulation demonstrations become available without additional teleoperation.
  • Whole-body policies can be trained that succeed on tasks requiring coordinated locomotion and manipulation in changing layouts.
  • A systematic comparison of data-generation choices and policy architectures is now possible on the nine-task benchmark.
  • Co-training with automatically generated data improves policy performance beyond what real demonstrations alone achieve.

Where Pith is reading between the lines

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

  • If the generated trajectories transfer to physical robots, the volume of real teleoperation needed for humanoid training could drop substantially.
  • The same adaptation-plus-planning loop might extend to multi-object or long-horizon tasks once the benchmark is expanded.
  • The nine-task benchmark supplies a concrete testbed for measuring how much simulation-to-real gap remains after co-training.
  • Similar data-generation pipelines could be applied to other high-dimensional platforms such as mobile manipulators.

Load-bearing premise

The generated demonstrations are realistic, stable, and diverse enough that adding them to training produces a measurable performance gain over real data alone.

What would settle it

Train identical whole-body visuomotor policies on the nine-task benchmark using only the real demonstrations versus the real demonstrations plus the HumanoidMimicGen data and check whether the success-rate difference remains at or above 20 percent.

Figures

Figures reproduced from arXiv: 2605.27724 by Ajay Mandlekar, Caelan Reed Garrett, Justin Tran, Kevin Lin, Linxi Fan, Nikita Chernyadev, Runyu Ding, Yu Fang, Yuke Zhu, Yuqi Xie.

Figure 1
Figure 1. Figure 1: HumanoidMimicGen Overview. We present HumanoidMimicGen, a method for humanoid legged loco￾manipulation demonstration generation. Left: a human teleoperator collects a loco-manipulation task demonstration. Center: HumanoidMimicGen generates thousands of demonstrations across scene and object layouts by adapting local segments of the human demonstration via whole-body planning and control. Right: these demon… view at source ↗
Figure 2
Figure 2. Figure 2: Method Overview. Left: HumanoidMimicGen takes a source demonstration with per-arm skill annotations (blue arrows) and constraints (orange box). Right: constraints determine execution order and target poses; whole-body planning produces locomotion and arm plans executed sequentially. quadratic programming approach to improving its tracking controller. Task and Motion Planning (TAMP) [19] algorithms have bee… view at source ↗
Figure 3
Figure 3. Figure 3: G1 Loco-Manipulation Benchmark. We introduce a simulation benchmark with nine loco-manipulation tasks and datasets generated by HumanoidMimicGen. Each task is shown with a sampled initial scene (left) and task-completion configuration (right). and the lower joint positions of 𝑞 ′′, where the robot switches from locomotion to manipulation. Then, it plans a locomotion trajectory 𝜏𝑙 between the current config… view at source ↗
Figure 4
Figure 4. Figure 4: Policy and data generation ablations. Left: Policy architecture ablation. VLA outperforms Flow Matching and Diffusion Policy when all are trained on 1,000 HumanoidMimicGen demonstrations. Right: Effect of data generation design on policy success rates. Removing motion noise or initialization noise reduces policy performance, highlighting the importance of these strategies. 6.2. HumanoidMimicGen Capabilitie… view at source ↗
Figure 5
Figure 5. Figure 5: Real-World Deployment. We evaluate policies co-trained on HumanoidMimicGen simulation data and real-world demonstrations across four real-world manipulation tasks: ThrowBottle (top left), BoxToCart (top right), PickCanister (bottom left), and PickCanisterWithObstruction (bottom right). Each left image is the initial state and right image is a goal state. Co-training improves average policy score from 0.51 … view at source ↗
Figure 6
Figure 6. Figure 6: Precedence and Coordination Constraints. Skill planning visualized on the Table-to-Shelf task. A human annotates precedence and coordination constraints among the skills (left). HumanoidMimicGen automatically compiles these constraints into partial orders on the skills, defining legal execution orders (right). Continuing on from Section 4.2, [PITH_FULL_IMAGE:figures/full_fig_p015_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Whole-Body Skill Adaptation. Left: skill 𝜓 for end-effector 𝑒 relative to object frame 𝑓 is adapted to a new state 𝑠. Right: spherical collision representation used for planning and IK configuration 𝑞 ′′ for adapted skill target pose 𝑇[𝑒]. kinematics and motion planning subproblems induced through manipulation, the initial configuration and goal end-effector poses are often in contact with manipulable obje… view at source ↗
read the original abstract

Imitation learning is a promising approach for training humanoid robots to both walk and manipulate, but it requires a large number of demonstrations, which are time-intensive and difficult to collect via teleoperation. Existing data-generation algorithms can automatically synthesize demonstrations for manipulators, but they are ineffective on humanoids because their high-dimensional composite action spaces involve arms, legs, and torsos. We present HumanoidMimicGen, a method for generating humanoid legged loco-manipulation data. Our method adapts contact-rich whole-body skills from a handful of source demonstrations to new states, generalizing across changes in object pose. By interleaving these single- and dual-arm skills with whole-body locomotion and manipulation planning, the method generates stable, collision-free data across diverse scenes and layouts. To evaluate our approach, we introduce a new simulated loco-manipulation benchmark containing nine diverse tasks that test humanoid loco-manipulation capabilities. There, we demonstrate that HumanoidMimicGen automatically generates large datasets for imitation learning and enables a systematic study of how data generation and policy learning decisions impact model performance. We show that whole-body visuomotor policies co-trained with data generated by HumanoidMimicGen outperform those trained only on real-world data by 20%.

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

2 major / 0 minor

Summary. The manuscript presents HumanoidMimicGen, a method for automatically generating large-scale whole-body loco-manipulation demonstrations for humanoids. It adapts contact-rich single- and dual-arm skills from a small set of source demonstrations to new object poses via whole-body planning, then interleaves these with locomotion and manipulation primitives to produce stable, collision-free trajectories across varied scenes. The authors introduce a new simulated benchmark consisting of nine diverse loco-manipulation tasks and report that whole-body visuomotor policies co-trained on real-world data augmented with HumanoidMimicGen data outperform policies trained on real-world data alone by 20%.

Significance. If the reported performance gains prove robust after proper controls, the work would meaningfully advance scalable data generation for humanoid imitation learning, addressing the high cost of teleoperated whole-body demonstrations. The new benchmark is a constructive contribution that could support systematic study of data and policy decisions in loco-manipulation. The core technical idea of combining skill adaptation with interleaved whole-body planning is a reasonable direction for composite action spaces.

major comments (2)
  1. [Abstract / Evaluation] Abstract and evaluation section: the central claim of a 20% performance improvement for co-trained policies is stated without reference to the number of tasks evaluated, the precise baselines, data volumes, error bars, or statistical tests. This detail is load-bearing for assessing whether the result supports the method.
  2. [Benchmark evaluation] Benchmark evaluation: the real-only versus real + HumanoidMimicGen comparison does not report ablations that isolate data volume (e.g., real + random simulated trajectories of matched length) or domain shift (e.g., dynamics or sensor mismatch between real collection and the simulation used for generation). Without these controls the attribution of any lift specifically to the quality and diversity of the generated loco-manipulation sequences remains unverified.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address each major comment below and commit to revisions that strengthen the presentation and evaluation of our results.

read point-by-point responses

  1. Referee: [Abstract / Evaluation] Abstract and evaluation section: the central claim of a 20% performance improvement for co-trained policies is stated without reference to the number of tasks evaluated, the precise baselines, data volumes, error bars, or statistical tests. This detail is load-bearing for assessing whether the result supports the method.

    Authors: We agree that additional detail in the abstract would improve clarity. The evaluation is performed across all nine tasks in the benchmark. The comparison uses real-world data only versus real-world data augmented with HumanoidMimicGen trajectories, with total demonstration counts matched between conditions. We will revise the abstract to explicitly state the nine-task scope, the matched data volumes, and the precise baselines. In the evaluation section we will also add error bars (standard deviation over three random seeds) and note the results of paired statistical tests. revision: yes

  2. Referee: [Benchmark evaluation] Benchmark evaluation: the real-only versus real + HumanoidMimicGen comparison does not report ablations that isolate data volume (e.g., real + random simulated trajectories of matched length) or domain shift (e.g., dynamics or sensor mismatch between real collection and the simulation used for generation). Without these controls the attribution of any lift specifically to the quality and diversity of the generated loco-manipulation sequences remains unverified.

    Authors: We acknowledge that the current manuscript lacks explicit controls for data volume and domain shift. We will add an ablation that augments the real dataset with an equal number of randomly sampled simulated trajectories (without HumanoidMimicGen planning) to isolate the contribution of data volume. Regarding domain shift, HumanoidMimicGen generates trajectories inside a simulation whose dynamics and sensor models were calibrated to the real robot; we will expand the manuscript to quantify residual mismatch via a small set of real-to-sim transfer experiments and discuss this calibration explicitly. These additions will be included in the revised version. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical method with no fitted predictions or self-referential derivations

full rationale

The paper describes an algorithmic pipeline for generating loco-manipulation demonstrations via whole-body planning and reports an empirical 20% policy improvement on a simulated benchmark. No equations, parameter-fitting steps, uniqueness theorems, or ansatzes appear in the provided text. The performance delta is framed as a direct experimental outcome of co-training on real versus generated data rather than any quantity derived from or equivalent to the method's own inputs by construction. The central claim therefore remains externally falsifiable via the benchmark results and does not reduce to self-definition or self-citation.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review supplies no information on free parameters, axioms, or invented entities.

pith-pipeline@v0.9.1-grok · 5781 in / 1083 out tokens · 37979 ms · 2026-06-29T16:33:27.768265+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

62 extracted references · 32 canonical work pages · 10 internal anchors

  1. [1]

    Whole-body motion and footstep planning for humanoid robots with multi-heuristic search.Robotics and Autonomous Systems, 116:51–63,

    Rizwan Asif, Ali Athar, Faisal Mehmood, Fahad Islam, and Yasar Ayaz. Whole-body motion and footstep planning for humanoid robots with multi-heuristic search.Robotics and Autonomous Systems, 116:51–63,

  2. [2]

    doi: 10.1016/j.robot.2019.03.007. 3

    work page doi:10.1016/j.robot.2019.03.007 2019

  3. [3]

    A framework for behavioural cloning

    Michael Bain and Claude Sammut. A framework for behavioural cloning. InMachine intelligence 15, pages 103–129, 1995. 4

    1995

  4. [4]

    A Careful Examination of Large Behavior Models for Multitask Dexterous Manipulation

    Jose Barreiros, Andrew Beaulieu, Aditya Bhat, Rick Cory, Eric Cousineau, Hongkai Dai, Ching-Hsin Fang, Kunimatsu Hashimoto, Muhammad Zubair Irshad, Masha Itkina, et al. A careful examination of large behavior models for multitask dexterous manipulation.arXiv preprint arXiv:2507.05331, 2025. 2

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  5. [5]

    Homie: Humanoid loco-manipulation with isomorphic exoskeleton cockpit,

    Qingwei Ben, Feiyu Jia, Jia Zeng, Junting Dong, Dahua Lin, and Jiangmiao Pang. Homie: Humanoid loco-manipulation with isomorphic exoskeleton cockpit.arXiv preprint arXiv:2502.13013, 2025. 3, 5

    work page arXiv 2025

  6. [6]

    Robot programming by demonstra- tion

    Aude Billard, Sylvain Calinon, Rüdiger Dillmann, and Stefan Schaal. Robot programming by demonstra- tion. InSpringer Handbook of Robotics, 2008. 3

    2008

  7. [7]

    GR00T N1: An Open Foundation Model for Generalist Humanoid Robots

    Johan Bjorck, Fernando Castañeda, Nikita Cherniadev, Xingye Da, Runyu Ding, Linxi Fan, Yu Fang, Dieter Fox, Fengyuan Hu, Spencer Huang, et al. Gr00t n1: An open foundation model for generalist humanoid robots.arXiv preprint arXiv:2503.14734, 2025. 1, 3, 7, 19

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  8. [8]

    Kevin Black, Noah Brown, Danny Driess, Adnan Esmail, Michael Equi, Chelsea Finn, Niccolo Fusai, Lachy Groom, Karol Hausman, Brian Ichter, et al.𝜋0: A vision-language-action flow model for general robot control.arXiv preprint arXiv:2410.24164, 2024. 1, 3

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  9. [9]

    RT-1: Robotics Transformer for Real-World Control at Scale

    Anthony Brohan, Noah Brown, Justice Carbajal, Yevgen Chebotar, Joseph Dabis, Chelsea Finn, Keerthana Gopalakrishnan, Karol Hausman, Alex Herzog, Jasmine Hsu, et al. Rt-1: Robotics transformer for real-world control at scale.arXiv preprint arXiv:2212.06817, 2022. 1

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  10. [10]

    RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Control

    Anthony Brohan, Noah Brown, Justice Carbajal, Yevgen Chebotar, Xi Chen, Krzysztof Choromanski, Tianli Ding, Danny Driess, Avinava Dubey, Chelsea Finn, et al. Rt-2: Vision-language-action models transfer web knowledge to robotic control.arXiv preprint arXiv:2307.15818, 2023. 3

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  11. [11]

    Whole-body motion planning for manipulation of articulated objects

    Felix Burget, Armin Hornung, and Maren Bennewitz. Whole-body motion planning for manipulation of articulated objects. In2013 IEEE International Conference on Robotics and Automation, pages 1656–1662. IEEE, 2013. doi: 10.1109/ICRA.2013.6630792. 3 10 HumanoidMimicGen: Data Generation for Loco-Manipulation via Whole-Body Planning

    work page doi:10.1109/icra.2013.6630792 2013

  12. [12]

    Sauser, Darwin G

    Sylvain Calinon, Florent D’halluin, Eric L. Sauser, Darwin G. Caldwell, and Aude Billard. Learning and reproduction of gestures by imitation.IEEE Robotics and Automation Magazine, 17, 2010. 3

    2010

  13. [13]

    Generalizable domain adaptation for sim-and-real policy co-training.arXiv preprint arXiv:2509.18631, 2025

    Shuo Cheng, Liqian Ma, Zhenyang Chen, Ajay Mandlekar, Caelan Garrett, and Danfei Xu. Generalizable domain adaptation for sim-and-real policy co-training.arXiv preprint arXiv:2509.18631, 2025. 2, 3

    work page arXiv 2025

  14. [14]

    Diffusion policy: Visuomotor policy learning via action diffusion.The Int’l Journal of Robotics Research, 2023

    Cheng Chi, Zhenjia Xu, Siyuan Feng, Eric Cousineau, Yilun Du, Benjamin Burchfiel, Russ Tedrake, and Shuran Song. Diffusion policy: Visuomotor policy learning via action diffusion.The Int’l Journal of Robotics Research, 2023. 3, 8

    2023

  15. [15]

    Imitating task and motion planning with visuomotor transformers.arXiv preprint arXiv:2305.16309,

    Murtaza Dalal, Ajay Mandlekar, Caelan Garrett, Ankur Handa, Ruslan Salakhutdinov, and Dieter Fox. Imitating task and motion planning with visuomotor transformers.arXiv preprint arXiv:2305.16309,

    work page arXiv

  16. [16]

    Manipulation of documented objects by a walking humanoid robot

    Sébastien Dalibard, Alireza Nakhaei, Florent Lamiraux, and Jean-Paul Laumond. Manipulation of documented objects by a walking humanoid robot. In2010 10th IEEE-RAS International Conference on Humanoid Robots, pages 518–523. IEEE, 2010. doi: 10.1109/ICHR.2010.5686827. 3

    work page doi:10.1109/ichr.2010.5686827 2010

  17. [17]

    Dynamic walking and whole-body motion planning for humanoid robots: an integrated approach.The International Journal of Robotics Research, 32(9-10):1089–1103, 2013

    Sébastien Dalibard, Antonio El Khoury, Florent Lamiraux, Alireza Nakhaei, Michel Taïx, and Jean-Paul Laumond. Dynamic walking and whole-body motion planning for humanoid robots: an integrated approach.The International Journal of Robotics Research, 32(9-10):1089–1103, 2013. doi: 10.1177/ 0278364913481250. 3

    2013

  18. [18]

    Bridge Data: Boosting Generalization of Robotic Skills with Cross- Domain Datasets

    Frederik Ebert, Yanlai Yang, Karl Schmeckpeper, Bernadette Bucher, Georgios Georgakis, Kostas Daniilidis, Chelsea Finn, and Sergey Levine. Bridge Data: Boosting Generalization of Robotic Skills with Cross- Domain Datasets. InRobotics: Science and Systems, 2022. 1

    2022

  19. [19]

    Tsagarakis, and Enrico Mingo Hoffman

    Paolo Ferrari, Luca Rossini, Francesco Ruscelli, Arturo Laurenzi, Giuseppe Oriolo, Nikos G. Tsagarakis, and Enrico Mingo Hoffman. Multi-contact planning and control for humanoid robots: Design and validation of a complete framework.Robotics and Autonomous Systems, 166:104448, 2023. doi: 10.1016/j.robot. 2023.104448. 3

    work page doi:10.1016/j.robot 2023

  20. [20]

    Integrated task and motion planning.Annual review of control, robotics, and autonomous systems, 4:265–293, 2021

    Caelan Reed Garrett, Rohan Chitnis, Rachel Holladay, Beomjoon Kim, Tom Silver, Leslie Pack Kaelbling, and Tomás Lozano-Pérez. Integrated task and motion planning.Annual review of control, robotics, and autonomous systems, 4:265–293, 2021. 3

    2021

  21. [21]

    Skillgen: Automated demonstration generation for efficient skill learning and deployment

    Caelan Reed Garrett, Ajay Mandlekar, Bowen Wen, and Dieter Fox. Skillgen: Automated demonstration generation for efficient skill learning and deployment. In8th Annual Conference on Robot Learning, 2024. URLhttps://openreview.net/forum?id=YOFrRTDC6d. 2, 4, 6, 9

    2024

  22. [22]

    Humanoid manipulation planning using backward-forward search

    Michael X Grey, Caelan R Garrett, C Karen Liu, Aaron D Ames, and Andrea L Thomaz. Humanoid manipulation planning using backward-forward search. In2016 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 5467–5473. IEEE, 2016. 3

    2016

  23. [23]

    Humanoid locomotion and manipulation: Current progress and challenges in control, planning, and learning,

    Zhaoyuan Gu, Junheng Li, Wenlan Shen, Wenhao Yu, Zhaoming Xie, Stephen McCrory, Xianyi Cheng, Abdulaziz Shamsah, Robert Griffin, C Karen Liu, et al. Humanoid locomotion and manipulation: Current progress and challenges in control, planning, and learning.arXiv preprint arXiv:2501.02116, 2025. 3

    work page arXiv 2025

  24. [24]

    Point bridge: 3d representations for cross domain policy learning.arXiv preprint arXiv:2601.16212, 2026

    Siddhant Haldar, Lars Johannsmeier, Lerrel Pinto, Abhishek Gupta, Dieter Fox, Yashraj Narang, and Ajay Mandlekar. Point bridge: 3d representations for cross domain policy learning.arXiv preprint arXiv:2601.16212, 2026. 2, 3

    work page arXiv 2026

  25. [25]

    Randomized multi-modal motion planning for a humanoid robot manipulation task.International Journal of Robotics Research (IJRR), 30(6):676–698, 2011

    Kris Hauser and Victor Ng-Thow-Hing. Randomized multi-modal motion planning for a humanoid robot manipulation task.International Journal of Robotics Research (IJRR), 30(6):676–698, 2011. 3 11 HumanoidMimicGen: Data Generation for Loco-Manipulation via Whole-Body Planning

    2011

  26. [26]

    Multi-modal motion planning for a humanoid robot manipulation task

    Kris Hauser, Victor Ng-Thow-Hing, and Hector Gonzalez-Baños. Multi-modal motion planning for a humanoid robot manipulation task. InRobotics Research, pages 307–317. Springer, 2011. 3

    2011

  27. [27]

    Hover: Versatile neural whole-body controller for humanoid robots

    Tairan He, Wenli Xiao, Toru Lin, Zhengyi Luo, Zhenjia Xu, Zhenyu Jiang, Jan Kautz, Changliu Liu, Guanya Shi, Xiaolong Wang, et al. Hover: Versatile neural whole-body controller for humanoid robots. In2025 IEEE International Conference on Robotics and Automation (ICRA), pages 9989–9996. IEEE, 2025. 3

    2025

  28. [28]

    Adaflow: Imitation learning with variance-adaptive flow-based policies.arXiv preprint arXiv:2402.04292, 2024

    Xixi Hu, Bo Liu, Xingchao Liu, and Qiang Liu. Adaflow: Imitation learning with variance-adaptive flow-based policies.arXiv preprint arXiv:2402.04292, 2024. 8

    work page arXiv 2024

  29. [29]

    Movement imitation with nonlinear dynamical systems in humanoid robots.Proceedings 2002 IEEE Int’l Conf on Robotics and Automation, 2, 2002

    Auke Jan Ijspeert, Jun Nakanishi, and Stefan Schaal. Movement imitation with nonlinear dynamical systems in humanoid robots.Proceedings 2002 IEEE Int’l Conf on Robotics and Automation, 2, 2002. 3

    2002

  30. [30]

    Shrinking sphere: A parallel algorithm for computing the thickness of 3d objects.Computer-Aided Design and Applications, 13(2):199–207, 2016

    Masatomo Inui, Nobuyuki Umezu, and Ryohei Shimane. Shrinking sphere: A parallel algorithm for computing the thickness of 3d objects.Computer-Aided Design and Applications, 13(2):199–207, 2016. 16

    2016

  31. [31]

    doi:10.48550/arXiv.2410.24185 , abstract =

    Zhenyu Jiang, Yuqi Xie, Kevin Lin, Zhenjia Xu, Weikang Wan, Ajay Mandlekar, Linxi Fan, and Yuke Zhu. Dexmimicgen: Automated data generation for bimanual dexterous manipulation via imitation learning. arXiv preprint arXiv:2410.24185, 2024. 2, 4, 6, 7, 8, 9, 18

    work page arXiv 2024

  32. [32]

    A unified approach for motion and force control of robot manipulators: The operational space formulation.IEEE Journal on Robotics and Automation, 3(1):43–53, 1987

    Oussama Khatib. A unified approach for motion and force control of robot manipulators: The operational space formulation.IEEE Journal on Robotics and Automation, 3(1):43–53, 1987. 4

    1987

  33. [33]

    DROID: A Large-Scale In-The-Wild Robot Manipulation Dataset

    Alexander Khazatsky, Karl Pertsch, Suraj Nair, Ashwin Balakrishna, Sudeep Dasari, Siddharth Karamcheti, SoroushNasiriany, MohanKumarSrirama, LawrenceYunliangChen, KirstyEllis, etal. Droid: Alarge-scale in-the-wild robot manipulation dataset.arXiv preprint arXiv:2403.12945, 2024. 1, 3

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  34. [34]

    Momagen: Generating demonstrations under soft and hard constraints for multi-step bimanual mobile manipulation.arXiv preprint arXiv:2510.18316, 2025

    Chengshu Li, Mengdi Xu, Arpit Bahety, Hang Yin, Yunfan Jiang, Huang Huang, Josiah Wong, Sujay Garlanka, Cem Gokmen, Ruohan Zhang, et al. Momagen: Generating demonstrations under soft and hard constraints for multi-step bimanual mobile manipulation.arXiv preprint arXiv:2510.18316, 2025. 2

    work page arXiv 2025

  35. [35]

    Constraint-preserving data generation for visuomotor policy generalization

    Kevin Lin, Varun Ragunath, Andrew McAlinden, Aaditya Prasad, Jimmy Wu, Yuke Zhu, and Jeannette Bohg. Constraint-preserving data generation for visuomotor policy generalization. In9th Annual Conference on Robot Learning, 2025. URLhttps://openreview.net/forum?id=KSKzA1mwKs. 2, 6, 9

    2025

  36. [36]

    Manipulation as in simulation: Enabling accurate geometry perception in robots.arXiv preprint arXiv:2509.02530, 2025

    Minghuan Liu, Zhengbang Zhu, Xiaoshen Han, Peng Hu, Haotong Lin, Xinyao Li, Jingxiao Chen, Jiafeng Xu, Yichu Yang, Yunfeng Lin, et al. Manipulation as in simulation: Enabling accurate geometry perception in robots.arXiv preprint arXiv:2509.02530, 2025. 2

    work page arXiv 2025

  37. [37]

    Smplolympics: Sports environments for physically simulated humanoids,

    Zhengyi Luo, Jiashun Wang, Kangni Liu, Haotian Zhang, Chen Tessler, Jingbo Wang, Ye Yuan, Jinkun Cao, Zihui Lin, Fengyi Wang, et al. Smplolympics: Sports environments for physically simulated humanoids. arXiv preprint arXiv:2407.00187, 2024. 3

    work page arXiv 2024

  38. [38]

    SONIC: Supersizing Motion Tracking for Natural Humanoid Whole-Body Control

    Zhengyi Luo, Ye Yuan, Tingwu Wang, Chenran Li, Sirui Chen, Fernando Castañeda, Zi-Ang Cao, Jiefeng Li, David Minor, Qingwei Ben, et al. Sonic: Supersizing motion tracking for natural humanoid whole-body control.arXiv preprint arXiv:2511.07820, 2025. 3

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  39. [39]

    Sim-and-

    Abhiram Maddukuri, Zhenyu Jiang, Lawrence Yunliang Chen, Soroush Nasiriany, Yuqi Xie, Yu Fang, Wenqi Huang, Zu Wang, Zhenjia Xu, Nikita Chernyadev, et al. Sim-and-real co-training: A simple recipe for vision-based robotic manipulation.arXiv preprint arXiv:2503.24361, 2025. 2, 3, 8

    work page arXiv 2025

  40. [40]

    What matters in learning from offline human demonstrations for robot manipulation

    Ajay Mandlekar, Danfei Xu, Josiah Wong, Soroush Nasiriany, Chen Wang, Rohun Kulkarni, Li Fei-Fei, Silvio Savarese, Yuke Zhu, and Roberto Martín-Martín. What matters in learning from offline human demonstrations for robot manipulation. InConference on Robot Learning ( CoRL), 2021. 2, 3 12 HumanoidMimicGen: Data Generation for Loco-Manipulation via Whole-Bo...

    2021

  41. [41]

    Human-in-the-loop task and motion planning for imitation learning

    Ajay Mandlekar, Caelan Garrett, Danfei Xu, and Dieter Fox. Human-in-the-loop task and motion planning for imitation learning. In7th Annual Conference on Robot Learning, 2023. 3

    2023

  42. [42]

    Mimicgen: A data generation system for scalable robot learning using human demonstrations

    Ajay Mandlekar, Soroush Nasiriany, Bowen Wen, Iretiayo Akinola, Yashraj Narang, Linxi Fan, Yuke Zhu, and Dieter Fox. Mimicgen: A data generation system for scalable robot learning using human demonstrations. In7th Annual Conference on Robot Learning, 2023. URL https://openreview.net/ forum?id=dk-2R1f_LR. 1, 2, 4, 6

    2023

  43. [43]

    Guided imitation of task and motion planning

    Michael James McDonald and Dylan Hadfield-Menell. Guided imitation of task and motion planning. In Conference on Robot Learning, pages 630–640. PMLR, 2022. 3

    2022

  44. [44]

    Humanoid loco-manipulation planning based on graph search and reachability maps.IEEE Robotics and Automation Letters, 6(2):1840–1847, 2021

    Masaki Murooka, Iori Kumagai, Mitsuharu Morisawa, Fumio Kanehiro, and Abderrahmane Kheddar. Humanoid loco-manipulation planning based on graph search and reachability maps.IEEE Robotics and Automation Letters, 6(2):1840–1847, 2021. doi: 10.1109/LRA.2021.3060728. 3

    work page doi:10.1109/lra.2021.3060728 2021

  45. [45]

    Open x-embodiment: Robotic learning datasets and rt-x models: Open x-embodiment collaboration 0

    Abby O’Neill, Abdul Rehman, Abhiram Maddukuri, Abhishek Gupta, Abhishek Padalkar, Abraham Lee, Acorn Pooley, Agrim Gupta, Ajay Mandlekar, Ajinkya Jain, et al. Open x-embodiment: Robotic learning datasets and rt-x models: Open x-embodiment collaboration 0. In2024 IEEE Int’l Conf on Robotics and Automation (ICRA), 2024. 1

    2024

  46. [46]

    Alvinn: An autonomous land vehicle in a neural network

    Dean A Pomerleau. Alvinn: An autonomous land vehicle in a neural network. InAdvances in neural information processing systems, pages 305–313, 1989. 3, 4

    1989

  47. [47]

    What matters in learning from large-scale datasets for robot manipulation.arXiv preprint arXiv:2506.13536, 2025

    Vaibhav Saxena, Matthew Bronars, Nadun Ranawaka Arachchige, Kuancheng Wang, Woo Chul Shin, Soroush Nasiriany, Ajay Mandlekar, and Danfei Xu. What matters in learning from large-scale datasets for robot manipulation.arXiv preprint arXiv:2506.13536, 2025. 2

    work page arXiv 2025

  48. [48]

    Is imitation learning the route to humanoid robots?Trends in cognitive sciences, 3, 1999

    Stefan Schaal. Is imitation learning the route to humanoid robots?Trends in cognitive sciences, 3, 1999. 3

    1999

  49. [49]

    Humanoidbench: Simulated humanoid benchmark for whole-body locomotion and manipulation, 2024

    Carmelo Sferrazza, Dun-Ming Huang, Xingyu Lin, Youngwoon Lee, and Pieter Abbeel. Humanoidbench: Simulated humanoid benchmark for whole-body locomotion and manipulation, 2024. 3

    2024

  50. [50]

    A sequential quadratic programming approach to the solution of open-loop generalized nash equilibria,

    Balakumar Sundaralingam, Siva Kumar Sastry Hari, Adam Fishman, Caelan Garrett, Karl Van Wyk, Valts Blukis, Alexander Millane, Helen Oleynikova, Ankur Handa, Fabio Ramos, Nathan Ratliff, and Dieter Fox. Curobo: Parallelized collision-free robot motion generation. In2023 IEEE International Conference on Robotics and Automation (ICRA), pages 8112–8119, 2023....

    work page doi:10.1109/icra48891.2023.10160765 2023

  51. [51]

    curobo: Parallelized collision-free minimum-jerk robot motion generation, 2023

    Balakumar Sundaralingam, Siva Kumar Sastry Hari, Adam Fishman, Caelan Garrett, Karl Van Wyk, Valts Blukis, Alexander Millane, Helen Oleynikova, Ankur Handa, Fabio Ramos, Nathan Ratliff, and Dieter Fox. curobo: Parallelized collision-free minimum-jerk robot motion generation, 2023. 16

    2023

  52. [52]

    Interndata-a1: Pioneering high-fidelity synthetic data for pre-training generalist policy.arXiv preprint arXiv:2511.16651, 2025

    Yang Tian, Yuyin Yang, Yiman Xie, Zetao Cai, Xu Shi, Ning Gao, Hangxu Liu, Xuekun Jiang, Zherui Qiu, Feng Yuan, et al. Interndata-a1: Pioneering high-fidelity synthetic data for pre-training generalist policy. arXiv preprint arXiv:2511.16651, 2025. 3

    work page arXiv 2025

  53. [53]

    Mujoco: A physics engine for model-based control

    Emanuel Todorov, Tom Erez, and Yuval Tassa. Mujoco: A physics engine for model-based control. In IEEE/RSJ International Conference on Intelligent Robots and Systems, pages 5026–5033, 2012. 7, 17

    2012

  54. [54]

    Empirical analysis of sim-and-real cotraining of diffusion policies for planar pushing from pixels,

    Adam Wei, Abhinav Agarwal, Boyuan Chen, Rohan Bosworth, Nicholas Pfaff, and Russ Tedrake. Empirical analysis of sim-and-real cotraining of diffusion policies for planar pushing from pixels.arXiv preprint arXiv:2503.22634, 2025. 2, 3, 8

    work page arXiv 2025

  55. [55]

    OmniRetarget: Interaction-Preserving Data Generation for Humanoid Whole-Body Loco-Manipulation and Scene Interaction

    Lujie Yang, Xiaoyu Huang, Zhen Wu, Angjoo Kanazawa, Pieter Abbeel, Carmelo Sferrazza, C Karen Liu, Rocky Duan, and Guanya Shi. Omniretarget: Interaction-preserving data generation for humanoid whole-body loco-manipulation and scene interaction.arXiv preprint arXiv:2509.26633, 2025. 3 13 HumanoidMimicGen: Data Generation for Loco-Manipulation via Whole-Bod...

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  56. [56]

    Physics-driven data generation for contact-rich manipulation via trajectory optimization.arXiv preprint arXiv:2502.20382, 2025

    Lujie Yang, HJ Suh, Tong Zhao, Bernhard Paus Graesdal, Tarik Kelestemur, Jiuguang Wang, Tao Pang, and Russ Tedrake. Physics-driven data generation for contact-rich manipulation via trajectory optimization. arXiv preprint arXiv:2502.20382, 2025. 2

    work page arXiv 2025

  57. [57]

    Genie Sim 3.0 : A High-Fidelity Comprehensive Simulation Platform for Humanoid Robot

    Chenghao Yin, Da Huang, Di Yang, Jichao Wang, Nanshu Zhao, Chen Xu, Wenjun Sun, Linjie Hou, Zhijun Li, Junhui Wu, et al. Genie sim 3.0: A high-fidelity comprehensive simulation platform for humanoid robot.arXiv preprint arXiv:2601.02078, 2026. 3

    work page internal anchor Pith review Pith/arXiv arXiv 2026

  58. [58]

    Reinforcegen: Hybrid skill policies with automated data generation and reinforcement learning.arXiv preprint arXiv:2512.16861, 2025

    Zihan Zhou, Animesh Garg, Ajay Mandlekar, and Caelan Garrett. Reinforcegen: Hybrid skill policies with automated data generation and reinforcement learning.arXiv preprint arXiv:2512.16861, 2025. 2

    work page arXiv 2025

  59. [59]

    robosuite: A Modular Simulation Framework and Benchmark for Robot Learning

    Yuke Zhu, Josiah Wong, Ajay Mandlekar, and Roberto Martín-Martín. robosuite: A modular simulation framework and benchmark for robot learning. InarXiv preprint arXiv:2009.12293, 2020. 7, 17 14 HumanoidMimicGen: Data Generation for Loco-Manipulation via Whole-Body Planning Overview The appendix contains the following content. •Skill Planning Example(Appendi...

    work page internal anchor Pith review Pith/arXiv arXiv 2009

  60. [60]

    We manually annotate source demonstration subtasks that will require locomotion

  61. [61]

    Thewhole-inv-kinematicsprocedure does not consider collisions, and all joints apart from the legs are unlocked and free to move from the current configuration

    For each subtask that requires locomotion, we invokewhole-inv-kinematicsto infer a target base pose for one (or both) arm poses at the start of each subtask. Thewhole-inv-kinematicsprocedure does not consider collisions, and all joints apart from the legs are unlocked and free to move from the current configuration

  62. [62]

    To move from a current base configuration to a new base configuration, a straight-line interpolated path is used (similar to interpolation segments for the arms in DexMimicGen [30]). This baseline lacks several crucial features introduced by HumanoidMimicGen, including the use of skill reasoning, motion planning for locomotion and arm movement, and collis...