Pith.

Computer Vision and Pattern Recognition · arXiv:2611.12885 v1 · submitted 9 Nov 2026 · 110 references

World In-Context Learning for Agentic Robot Manipulation

Heng Fang, Zehui Guo, Yuehan Zhang, Mingyang Du, Dingkang Liang, Xiang Bai

The paper says

“Trained action policies write manipulation skill into network weights, tying what is learned to the distribution it is learned from.”

more of the abstract

In-context learning (ICL) avoids this coupling but has not matched trained policies, because a demonstration recorded in another scene carries no feedback from the world the robot currently faces. We observe that what an agent learns in context is semantic, recording task steps and object roles rather than visual appearance, so experience gathered in one world transfers to another. Based on this insight, we propose World In-Context Learning (WICL), a paradigm in which a frozen model starts from a demonstration, rehearses the task in a world such as a simulator, and writes the feedback into its context rather than its weights. We build RoboScript, a modular infrastructure that implements WICL for agentic manipulation. From a single demonstration, RoboScript learns each task through rehearsal in simulation, freezes the learned knowledge as a Task Memory, and deploys the frozen memory on new scenes with data isolation that ensures every deployment input is available on a real robot. On all 50 tasks of RoboTwin 2.0, RoboScript reaches 83.4% mean success across clean and randomized settings with one demonstration per task and no gradient updates, exceeding the strongest trained action policy by over 20 points. The method generalizes across visual distributions, improves through reflection, scales with the foundation model, and transfers to a physical dual-arm robot through cross-embodiment memory adaptation. With a world in the loop, a robot can learn a manipulation task in its context as well as it learns one in its weights. All code is available at https://github.com/ArcheOne-AI/RoboScript

from the abstract

Pith says

Machine review · deepseek-v4-flash · 10 Nov 2026 · what was read was not recorded · not checked by a person

It can be wrong. Challenge this review or reply to it.

the referee pass did not run

How often objections from reviews like this one are right: not yet measured

Are you an author? What you can do.

Not checked: figures, code, what the 110 references say.

What would settle it

Evaluate the frozen Task Memories (GPT-6.1-Sol) on a different simulator (e.g., Isaac Sim or MuJoCo) with the same 50 tasks but different object meshes, textures, and lighting, without any retraining or re-rehearsing. If the success rate drops significantly (e.g., >20 points), the claim of semantic transfer across worlds is weakened because the memory may be overfit to the specific simulator's visual and physical regularities.

from the skeptic's reading

How the review went

2 readings by deepseek-v4-flash · 10 Nov 2026

  1. Reader holds up, with conditions confidence high
  2. Skeptic agrees 15 s · 38,602 tokens read, 863 written

Not run on this paper: the referee, the editor, the simulated authors' defence.

Pith did not check this review's objections against the paper.

The review in full

The strongest objection

The central claim is that WICL decouples skill from visual appearance because the Task Memory stores only semantic knowledge (object roles, relative geometry). The paper states that a world must be 'faithful to the task' (Sec. 3, [b120]) and that the learned knowledge transfers because it is semantic ([b108]–[b109]). However, the only world used is the RoboTwin 2.0 simulator, and the only evaluation is on the same simulator (clean and randomized variants). The randomized setting changes textures, lighting, table height, and distractors, but the underlying object geometry, physics, and task semantics are identical to the clean setting. The paper's own failure analysis (Fig. 5b) shows that knowledge failures do not increase under randomization, which is consistent with the memory being semantic—but it is also consistent with the simulator's randomization being too mild to change the semantic content that the memory relies on. The real-robot transfer (Sec. 5.7) is on only 4 tasks with small sample sizes (10 episodes each) and uses a lightweight adapter that remaps embodiment parameters, not a test of whether the memory transfers to a visually and physically different world. The paper acknowledges that deformable objects and tight bimanual coordination remain challenging ([b542]), but the core assumption of faithfulness for the tasks tested is not independently verified beyond the reported success rates on the same simulator. The claim that 'what is learned in context holds across distributions' ([b213]) is supported only by the simulator's randomized setting, which shares the same underlying task semantics. A stronger test would be to evaluate on a different simulator or a different real-world setup with different object instances, lighting, and backgrounds, without retraining or re-rehearsing.

The reader's weakest_assumption is that the simulator is 'faithful to the task' in the sense that behaviors change the semantic content as they would in the target environment. The paper's central claim is that the learned knowledge is semantic and transfers across worlds, but the only evidence for this transfer is within the same simulator (clean to randomized) and a limited real-robot test. The randomized setting changes visual appearance but not the underlying task semantics, so it does not test whether the memory would transfer to a world with different physics, object geometry, or task semantics. The paper acknowledges this limitation indirectly by noting that deformable objects and tight bimanual coordination remain challenging, but the core assumption of faithfulness for the tasks tested is not independently verified. The concrete test proposed would directly test whether the memory transfers to a different simulator with different object meshes and physics, which would be a stronger test of the semantic transfer claim. The reader's verdict is CONDITIONAL, and I agree with that assessment. The concern is load-bearing because if the memory does not transfer to a different simulator, the claim that 'what is learned in context holds across distributions' is overstated. The paper's own failure analysis shows that knowledge failures do not increase under randomization, but this is consistent with the randomization being too mild. The real-robot test is too limited to rule out overfitting to the simulator. Therefore, the verdict should remain CONDITIONAL, with the condition being that the authors provide evidence of transfer to a different simulator or a more diverse real-world setup.

Pith's reading of the paper (a machine's summary, not the authors')

This paper introduces World In-Context Learning (WICL), a paradigm that lets a frozen foundation model learn a manipulation task from a single demonstration by rehearsing in a simulator and storing the learned skill in its context as readable text. The authors build RoboScript, an infrastructure that implements WICL, and show that on 50 tasks of the RoboTwin 2.0 benchmark it reaches 83.4% mean success with one demonstration and no gradient updates, exceeding the best trained action policy by over 20 points. The learned knowledge is semantic (object roles, task steps, relative geometry) rather than visual, so it transfers across visual distributions and even to a different physical robot without retraining. The paper argues that task skill need not reside in network weights and that a world in the loop—a simulator providing feedback during rehearsal—makes in-context learning competitive with trained policies for the first time.

What carries the argument. The central object is the Task Memory, a structured Markdown document that records the semantic content of a task: object roles, workflow stages, relative grasp and placement specifications (e.g., 'approach the handle from the left and grasp at its midpoint'), learned parameters, and lessons from failures. It is built through a two-stage process: Imitation (S1) where the agent writes an initial memory from the demonstration, and Rehearsal (S2) where it practices on new simulator instances, revises the memory after failures, and freezes it. The key identity is that the knowledge in the Task Memory refers only to semantic content σ (object roles, relative geometry, task steps) and to quantities measurable at execution time, so it transfers across worlds that share the same σ, even if they differ in appearance, lighting, or physics parameters. A memory linter enforces this by rejecting coordinates, ground-truth terms, and instance identifiers.

Core claim. The central claim is that a frozen foundation model can learn a robot manipulation task as well in its context as a trained policy learns in its weights, provided the model can rehearse in a world (e.g., a simulator) that gives feedback. The paper formalizes this as World In-Context Learning (WICL): from one demonstration, the agent builds an initial Task Memory, practices in the simulator on training instances, revises the memory after failures, and freezes the result. The frozen memory is a structured Markdown document containing object roles, workflow stages, relative grasp and placement specifications, and lessons—all semantic, not tied to any scene's visual appearance. Because the knowledge is semantic, it transfers to new visual distributions, to unseen instances, and even to a different physical robot via a lightweight adapter. The paper reports 83.4% mean success on 50 tasks with one demonstration, versus 61.7% for the best trained policy (4D-WAM) trained on 50 demonstrations per task. Under domain randomization, RoboScript retains over 95% of its clean success while trained policies retain at most 65.1%.

The premise it rests on. The simulator used for rehearsal must be faithful to the task: behaviors in the simulator must change the semantic content (object roles, relative geometry, task steps) in the same way they would in the target environment, whether that is a randomized simulator or a real robot.

If the paper is right (Pith's inferences, not the authors')

If WICL holds, a robot can acquire a new manipulation task from a single human demonstration and a few minutes of simulator rehearsal, without any gradient training or dataset collection.

The decoupling of task skill from network weights means the same foundation model can serve as the base for an arbitrary number of tasks, with each task's memory added to its context—no weight interference between tasks.

Because the Task Memory is readable text, the agent can review its own evaluations and rewrite what it has learned, enabling recursive self-improvement (RSI) without gradient updates. The paper shows three rounds of RSI on 17 weak tasks, with net gains in each round.

The paradigm extends to any embodiment where a faithful simulator exists: the paper demonstrates cross-embodiment transfer from a simulated ALOHA to a physical PiPER robot via a lightweight parameter adapter, without re-rehearsing.

The paper's claim that 'what an agent learns in context is semantic' could be tested more broadly: if true, then WICL should work for tasks where the semantic content is complex (e.g., assembly with many parts) but the visual appearance is simple, and should fail for tasks where the semantics are ambiguous or the simulator is not faithful (e.g., deformable objects). The paper acknowledges deformab

The reliance on a faithful simulator is the weakest point: the paper's results on RoboTwin 2.0 use the same simulator for rehearsal and evaluation, which is a best-case scenario. The real-robot transfer is only on four tasks with a lightweight adapter and no online replanning. A stronger test would be to rehearse in one simulator (e.g., RoboTwin) and deploy in a very different simulator (e.g., Isa

The paper's data isolation protocol (linter, deployable vs. privileged tiers) is a methodological contribution that could be adopted by other agentic manipulation systems to ensure simulator-learned skills transfer to real hardware. However, the paper does not measure how much the linter degrades performance relative to a version that uses privileged information—the privilege-free ablation shows n

Check this review

Pith stored no record of this review's passes and took no fingerprint of it. Reviews made by code from 13 Nov 2026 on carry both.

How often reviews like this one are right

Reviews like this one: written by deepseek-v4-flash, where what was read was not recorded.

How often objections from reviews like this one are right: not yet measured

Every value above, in the review's public record (JSON)

Reviews by people

This review was written by a machine. No person has reviewed this paper on Pith yet. A scientist listed on Pith can review it for a fee, and their review is published here whatever it concludes.

Request a human review

How Pith checks papers · The full audit of this paper

The paper

Abstract

Trained action policies write manipulation skill into network weights, tying what is learned to the distribution it is learned from. In-context learning (ICL) avoids this coupling but has not matched trained policies, because a demonstration recorded in another scene carries no feedback from the world the robot currently faces. We observe that what an agent learns in context is semantic, recording task steps and object roles rather than visual appearance, so experience gathered in one world transfers to another. Based on this insight, we propose World In-Context Learning (WICL), a paradigm in which a frozen model starts from a demonstration, rehearses the task in a world such as a simulator, and writes the feedback into its context rather than its weights. We build RoboScript, a modular infrastructure that implements WICL for agentic manipulation. From a single demonstration, RoboScript learns each task through rehearsal in simulation, freezes the learned knowledge as a Task Memory, and deploys the frozen memory on new scenes with data isolation that ensures every deployment input is available on a real robot. On all 50 tasks of RoboTwin 2.0, RoboScript reaches 83.4% mean success across clean and randomized settings with one demonstration per task and no gradient updates, exceeding the strongest trained action policy by over 20 points. The method generalizes across visual distributions, improves through reflection, scales with the foundation model, and transfers to a physical dual-arm robot through cross-embodiment memory adaptation. With a world in the loop, a robot can learn a manipulation task in its context as well as it learns one in its weights. All code is available at https://github.com/ArcheOne-AI/RoboScript

The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. No sentence in this record was written by a model.

abstract compiled · 2026-11-11

The authors' main claim, as the classifier picked it out 0.99

“With a world in the loop, a robot can learn a manipulation task in its context as well as it learns one in its weights.”

How the main claim is stated: hedged possibility 0.52

What the abstract itself states, as probabilities: claims first 0.10 · states limitation 0.19 · human participants 0.03 · controlled comparison 0.86 · randomized assignment 0.16 · states quantitative result 0.99 · compares with prior results 0.97 · mentions code or data availability 0.98

The abstract, sentence by sentence

This table labels the abstract only (arXiv listing abstract v1). The number is the probability the classifier gave the label.

SentenceRoleThe authors' own workStates the main conclusionStrength
Trained action policies write manipulation skill into network weights, tying what is learned to the distribution it is learned from. background 1.00 0.09 0.03 asserted 1.00
In-context learning (ICL) avoids this coupling but has not matched trained policies, because a demonstration recorded in another scene carries no feedback from the world the robot currently faces. gap 0.75 0.11 0.04 asserted 1.00
We observe that what an agent learns in context is semantic, recording task steps and object roles rather than visual appearance, so experience gathered in one world transfers to another. result 0.66 0.84 0.21 asserted 1.00
Based on this insight, we propose World In-Context Learning (WICL), a paradigm in which a frozen model starts from a demonstration, rehearses the task in a world such as a simulator, and writes the feedback into its context rather than its weights. contribution 0.58 0.54 0.12 asserted 0.96
We build RoboScript, a modular infrastructure that implements WICL for agentic manipulation. contribution 0.62 0.91 0.03 asserted 1.00
From a single demonstration, RoboScript learns each task through rehearsal in simulation, freezes the learned knowledge as a Task Memory, and deploys the frozen memory on new scenes with data isolation that ensures every deployment input is available on a real robot. method 0.92 0.91 0.08 asserted 1.00
On all 50 tasks of RoboTwin 2.0, RoboScript reaches 83.4% mean success across clean and randomized settings with one demonstration per task and no gradient updates, exceeding the strongest trained action policy by over 20 points. result 1.00 0.96 0.27 asserted 1.00
The method generalizes across visual distributions, improves through reflection, scales with the foundation model, and transfers to a physical dual-arm robot through cross-embodiment memory adaptation. result 0.96 0.92 0.12 asserted 1.00
With a world in the loop, a robot can learn a manipulation task in its context as well as it learns one in its weights. interpretation 0.95 0.67 0.83 hedged possibility 0.52
All code is available at https://github.com/ArcheOne-AI/RoboScript contribution 0.59 0.47 0.15 hedged possibility 0.66

What kind of paper the classifier reports this is: method or tool 1.00 · evidence from observation or experiment 0.83

What was examined, and what was not

Sources read

  • arXiv listing abstract v1

Not examined

  • full text
  • figures and tables
  • references and citing sentences
  • relations to other papers

Levels run

  • P abstract

Not run

  • S sections
  • T body sentences
  • C citations
  • R cross-paper relations

How the labels were made

Compiled by classifier-1.13.0 via TypeSafe API, question bank abstract-p1-20260923, 2026-11-11: 1 request, 7,732 input tokens.

Every quoted sentence on this record is the authors'. Every label is from Pith's closed vocabulary, with the probability shown beside it. This record as JSON

Figures (13)
Figure 1 of the paper
Figure 1. (a) Three paradigms for learning manipulation. Skill in weights is bound to the training domain. ICL avoids retraining but is bound to the shown scene. WICL rehearses in a world and stores semantic knowledge that transfers across settings. (b) RoboScript maps the coding-agent stack to robot manipulation, with the world replacing the terminal. (c) Success rate on all 50 RoboTwin 2.0 tasks with one demonstration and no gradient updates. no feedback from the world. A demonstration is recorded in another scene and cannot tell the robot what will happen on the table in front of it. A…
Figure 2 of the paper
Figure 2. Infrastructure of RoboScript. Every component runs as an independent service, and any method, benchmark, embodiment, or extension plugs in as a replaceable module. All model calls pass through one gateway. The agent plane reaches the world only through the deployable interface (D), and only the control plane accesses privileged data (P). frozen memory on new scenes, anticipating how the task should unfold and recovering from failures. After evaluation, a reflection agent reviews the records and writes guidance for the next round, realizing RSI over several iterations. Data isolation…
Figure 3 of the paper
Figure 3. (a) The WICL pipeline. From one demonstration, the agent rehearses in a world, freezes the context, and deploys it on unseen instances. (b) Recursive self-improvement over disjoint evaluation blocks with frozen weights. (c) Context knowledge refers to shared semantics σ and transfers across worlds, whereas trained weights do not. rule is met, and the final context C ⋆ is frozen. The learner then acts on new instances with the frozen context and receives no feedback from the world, at ∼ πθ(· | ot, C⋆ ), e ∈ Etest. (2) Since no evaluation instance is rehearsed, C ⋆ is tested on genera
Figure 4 of the paper
Figure 4. The four stages of RoboScript and the data tier of each role. The agent builds reusable skills, learns a Task Memory from one demonstration through imitation (S1) and rehearsal (S2), deploys the frozen memory with anticipation and recovery, and improves through reflection over disjoint evaluation blocks. Green borders mark deployable data (real-robot inputs), and orange borders mark privileged data (simulator only). 4 RoboScript RoboScript implements WICL for robot manipulation
Figure 10 of the paper
Figure 10. Real-robot setup with a dual-arm PiPER and RGB-D cameras. We deploy RoboScript on a physical dual-arm PiPER robot (
Figure 11 of the paper
Figure 11. Execution frames for four real-robot tasks. Each row shows key frames from a successful episode. The Task Memory was rehearsed on a simulated ALOHA and transferred to the PiPER through a lightweight adapter without re-rehearsing. 6 Conclusion We have introduced World In-Context Learning (WICL), a paradigm that extends in-context learning with feedback from a world, and implemented it in RoboScript, a modular infrastructure for agentic manipulation with strict data isolation. On all 50 tasks of RoboTwin 2.0, the agent builds its own skills, learns each task from one demonstration…
Figure 12 of the paper
Figure 12. Detailed architecture of RoboScript, expanding

Showing 7 of the paper's 13 figures. Figures are the authors', reproduced from the paper.

References (110)
  1. [1] Lakshya A Agrawal, Shangyin Tan, Dilara Soylu, Noah Ziems, Rishi Khare, Krista Opsahl-Ong, Ar- nav Singhvi, Herumb Shandilya, Michael J Ryan, Meng Jiang, Christopher Potts, Koushik Sen, Alexan- dros G. Dimakis, Ion Stoica, Dan Klein, Matei Zaharia, and Omar Khattab. GEPA: Refl... on Pith
  2. [2] Kevin Black, Noah Brown, James Darpinian, Karan Dhabalia, Danny Driess, Adnan Esmail, Michael Robert Equi, Chelsea Finn, Niccolo Fusai, Manuel Y. Galliker, Dibya Ghosh, Lachy Groom, Karol Hausman, Brian Ichter, Szymon Jakubczak, Tim Jones, Liyiming Ke, Devin LeBlanc, Sergey Le... on Pith
  3. [3] Kevin Black, Noah Brown, Danny Driess, Adnan Esmail, Michael Robert Equi, Chelsea Finn, Niccolo Fusai, Lachy Groom, Karol Hausman, Brian Ichter, Szymon Jakubczak, Tim Jones, Liyiming Ke, Sergey Levine, Adrian Li-Bell, Mohith Mothukuri, Suraj Nair, Karl Pertsch, Lucy Xiaoyang S... on Pith
  4. [4] Konstantinos Bousmalis, Giulia Vezzani, Dushyant Rao, Coline Manon Devin, Alex X. Lee, Maria Bauza Villa- longa, Todor Davchev, Yuxiang Zhou, Agrim Gupta, Akhil Raju, Antoine Laurens, Claudio Fantacci, Valentin Dalibard, Martina Zambelli, Murilo Fernandes Martins, Rugile Pevce... on Pith
  5. [5] Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel Ziegler, Jeffrey W... on Pith
  6. [6] Qingwen Bu, Jisong Cai, Li Chen, Xiuqi Cui, Yan Ding, Siyuan Feng, Xindong He, Xu Huang, et al. AgiBot World Colosseo: A large-scale manipulation platform for scalable and intelligent embodied systems. In2025 IEEE/RSJ International Conference on Intelligent Robots and Systems ... on Pithdoi:10.1109/iros60139.2025.11247088
  7. [7] Jisong Cai, Long Ling, Shiwei Chu, Zhongshan Liu, Jiayue Kang, Zhixuan Liang, Wenjie Xu, Yinan Mao, Weinan Zhang, Xiaokang Yang, Ru Ying, Ran Zheng, and Yao Mu. AHA-WAM: Asynchronous horizon-adaptive world-action modeling with observation-guided context routing.arXiv preprint ... on PithPith's reviewarXiv
  8. [8] Rui Cai, Jun Guo, Xinze He, Piaopiao Jin, Jie Li, Bingxuan Lin, Futeng Liu, Wei Liu, Fei Ma, Kun Ma, Feng Qiu, Heng Qu, Yifei Su, Qiao Sun, Dong Wang, Donghao Wang, Yunhong Wang, Rujie Wu, Diyun Xiang, Yu Yang, Hangjun Ye, Yuan Zhang, and Quanyun Zhou. Xiaomi-robotics-0: An op... on PitharXiv
  9. [9] Chi-Lam Cheang, Guangzeng Chen, Ya Jing, Tao Kong, Hang Li, Yifeng Li, Yuxiao Liu, Hongtao Wu, Jiafeng Xu, Yichu Yang, Hanbo Zhang, and Minzhao Zhu. GR-2: A generative video-language-action model with web-scale knowledge for robot manipulation.arXiv preprint arXiv:2410.06158, 2024 on PithPith's reviewarXiv
  10. [10] Siyi Chen, Hugo Hadfield, Alex Zook, Mikaela Angelina Uy, Chan Hee Song, Erwin Coumans, Xuning Yang, 25 Faisal Ladhak, Qing Qu, Stan Birchfield, Jonathan Tremblay, and Valts Blukis. VoLo: A physical orchestrator for open-vocabulary long-horizon manipulation. InConference on Ro... on Pith
  11. [11] Tianxing Chen, Zanxin Chen, Baijun Chen, Zijian Cai, Yibin Liu, Zixuan Li, Qiwei Liang, Xianliang Lin, Yiheng Ge, Zhenyu Gu, Weiliang Deng, Yubin Guo, Tian Nian, Xuanbing Xie, Qiangyu Chen, Kailun Su, Tianling Xu, Guodong Liu, Mengkang Hu, Huan-Ang Gao, Kaixuan Wang, Zhixuan L... on Pith
  12. [12] Yixiang Chen, Peiyan Li, Yuan Xu, Qisen Ma, Jiabing Yang, Kai Wang, Jianhua Yang, Dong An, He Guan, Gaoteng Liu, Jianlou Si, Jun Huang, Jing Liu, Nianfeng Liu, Yan Huang, and Liang Wang. FlowWAM: Optical flow as a unified action representation for world action models.arXiv pre... on PithPith's reviewarXiv
  13. [13] Norman Di Palo and Edward Johns. Keypoint action tokens enable in-context imitation learning in robotics. In Proceedings of Robotics: Science and Systems, Delft, Netherlands, July 2024. doi: 10.15607/RSS.2024.XX.096 on Pithdoi:10.15607/rss.2024.xx.096
  14. [14] Qingxiu Dong, Lei Li, Damai Dai, Ce Zheng, Jingyuan Ma, Rui Li, Heming Xia, Jingjing Xu, Zhiyong Wu, Baobao Chang, Xu Sun, Lei Li, and Zhifang Sui. A survey on in-context learning. InProceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages ... on Pithdoi:10.18653/v1/2024.emnlp-main.64
  15. [15] Danny Driess, Fei Xia, Mehdi S. M. Sajjadi, Corey Lynch, Aakanksha Chowdhery, Brian Ichter, Ayzaan Wahid, Jonathan Tompson, Quan Vuong, Tianhe Yu, Wenlong Huang, Yevgen Chebotar, Pierre Sermanet, Daniel Duckworth, Sergey Levine, Vincent Vanhoucke, Karol Hausman, Marc Toussaint... on Pith
  16. [16] Yilun Du, Sherry Yang, Bo Dai, Hanjun Dai, Ofir Nachum, Josh Tenenbaum, Dale Schuurmans, and Pieter Abbeel. Learning universal policies via text-guided video generation. InAdvances in Neural Information Processing Systems, volume 36, pages 9156–9172, 2023. doi: 10.52202/075280-0403 on Pithdoi:10.52202/075280-0403
  17. [17] Yan Duan, Marcin Andrychowicz, Bradly Stadie, Jonathan Ho, Jonas Schneider, Ilya Sutskever, Pieter Abbeel, and Wojciech Zaremba. One-shot imitation learning. InAdvances in Neural Information Processing Systems, volume 30. Curran Associates, Inc., 2017. URLhttps://proceedings.n... on Pith
  18. [18] Heng Fang, Shangru Li, Shuhan Wang, Xuanyang Xi, Dingkang Liang, and Xiang Bai. Towards generalizable robotic manipulation in dynamic environments. InEuropean Conference on Computer Vision, pages 314–333. Springer, 2026 on Pith
  19. [19] Kuan Fang, Fangchen Liu, Pieter Abbeel, and Sergey Levine. MOKA: Open-world robotic manipulation through mark-based visual prompting. InProceedings of Robotics: Science and Systems, Delft, Netherlands, July 2024. doi: 10.15607/RSS.2024.XX.062 on Pithdoi:10.15607/rss.2024.xx.062
  20. [20] Chelsea Finn, Tianhe Yu, Tianhao Zhang, Pieter Abbeel, and Sergey Levine. One-shot visual imitation learning via meta-learning. In Sergey Levine, Vincent Vanhoucke, and Ken Goldberg, editors,Proceedings of the 1st Annual Conference on Robot Learning, volume 78 ofProceedings of... on Pith
show 80 more
  1. [21] Haoyu Fu, Diankun Zhang, Zongchuang Zhao, Jianfeng Cui, Dingkang Liang, Chong Zhang, Dingyuan Zhang, Hongwei Xie, Bing Wang, and Xiang Bai. Orion: A holistic end-to-end autonomous driving framework by vision-language instructed action generation. InProceedings of the IEEE/CVF ... on Pith
  2. [22] Haoyu Fu, Diankun Zhang, Zongchuang Zhao, Jianfeng Cui, Hongwei Xie, Bing Wang, Guang Chen, Dingkang Liang, and Xiang Bai. Minddrive: A vision-language-action model for autonomous driving via online reinforcement learning. InProceedings of the European Conference on Computer V... on Pith
  3. [23] Letian Fu, Justin Yu, Karim El-Refai, Ethan Kou, Haoru Xue, Huang Huang, Wenli Xiao, Li Fei-Fei, Guanya Shi, Jiajun Wu, S. Shankar Sastry, Yuke Zhu, Ken Goldberg, and Linxi Fan. CaP-X: A framework for benchmarking and improving coding agents for robot manipulation. InProceedin... on Pith
  4. [24] Max Fu, Huang Huang, Gaurav Datta, Lawrence Yunliang Chen, Will Panitch, Fangchen Liu, Hui Li, and Ken Goldberg. ICRT: In-context imitation learning via next-token prediction. In2025 IEEE International Conference on Robotics and Automation (ICRA), pages 5937–5944. IEEE, 2025. ... on Pithdoi:10.1109/icra55743.2025.11128272
  5. [25] Ning Gao, Jinliang Zheng, Xing Gao, Haoxiang Ma, Hanqing Wang, Yukai Wang, Jiantong Chen, Zanxin Chen, Shujie Zhang, Mingda Jia, Xuekun Jiang, Zihou Zhu, Xinyu Li, Shuai Wang, Hao Li, Wenzhe Cai, Yuqiang Yang, Xudong Xu, Zhaoyang Lyu, Yao Mu, Tai Wang, Jiangmiao Pang, Jia Zeng... on PithPith's reviewarXiv
  6. [26] Generalist Team. GEN-1.5: Embodied foundation models are one-shot learners. Generalist AI Blog,https: //generalistai.com/blog/gen-1.5, August 2026 on Pith
  7. [27] Zhaopeng Gu, Bingke Zhu, Tianxi Lin, Guibo Zhu, Yingying Chen, Kai Wang, Tingyu Yuan, Fan Yang, Chaoyang Zhao, Zhaowen Li, Peng Su, and Jinqiao Wang. HarnessWAM: Bridging prediction and deliberation in world action models.arXiv preprint arXiv:2608.09516, 2026 on PithPith's reviewarXiv
  8. [28] Jun Guo, Qiwei Li, Peiyan Li, Zilong Chen, Nan Sun, Yifei Su, Heyun Wang, Yuan Zhang, Xinghang Li, and Huaping Liu. Unified 4D world action modeling from video priors with asynchronous denoising.arXiv preprint arXiv:2604.26694, 2026 on PithPith's reviewarXiv
  9. [29] Yanjiang Guo, Lucy Xiaoyang Shi, Jianyu Chen, and Chelsea Finn. Ctrl-World: A controllable generative world model for robot manipulation. InInternational Conference on Learning Represen- tations, pages 6121–6138, 2026. URL https://proceedings.iclr.cc/paper_files/paper/2026/fil... on Pith
  10. [30] Danijar Hafner, Jurgis Pasukonis, Jimmy Ba, and Timothy Lillicrap. Mastering diverse control tasks through world models.Nature, 640(8059):647–653, 2025. doi: 10.1038/s41586-025-08744-2 on Pithdoi:10.1038/s41586-025-08744-2
  11. [31] Jinbang Huang, Zhiyuan Li, Yuanzhao Hu, Ran Qi, Yixin Xiao, Mark Coates, Tongtong Cao, Zhanguang Zhang, and Yingxue Zhang. RoboHarness: Memory-driven orchestration of heterogeneous robot policies for long-horizon planning.arXiv preprint arXiv:2607.18060, 2026 on PithPith's reviewarXiv
  12. [32] Wenlong Huang, Chen Wang, Ruohan Zhang, Yunzhu Li, Jiajun Wu, and Fei-Fei Li. VoxPoser: Composable 3D value maps for robotic manipulation with language models. In Jie Tan, Marc Toussaint, and Kourosh Darvish, editors,Proceedings of The 7th Conference on Robot Learning, volume ... on Pith
  13. [33] Wenlong Huang, Fei Xia, Ted Xiao, Harris Chan, Jacky Liang, Pete Florence, Andy Zeng, Jonathan Tompson, Igor Mordatch, Yevgen Chebotar, Pierre Sermanet, Tomas Jackson, Noah Brown, Linda Luu, Sergey Levine, Karol Hausman, and Brian Ichter. Inner monologue: Embodied reasoning th... on Pith
  14. [34] Wenlong Huang, Chen Wang, Yunzhu Li, Ruohan Zhang, and Li Fei-Fei. Rekep: Spatio-temporal reasoning of relationalkeypointconstraintsforroboticmanipulation. InPulkitAgrawal, OliverKroemer, andWolframBurgard, editors,Proceedings of The 8th Conference on Robot Learning, volume 27... on Pith
  15. [35] Brian Ichter, Anthony Brohan, Yevgen Chebotar, Chelsea Finn, Karol Hausman, Alexander Herzog, Daniel Ho, Julian Ibarz, Alex Irpan, Eric Jang, Ryan Julian, Dmitry Kalashnikov, Sergey Levine, Yao Lu, Carolina Parada, Kanishka Rao, Pierre Sermanet, Alexander T Toshev, Vincent Van... on Pith
  16. [36] Stephen James, Michael Bloesch, and Andrew J. Davison. Task-embedded control networks for few-shot imitation learning. In Aude Billard, Anca Dragan, Jan Peters, and Jun Morimoto, editors,Proceedings of The 2nd Conference on Robot Learning, volume 87 ofProceedings of Machine Le... on Pith
  17. [37] Joel Jang, Seonghyeon Ye, Zongyu Lin, Jiannan Xiang, Johan Bjorck, Yu Fang, Fengyuan Hu, Spencer Huang, Kaushil Kundalia, Yen-Chen Lin, Loïc Magne, Ajay Mandlekar, Avnish Narayan, You Liang Tan, Guanzhi Wang, Jing Wang, Qi Wang, Yinzhen Xu, Xiaohui Zeng, Kaiyuan Zheng, Ruijie ... on Pith
  18. [38] Tao Jiang, Tianyuan Yuan, Yicheng Liu, Chenhao Lu, Jianning Cui, Xiao Liu, Shuiqi Cheng, Jiyang Gao, Huazhe Xu, and Hang Zhao. Galaxy open-world dataset and G0 dual-system VLA model. In2026 IEEE International Conference on Robotics and Automation (ICRA), pages 20749–20756, Vie... on Pithdoi:10.1109/icra57385.2026.11696535
  19. [39] Dmitry Kalashnikov, Alex Irpan, Peter Pastor, Julian Ibarz, Alexander Herzog, Eric Jang, Deirdre Quillen, Ethan Holly, Mrinal Kalakrishnan, Vincent Vanhoucke, and Sergey Levine. Scalable deep reinforcement learning for vision-based robotic manipulation. In Aude Billard, Anca D... on Pith
  20. [40] Alexander Khazatsky, Karl Pertsch, Suraj Nair, Ashwin Balakrishna, Sudeep Dasari, Siddharth Karamcheti, Soroush Nasiriany, Mohan Kumar Srirama, Lawrence Yunliang Chen, Kirsty Ellis, Peter David Fagan, Joey Hejna, Masha Itkina, Marion Lepert, Yecheng Jason Ma, Patrick Tree Mill... on Pith
  21. [41] Moo Jin Kim, Karl Pertsch, Siddharth Karamcheti, Ted Xiao, Ashwin Balakrishna, Suraj Nair, Rafael Rafailov, Ethan P Foster, Pannag R Sanketi, Quan Vuong, Thomas Kollar, Benjamin Burchfiel, Russ Tedrake, Dorsa Sadigh, Sergey Levine, Percy Liang, and Chelsea Finn. OpenVLA: An op... on Pith
  22. [42] Moo Jin Kim, Yihuai Gao, Tsung-Yi Lin, Yen-Chen Lin, Yunhao Ge, Grace Lam, Percy Liang, Shuran Song, Ming-Yu Liu, Chelsea Finn, and Jinwei Gu. Cosmos policy: Fine-tuning video models for visuomotor control and planning. InInternational Conference on Learning Representa- tions,... on Pith
  23. [43] Michael Laskin, Luyu Wang, Junhyuk Oh, Emilio Parisotto, Stephen Spencer, Richie Steigerwald, DJ Strouse, Steven Hansen, Angelos Filos, Ethan Brooks, Maxime Gazeau, Himanshu Sahni, Satinder Singh, and Volodymyr Mnih. In-context reinforcement learning with algorithm distillatio... on Pith
  24. [44] Jonathan N. Lee, Annie Xie, Aldo Pacchiano, Yash Chandak, Chelsea Finn, Ofir Nachum, and Emma Brunskill. Supervised pretraining can learn in-context reinforcement learning. InAdvances in Neural Information Processing Systems, volume 36, pages 43057–43083, 2023. doi: 10.52202/0... on Pithdoi:10.52202/075280-1866
  25. [45] Fuhao Li, Wenxuan Song, Han Zhao, Jingbo Wang, Pengxiang Ding, Donglin Wang, Long Zeng, and Haoang Li. Spatial forcing: Implicit spatial representation alignment for vision-language-action model. InInternational Conference on Learning Representations, volume 2026, pages 132324... on Pith
  26. [46] Haozhan Li, Yuxin Zuo, Jiale Yu, Yuhao Zhang, Zhaohui Yang, Kaiyan Zhang, Xuekai Zhu, Yuchen Zhang, Tianxing Chen, Ganqu Cui, Dehui Wang, Dingxiang Luo, Yuchen Fan, Youbang Sun, Jia Zeng, Jiangmiao Pang, Shanghang Zhang, Yu Wang, Yao Mu, Bowen Zhou, and Ning Ding. SimpleVLA-RL... on Pith
  27. [47] Lin Li, Qihang Zhang, Yiming Luo, Shuai Yang, Ruilin Wang, Luyao Zhang, Mingrui Yu, Zelin Gao, Nan Xue, Boyu Zhou, Xing Zhu, Mingyu Ding, Yujun Shen, and Yinghao Xu. Causal World Modeling for Robot Control. InProceedings of Robotics: Science and Systems, Sydney, Australia, Jul... on Pithdoi:10.15607/rss.2026.xxii.016
  28. [48] Yaxuan Li, Yichen Zhu, Junjie Wen, Chaomin Shen, and Yi Xu. WorldEval: World model as real-world robot policies evaluator.arXiv preprint arXiv:2505.19017, 2025 on PithPith's reviewarXiv
  29. [49] Dingkang Liang, Xin Zhou, Wei Xu, Xingkui Zhu, Zhikang Zou, Xiaoqing Ye, Xiao Tan, and Xiang Bai. Pointmamba: A simple state space model for point cloud analysis. InAdvances in Neural Information Processing Systems, 2024 on Pith
  30. [50] Dingkang Liang, Tianrui Feng, Xin Zhou, Yumeng Zhang, Zhikang Zou, and Xiang Bai. Parameter-efficient fine-tuning in spectral domain for point cloud learning.IEEE transactions on pattern analysis and machine intelligence, 2025 on Pith
  31. [51] Dingkang Liang, Cheng Zhang, Xiaopeng Xu, Jianzhong Ju, Zhenbo Luo, and Xiang Bai. Cook and clean together: Teaching embodied agents for parallel task execution. InProceedings of the AAAI Conference on Artificial Intelligence, 2026 on Pith
  32. [52] Dingkang Liang, Dingyuan Zhang, Xin Zhou, Sifan Tu, Tianrui Feng, Xiaofan Li, Yumeng Zhang, Mingyang Du, Xiao Tan, and Xiang Bai. Unifuture: A 4d driving world model for future generation and perception. InIEEE International Conference on Robotics and Automation, 2026 on Pith
  33. [53] Jacky Liang, Wenlong Huang, Fei Xia, Peng Xu, Karol Hausman, Brian Ichter, Pete Florence, and Andy Zeng. Code as policies: Language model programs for embodied control. In2023 IEEE International Conference on Robotics and Automation (ICRA), pages 9493–9500. IEEE, 2023. doi: 10... on Pithdoi:10.1109/icra48891.2023.10160591
  34. [54] Haowen Liu, Xirui Li, Shaoxiong Yao, Peng Shi, Tianyi Zhou, Jia-Bin Huang, Furong Huang, and Jiayuan Mao. Guava: Distilling frontier VLMs into a compact agent through a robotic manipulation harness.arXiv preprint arXiv:2606.18363, 2026 on PithPith's reviewarXiv
  35. [55] Songming Liu, Lingxuan Wu, Bangguo Li, Hengkai Tan, Huayu Chen, Zhengyi Wang, Ke Xu, Hang Su, and Jun Zhu. RDT-1B: a diffusion foundation model for bimanual manipulation. InInternational Conference on Learning Representations, volume 2025, pages 29982–30009, 2025. URLhttps://p... on Pith
  36. [56] Jianlan Luo, Zheyuan Hu, Charles Xu, You Liang Tan, Jacob Berg, Archit Sharma, Stefan Schaal, Chelsea Finn, Abhishek Gupta, and Sergey Levine. SERL: A software suite for sample-efficient robotic reinforcement learning. In2024 IEEE International Conference on Robotics and Autom... on Pithdoi:10.1109/icra57147.2024.10610040
  37. [57] Jianlan Luo, Charles Xu, Jeffrey Wu, and Sergey Levine. Precise and dexterous robotic manipulation via human- in-the-loop reinforcement learning.Science Robotics, 10(105):eads5033, 2025. doi: 10.1126/scirobotics.ads5033 on Pithdoi:10.1126/scirobotics.ads5033
  38. [58] Yecheng Jason Ma, William Liang, Guanzhi Wang, De-An Huang, Osbert Bastani, Dinesh Jayaraman, Yuke Zhu, Linxi Fan, and Anima Anandkumar. Eureka: Human-level reward design via coding large language models. InInternational Conference on Learning Representations, volume 2024, pag... on Pith
  39. [59] NVIDIA, Johan Bjorck, Fernando Castañeda, Nikita Cherniadev, Xingye Da, Runyu Ding, Linxi “Jim” Fan, Yu Fang, Dieter Fox, Fengyuan Hu, Spencer Huang, Joel Jang, Zhenyu Jiang, Jan Kautz, Kaushil Kundalia, Lawrence Lao, Zhiqi Li, Zongyu Lin, Kevin Lin, Guilin Liu, Edith Llontop,... on PithPith's reviewarXiv
  40. [60] Hanbit Oh, Andrea M. Salcedo-Vázquez, Ixchel G. Ramirez-Alpizar, and Yukiyasu Domae. Robust instant policy: Leveraging student’s t-regression model for robust in-context imitation learning of robot manipulation. In2025 IEEE/RSJ International Conference on Intelligent Robots an... on Pith
  41. [61] Open X-Embodiment Collaboration. Open X-Embodiment: Robotic learning datasets and RT-X models. In2024 IEEE International Conference on Robotics and Automation (ICRA), pages 6892–6903, Yokohama, Japan, 2024. IEEE. doi: 10.1109/ICRA57147.2024.10611477. URLhttps://doi.org/10.1109... on Pithdoi:10.1109/icra57147.2024.10611477
  42. [62] OpenAI, Marcin Andrychowicz, Bowen Baker, Maciek Chociej, Rafal Józefowicz, Bob McGrew, Jakub Pachocki, Arthur Petron, Matthias Plappert, Glenn Powell, Alex Ray, Jonas Schneider, Szymon Sidor, Josh Tobin, Peter Welinder, Lilian Weng, and Wojciech Zaremba. Learning dexterous in... on Pithdoi:10.1177/0278364919887447
  43. [63] Alessio Palma, Indro Spinelli, Vignesh Prasad, Luca Scofano, Yufeng Jin, Georgia Chalvatzaki, and Fabio Galasso. Bimanual robot manipulation via multi-agent in-context learning. InConference on Robot Learning (CoRL), on Pith
  44. [64] Seyed Kamyar Seyed Ghasemipour, Ayzaan Wahid, Jonathan Tompson, Pannag Sanketi, and Igor Mordatch. Self-improving embodied foundation models. InAdvances in Neural Information Processing Systems, volume 38, on Pith
  45. [65] Lucy Xiaoyang Shi, Brian Ichter, Michael Robert Equi, Liyiming Ke, Karl Pertsch, Quan Vuong, James Tanner, Anna Walling, Haohuan Wang, Niccolo Fusai, Adrian Li-Bell, Danny Driess, Lachy Groom, Sergey Levine, and Chelsea Finn. Hi robot: Open-ended instruction following with hie... on Pith
  46. [66] Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik Narasimhan, and Shunyu Yao. Reflexion: Language agents with verbal reinforcement learning. InAdvances in Neural Information Processing Systems, volume 36, pages 8634–8652, 2023. doi: 10.52202/075280-0377 on Pithdoi:10.52202/075280-0377
  47. [67] Ishika Singh, Valts Blukis, Arsalan Mousavian, Ankit Goyal, Danfei Xu, Jonathan Tremblay, Dieter Fox, Jesse Thomason, and Animesh Garg. ProgPrompt: Generating situated robot task plans using large language models. In2023 IEEE International Conference on Robotics and Automation... on Pithdoi:10.1109/icra48891.2023.10161317
  48. [68] doi: 10.52202/085713-3762 on Pithdoi:10.52202/085713-3762
  49. [69] Kaustubh Sridhar, Souradeep Dutta, Dinesh Jayaraman, and Insup Lee. RICL: Adding in-context adaptability to pre-trained vision-language-action models. In Joseph Lim, Shuran Song, and Hae-Won Park, editors,Proceedings of The 9th Conference on Robot Learning, volume 305 ofProcee... on Pith
  50. [70] StarVLA Community. Starvla: A lego-like codebase for vision-language-action model developing.arXiv preprint arXiv:2604.05014, 2026 on PithPith's reviewarXiv
  51. [71] Marcel Torne Villasevil, Arhan Jain, Jiayi Yuan, Vidyaaranya Macha, Lars Lien Ankile, Anthony Simeonov, Pulkit Agrawal, and Abhishek Gupta. Robot learning with super-linear scaling. InProceedings of Robotics: Science and Systems, Los Angeles, CA, USA, June 2025. doi: 10.15607/... on Pithdoi:10.15607/rss.2025.xxi.025
  52. [72] Skild AI. Introducing S1: In-context learning for robotics. Skild AI blog,https://www.skild.ai/blogs/s1, August 2026 on Pith
  53. [73] Vitalis Vosylius and Edward Johns. Instant policy: In-context imitation learning via graph diffusion. InInterna- tional Conference on Learning Representations (ICLR), pages 35011–35030, 2025. URLhttps://proceedings. iclr.cc/paper_files/paper/2025/hash/5692c7dbc4abcaa50f9ce6098... on Pith
  54. [74] Guanzhi Wang, Yuqi Xie, Yunfan Jiang, Ajay Mandlekar, Chaowei Xiao, Yuke Zhu, Linxi Fan, and Anima Anandkumar. Voyager: An open-ended embodied agent with large language models.Transactions on Machine Learning Research, 2024. ISSN 2835-8856. URLhttps://openreview.net/forum?id=e... on Pith
  55. [75] Qi Wang, Tianyi Wang, Chengyang Li, Shikun Ban, Yurun Chen, Yizhong Ge, Jason Qin, Chengtai Li, and Wentao Zhu. Towards the harness of embodied agents.arXiv preprint arXiv:2608.11246, 2026 on PithPith's reviewarXiv
  56. [76] Brian Y. Tsui, Alan Y. Fang, and Tiffany J. Hwu. Demonstration-free robotic control via LLM agents. In 30 2026 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2026. URL https: //arxiv.org/abs/2601.20334 on PithPith's reviewarXiv
  57. [77] Zora Zhiruo Wang, Jiayuan Mao, Daniel Fried, and Graham Neubig. Agent workflow memory. InProceedings of the 42nd International Conference on Machine Learning, volume 267 ofProceedings of Machine Learning Research, pages 63897–63911. PMLR, 2025 on Pith
  58. [78] Philipp Wu, Alejandro Escontrela, Danijar Hafner, Pieter Abbeel, and Ken Goldberg. DayDreamer: World models for physical robot learning. In Karen Liu, Dana Kulic, and Jeff Ichnowski, editors,Proceedings of The 6th Conference on Robot Learning, volume 205 ofProceedings of Machi... on Pith
  59. [79] Wei Wu, Fan Lu, Yunnan Wang, Shuai Yang, Shi Liu, Fangjing Wang, Qian Zhu, He Sun, Yong Wang, Shuailei Ma, Yiyu Ren, Kejia Zhang, Hui Yu, Jingmei Zhao, Shuai Zhou, Zhenqi Qiu, Houlong Xiong, Ziyu Wang, Zechen Wang, Ran Cheng, Yong-Lu Li, Yongtao Huang, Xing Zhu, Yujun Shen, an... on PithPith's reviewarXiv
  60. [80] Xingyao Wang, Boxuan Li, Yufan Song, Frank F. Xu, Xiangru Tang, Mingchen Zhuge, Jiayi Pan, Yueqi Song, Bowen Li, Jaskirat Singh, Hoang Tran, Fuqiang Li, Ren Ma, Mingzhang Zheng, Bill Qian, Daniel Shao, Niklas Muennighoff, Yizhe Zhang, Binyuan Hui, Junyang Lin, Robert Brennan, ... on Pith
  61. [81] Hengyi Xie, Chenfei Yao, Xianjin Wu, Xuanyang Xi, Yiping Tang, Di Xu, Yingying Zhu, Dingkang Liang, Xiang Bai, and Han Ding. Turbovla: Real-time vision-language-action model at 32 hz on an rtx 4090 with <1 gb vram. arXiv preprint arXiv:2607.27205, 2026 on PithPith's reviewarXiv
  62. [82] Siyu Xu, Yunke Wang, Zijian Wang, Dihao Zhu, Chenghao Xia, Chengbin Du, Daochang Liu, Tao Huang, and Chang Xu. StellaVLA: In-context structured demonstration for generalizable vision-language-action models. arXiv preprint arXiv:2608.11671, 2026 on PithPith's reviewarXiv
  63. [83] John Yang, Carlos Jimenez, Alexander Wettig, Kilian Lieret, Shunyu Yao, Karthik Narasimhan, and Ofir Press. SWE-agent: Agent-computer interfaces enable automated software engineering. In A. Glober- son, L. Mackey, D. Belgrave, A. Fan, U. Paquet, J. Tomczak, and C. Zhang, edito... on Pith
  64. [84] Xianjin Wu, Dingkang Liang, Tianrui Feng, Kui Xia, Yumeng Zhang, Xiaofan Li, Xiao Tan, and Xiang Bai. Generation models know space: Unleashing implicit 3d priors for scene understanding. InProceedings of the European Conference on Computer Vision (ECCV), 2026 on Pith
  65. [85] Yandan Yang, Shuang Zeng, Tong Lin, Xinyuan Chang, Dekang Qi, Junjin Xiao, Haoyun Liu, Ronghan Chen, Yuzhi Chen, Dongjie Huo, Feng Xiong, Xing Wei, Zhiheng Ma, and Mu Xu. Abot-m0: Vla foundation model for robotic manipulation with action manifold learning.arXiv preprint arXiv:... on PithPith's reviewarXiv
  66. [86] Seonghyeon Ye, Yunhao Ge, Kaiyuan Zheng, Shenyuan Gao, Sihyun Yu, George Kurian, Suneel Indupuru, You Liang Tan, Chuning Zhu, Jiannan Xiang, Ayaan Malik, Kyungmin Lee, William Liang, Nadun Ranawaka, 31 Jiasheng Gu, Yinzhen Xu, Guanzhi Wang, Fengyuan Hu, Avnish Narayan, Johan B... on Pith
  67. [87] Yida Yin, Zekai Wang, Yuvan Sharma, Dantong Niu, Trevor Darrell, and Roei Herzig. In-context learning enables robot action prediction in LLMs. In2025 IEEE International Conference on Robotics and Automation (ICRA), pages 8972–8979, Atlanta, GA, USA, 2025. IEEE. doi: 10.1109/IC... on Pithdoi:10.1109/icra55743.2025.11128807
  68. [88] Wenhao Yu, Nimrod Gileadi, Chuyuan Fu, Sean Kirmani, Kuang-Huei Lee, Montserrat Gonzalez Arenas, Hao-Tien Lewis Chiang, Tom Erez, Leonard Hasenclever, Jan Humplik, Brian Ichter, Ted Xiao, Peng Xu, Andy Zeng, Tingnan Zhang, Nicolas Heess, Dorsa Sadigh, Jie Tan, Yuval Tassa, and... on Pith
  69. [89] Lishan Yang, Wenxuan Song, Xi Wang, Pingyue Sheng, Zheng Fang, Ziyang Zhou, Junjie He, Haodong Yan, Jiayi Chen, Nan Sun, Qiao Sun, Pengwei Wang, Lingqiao Liu, Yan Wang, Yuxiang Gao, Feras Dayoub, and Haoang Li. 4d-wam: Infusing spatiotemporal awareness into world action models... on PithPith's reviewarXiv
  70. [90] Borong Zhang, Jiahao Li, Jiachen Shen, Yuhao Zhang, Yishuai Cai, Yuanpei Chen, Juntao Dai, Jiaming Ji, and Yaodong Yang. VLA-arena: An open-source framework for benchmarking vision-language-action models. InProceedings of the 43rd International Conference on Machine Learning, ... on Pith
  71. [91] Qizheng Zhang, Changran Hu, Shubhangi Upasani, Boyuan Ma, Fenglu Hong, Vamsidhar Kamanuru, Jay Rainton, Chen Wu, Mengmeng Ji, Hanchen Li, Urmish Thakker, James Y. Zou, and Kunle Olukotun. Agentic context engineering: Evolving contexts for self-improving language models. InInte... on Pith
  72. [92] Wenbo Zhang, Kaixuan Wang, Yutao Ouyang, Xiaoyu Huang, Liyang Li, Kailun Su, Weiyang Jin, Wenhao Chai, Haotian Liang, Zhiyang Dou, Yue Chen, and Tianxing Chen. An unexpected robot policy: Early evaluations of GPT-6 Astra on RoboDojo and beyond.arXiv preprint arXiv:2609.24170, 2026 on PithPith's reviewarXiv
  73. [93] Yixian Zhang, Huanming Zhang, Feng Gao, Xiao Li, Zhihao Liu, Yi Nie, Chunyang Zhu, Jiaxing Qiu, Yuchen Yan, Jiyuan Liu, Wenhao Tang, Jiaji Rao, Zhengru Fang, Changxu Wei, Yu Wang, Wenbo Ding, and Chao Yu. Harness VLA: Steering frozen VLAs into reliable manipulation primitives ... on PithPith's reviewarXiv
  74. [94] Andrew Zhao, Daniel Huang, Quentin Xu, Matthieu Lin, Yong-Jin Liu, and Gao Huang. ExpeL: LLM agents are experiential learners. InProceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 19632–19642. Association for the Advancement of Artificial Intellige... on Pithdoi:10.1609/aaai.v38i17.29936
  75. [95] Tianyuan Yuan, Zibin Dong, Yicheng Liu, and Hang Zhao. Fast-WAM: Do world action models need test-time future imagination?arXiv preprint arXiv:2603.16666, 2026 on PithPith's reviewarXiv
  76. [96] Jinliang Zheng, Jianxiong Li, Zhihao Wang, Dongxiu Liu, Xirui Kang, Yuchun Feng, Yinan Zheng, Jiayin Zou, Yilun Chen, Jia Zeng, Tai Wang, Ya-Qin Zhang, Jingjing Liu, and Xianyuan Zhan. X-VLA: Soft-prompted transformer as scalable cross-embodiment vision-language-action model. ... on Pith
  77. [97] Jiaming Zhou, Qihang Zhang, Gangwei Xu, Cunxin Fan, Yujie Zhao, Ruilin Wang, Yiming Luo, Shuai Yang, Xing Zhu, Yujun Shen, Junwei Liang, and Yinghao Xu. Zero-WAM: In-context world-action modeling from human videos for open-ended task generalization.arXiv preprint arXiv:2608.26... on PithPith's reviewarXiv
  78. [98] Xin Zhou, Dingkang Liang, Sifan Tu, Xiwu Chen, Yikang Ding, Dingyuan Zhang, Feiyang Tan, Hengshuang Zhao, and Xiang Bai. Hermes: A unified self-driving world model for simultaneous 3d scene understanding and generation. InProceedings of the IEEE/CVF International Conference on... on Pith
  79. [99] Xin Zhou, Dingkang Liang, Xiwu Chen, Feiyang Tan, Dingyuan Zhang, Hengshuang Zhao, and Xiang Bai. Hermes++: Toward a unified driving world model for 3d scene understanding and generation.IEEE Transactions on Pattern Analysis and Machine Intelligence, 2026. doi: 10.1109/TPAMI.2... on Pithdoi:10.1109/tpami.2026.3735606
  80. [100] Xin Zhou, Zongchuang Zhao, Zhibo Yang, Mingsheng Li, Humen Zhong, Shuai Bai, Du Chu, Ruizhe Chen, Zhaohai Li, Jun Tang, Qiuyue Wang, Mingkun Yang, Jiazhao Zhang, Dayiheng Liu, Dingkang Liang, and Xiang Bai. Qwen-drive-1.0: An initial step towards a vision-language foundation m... on PithPith's reviewarXiv

The first 100 of 110 references are listed here. All of them are in this paper's citation record.

Challenges and discussion

Disagree with the review, or know something it missed? Leave a comment, or challenge this review formally: a challenge becomes a permanent entry in this paper's record, and its decision is published here.

No comments yet.

Sign in to comment