arxiv: 2604.13800 · v1 · submitted 2026-04-15 · 💻 cs.RO

Recognition: unknown

EmbodiedClaw: Conversational Workflow Execution for Embodied AI Development

Xueyang Zhou , Yihan Sun , Xijie Gong , Guiyao Tie , Pan Zhou , Lichao Sun , Yongchao Chen

Authors on Pith no claims yet

Pith reviewed 2026-05-10 13:31 UTC · model grok-4.3

classification 💻 cs.RO

keywords embodied AIconversational agentsworkflow automationrobotics developmentAI research toolsenvironment constructionbenchmark transformation

0 comments

The pith

A conversational agent automates the full workflow of embodied AI research from user goals alone.

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

Embodied AI work has shifted from single-task policies to multi-task, multi-scene, multi-model projects, which multiplies the engineering steps for setting up environments, collecting data, training, and testing. The paper presents EmbodiedClaw as a system that lets researchers state goals and limits in ordinary conversation and then automatically plans and runs those steps as executable skills. Experiments covering complete workflows, specific capabilities, human researchers, and ablations indicate lower manual effort together with higher consistency and reproducibility. If this holds, research teams could spend less time on custom scripting and more on the scientific questions.

Core claim

EmbodiedClaw is a conversational agent that turns high-level user instructions into planned sequences of embodied-AI development skills, covering environment creation and revision, benchmark transformation, trajectory synthesis, model evaluation, and asset expansion. When evaluated on end-to-end tasks, targeted capability tests, researcher studies, and component ablations, the agent lowers manual engineering time while raising executability, consistency, and reproducibility of the resulting pipelines.

What carries the argument

EmbodiedClaw, the conversational agent that maps user dialogue directly into executable workflow skills for environment, data, training, and evaluation steps.

If this is right

Environment revision and benchmark changes can be performed by stating the desired change rather than editing code by hand.
Trajectory collection and model evaluation become reproducible across different researchers because the steps are generated from the same dialogue record.
The overall development loop for multi-model, multi-scene projects shortens because each stage is executed by the same automated pipeline.
Human studies show researchers complete the same workflow with measurably less time spent on implementation details.

Where Pith is reading between the lines

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

The same conversational-execution pattern could be reused for non-embodied AI pipelines that also involve repetitive data and evaluation steps.
Standardized skill libraries built this way might allow different research groups to share and compose workflows without exchanging custom scripts.
If planning errors decrease with more examples, the system could eventually handle novel combinations of environments and models with little extra training.

Load-bearing premise

The conversational agent can interpret high-level goals and constraints accurately enough to plan and run complex embodied workflows without needing frequent human fixes.

What would settle it

A head-to-head trial on a multi-step task such as benchmark transformation in which the number of human interventions or final error rate is higher with EmbodiedClaw than with standard manual scripting.

Figures

Figures reproduced from arXiv: 2604.13800 by Guiyao Tie, Lichao Sun, Pan Zhou, Xijie Gong, Xueyang Zhou, Yihan Sun, Yongchao Chen.

**Figure 1.** Figure 1: Overview of EmbodiedClaw capabilities. The framework supports batch simulation environment synthesis, including automatic scene creation and controllable scene editing, batch trajectory collection for embodied manipulation tasks, and adaptive export of embodied data to HDF5, LeRobot, RLDS, and video formats. It further enables model deployment with vision-language and world models, model training through i… view at source ↗

**Figure 2.** Figure 2: EmbodiedClaw pipeline. The system maps user requests to intent-aware workflows, grounds abstract skills to platform-specific actions, and verifies each step in a closed loop. where o D t may contain collected trajectories, relabeled transitions, and transformed training samples, ∆task measures how well the synthesized data supports the target task, ∆stab measures consistency and robustness across repeated … view at source ↗

**Figure 3.** Figure 3: Efficiency comparison across four embodied AI development tasks. The four subfigures report the efficiency of Layperson, Expert, Claude Code, and EmbodiedClaw on image-tosimulation, revise-to-simulation, simulation-to-data, and VLA evaluation, respectively. Each subfigure compares the time cost of completing the corresponding task, where lower is better. We evaluate four representative tasks: (1) Environm… view at source ↗

**Figure 4.** Figure 4: Task completion rate comparison across four embodied AI development tasks. The four subfigures report the accuracy of Layperson, Expert, Claude Code, and EmbodiedClaw on image-to-simulation, revise-to-simulation, simulation-to-data, and VLA evaluation, respectively. Higher is better. We further evaluate EmbodiedClaw in terms of task completion rate, which measures whether an embodied development workflow i… view at source ↗

**Figure 5.** Figure 5: Case study of batch editing for existing environments. EmbodiedClaw supports parallel editing of existing environments from natural-language instructions, including inserting new objects, modifying irrelevant viewpoints, and adjusting scene lighting conditions. The three case studies in [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗

**Figure 6.** Figure 6: Case studies of EmbodiedClaw for environment construction and editing. Given natural-language instructions, EmbodiedClaw translates user requests into executable workflows and completes the corresponding environment construction and editing procedures automatically. AI2-THOR [12], Habitat [13, 14], ManiSkill3 [18], Isaac Gym [16], SAPIEN [15], and VirtualHome [40]—each providing distinct interfaces, asset… view at source ↗

**Figure 7.** Figure 7: Case study of the VLA evaluation workflow. EmbodiedClaw executes VLA evaluation as a complete workflow, covering environment preparation, trajectory collection, data transformation, model deployment, and downstream evaluation. plausible interactive environments. Despite these advances, existing work remains fragmented and mutually incompatible: each system targets a single pipeline stage, exposes distinct … view at source ↗

read the original abstract

Embodied AI research is increasingly moving beyond single-task, single-environment policy learning toward multi-task, multi-scene, and multi-model settings. This shift substantially increases the engineering overhead and development time required for stages such as evaluation environment construction, trajectory collection, model training, and evaluation. To address this challenge, we propose a new paradigm for embodied AI development in which users express goals and constraints through conversation, and the system automatically plans and executes the development workflow. We instantiate this paradigm with EmbodiedClaw, a conversational agent that turns high-frequency, high-cost embodied research activities, including environment creation and revision, benchmark transformation, trajectory synthesis, model evaluation, and asset expansion, into executable skills. Experiments on end-to-end workflow tasks, capability-specific evaluations, human researcher studies, and ablations show that EmbodiedClaw reduces manual engineering effort while improving executability, consistency, and reproducibility. These results suggest a shift from manual toolchains to conversationally executable workflows for embodied AI development.

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.

Desk Editor's Note private letter to a colleague

EmbodiedClaw applies conversation to embodied workflow automation but needs stronger evidence on actual effort reduction.

read the letter

The punchline is that EmbodiedClaw presents a conversational agent designed to automate the engineering-heavy parts of embodied AI development, but the reported experiments leave open how much ongoing human oversight is still required for reliable results. What stands out as new is the focus on turning specific embodied research activities—environment creation and revision, benchmark transformation, trajectory synthesis, model evaluation, and asset expansion—into skills that a conversational system can plan and execute. While conversational coding tools exist elsewhere, applying this directly to the multi-task, multi-scene embodied setting with these concrete tasks appears fresh. The paper handles the motivation well by describing the shift from single-task to more complex embodied setups and the resulting overhead. Running evaluations across end-to-end workflows, capability-specific tests, studies with human researchers, and ablations is a solid structure for checking the idea from multiple sides. The soft spots come in the execution details. The abstract states that the system reduces manual engineering effort and improves executability, consistency, and reproducibility, yet it provides no metrics, baselines, or information on intervention frequency. If the agent frequently needs corrections on complex steps like environment revision or trajectory synthesis, the claimed benefits would not hold up as strongly. The weakest assumption here is that high-level user goals can be interpreted and carried out autonomously without substantial back-and-forth. This work is aimed at embodied AI researchers dealing with repetitive setup tasks. Someone looking for ideas on agent-assisted research pipelines could find the framing and skill breakdown useful as a starting point. Overall, the paper shows clear thinking on the problem and honest engagement with the practical challenges. It deserves a serious referee because the topic is timely and the approach has potential, though the evaluation section would likely need expansion with quantitative data on human effort before publication.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes EmbodiedClaw, a conversational agent that allows users to express high-level goals and constraints for embodied AI development tasks (environment creation/revision, benchmark transformation, trajectory synthesis, model evaluation, asset expansion) and automatically plans and executes the corresponding workflows. It reports results from end-to-end workflow tasks, capability-specific evaluations, human researcher studies, and ablations, claiming that the system reduces manual engineering effort while improving executability, consistency, and reproducibility compared to traditional manual toolchains.

Significance. If the experimental claims hold with proper quantification, the work could meaningfully lower the engineering barrier for multi-task, multi-environment embodied AI research by shifting from manual scripting to conversational workflow execution. This addresses a genuine practical bottleneck in the field and, if validated, would represent a useful systems contribution for accelerating development cycles.

major comments (2)

[Experiments] Experiments section (end-to-end workflow tasks and human studies): The central claim that EmbodiedClaw reduces manual engineering effort and improves executability requires evidence that the agent can handle error-prone steps (environment revision, benchmark transformation, trajectory synthesis) with minimal human correction. The reported results do not quantify intervention frequency, failure recovery rates, or effort reduction against an explicit manual baseline, so the asserted improvements in executability and reduced overhead do not yet follow from the presented data.
[Human studies and ablations] Human researcher studies and ablations: These evaluations are described as showing positive outcomes, but without reported metrics (e.g., task completion time, number of corrections per workflow, inter-rater reliability, or ablation-specific deltas with error bars), it is impossible to assess whether the conversational interface delivers reliable autonomy or merely shifts effort to prompt engineering.

minor comments (1)

[Abstract] Abstract: The summary of experimental outcomes asserts improvements without any high-level metrics, baselines, or sample sizes; adding one sentence summarizing key quantitative findings would improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive review and for recognizing the potential significance of EmbodiedClaw in addressing engineering bottlenecks in embodied AI research. We address each major comment below and have revised the manuscript to incorporate the requested quantifications and metrics.

read point-by-point responses

Referee: [Experiments] Experiments section (end-to-end workflow tasks and human studies): The central claim that EmbodiedClaw reduces manual engineering effort and improves executability requires evidence that the agent can handle error-prone steps (environment revision, benchmark transformation, trajectory synthesis) with minimal human correction. The reported results do not quantify intervention frequency, failure recovery rates, or effort reduction against an explicit manual baseline, so the asserted improvements in executability and reduced overhead do not yet follow from the presented data.

Authors: We agree that explicit quantification of intervention frequency, failure recovery, and effort reduction against a manual baseline would strengthen the central claims. In the revised manuscript, we have expanded the end-to-end workflow tasks subsection to report the average number of human interventions per task for error-prone steps, autonomous failure recovery rates, and a direct comparison of total researcher effort (in hours) using EmbodiedClaw versus a traditional manual toolchain on identical tasks performed by the same participants. These additions provide the missing evidence that the system handles such steps with minimal correction and reduces overhead. revision: yes
Referee: [Human studies and ablations] Human researcher studies and ablations: These evaluations are described as showing positive outcomes, but without reported metrics (e.g., task completion time, number of corrections per workflow, inter-rater reliability, or ablation-specific deltas with error bars), it is impossible to assess whether the conversational interface delivers reliable autonomy or merely shifts effort to prompt engineering.

Authors: We acknowledge that the original presentation lacked the specific metrics needed for rigorous assessment. We have revised the Human Researcher Studies and Ablations sections to include average task completion times, number of corrections per workflow, inter-rater reliability coefficients for the qualitative evaluations, and ablation results reported as mean deltas with standard error bars. The updated metrics demonstrate reliable autonomy from the conversational interface, with performance gains not attributable solely to prompt engineering. revision: yes

Circularity Check

0 steps flagged

No circularity; empirical claims rest on direct experiments without self-referential definitions or fitted predictions

full rationale

The paper introduces EmbodiedClaw as an instantiation of a conversational workflow paradigm for embodied AI tasks and evaluates it via end-to-end experiments, capability-specific tests, human studies, and ablations. No mathematical derivation chain, equations, or parameters exist in the provided text. Claims of reduced engineering effort and improved executability are presented as outcomes of the implemented system and measured results rather than reductions by construction to inputs or self-citations. This is a systems paper whose central assertions are externally falsifiable through the described experiments and do not rely on self-definitional loops or imported uniqueness theorems.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The central claim rests on the domain assumption that conversational interfaces can capture and translate research goals into reliable executable workflows; no free parameters or invented physical entities are introduced.

axioms (1)

domain assumption Conversational interfaces can effectively capture high-level research goals and constraints and translate them into executable embodied AI workflows.
This premise underpins the entire proposed paradigm and is required for the agent to replace manual engineering.

invented entities (1)

EmbodiedClaw no independent evidence
purpose: Conversational agent that converts user goals into automated skills for environment creation, trajectory synthesis, model evaluation, and related tasks.
The system itself is the primary contribution; no independent evidence outside the paper is provided.

pith-pipeline@v0.9.0 · 5489 in / 1224 out tokens · 55028 ms · 2026-05-10T13:31:15.589991+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

50 extracted references · 22 canonical work pages · 5 internal anchors

[2]

Aligning cyber space with physical world: A comprehensive survey on embodied AI,

Minghua Liu et al. Aligning cyber space with physical world: A comprehensive survey on embodied AI. arXiv preprint arXiv:2407.06886, 2023

work page arXiv 2023
[3]

Multi-agent embodied ai: Advances and future directions.arXiv preprint arXiv:2505.05108,

Zhaocan Feng et al. Multi-agent embodied AI: Advances and future directions.arXiv preprint arXiv:2505.05108, 2025

work page arXiv 2025
[4]

arXiv preprint arXiv:2508.13073 (2025)

Rui Shao, Wei Li, Lingsen Zhang, Renshan Zhang, Zhiyang Liu, Ran Chen, and Liqiang Nie. Large vlm- based vision-language-action models for robotic manipulation: A survey.arXiv preprint arXiv:2508.13073, 2025

work page arXiv 2025
[5]

Embodied intelligence: A synergy of morphology, action, perception and learning.ACM Computing Surveys, 57(7):1–36, 2025

Huaping Liu, Di Guo, and Angelo Cangelosi. Embodied intelligence: A synergy of morphology, action, perception and learning.ACM Computing Surveys, 57(7):1–36, 2025

2025
[6]

Machine learning meets advanced robotic manipulation.Information Fusion, 105:102221, 2024

Saeid Nahavandi, Roohallah Alizadehsani, Darius Nahavandi, Chee Peng Lim, Kevin Kelly, and Fernando Bello. Machine learning meets advanced robotic manipulation.Information Fusion, 105:102221, 2024

2024
[7]

Foundation models defining a new era in vision: a survey and outlook.IEEE Transactions on Pattern Analysis and Machine Intelligence, 47(4): 2245–2264, 2025

Muhammad Awais, Muzammal Naseer, Salman Khan, Rao Muhammad Anwer, Hisham Cholakkal, Mubarak Shah, Ming-Hsuan Yang, and Fahad Shahbaz Khan. Foundation models defining a new era in vision: a survey and outlook.IEEE Transactions on Pattern Analysis and Machine Intelligence, 47(4): 2245–2264, 2025

2025
[8]

Orchestrating embodied systems through the embodied context protocol: Motivation, progress, and directions.Research, 2025

Fuyu Ma et al. Orchestrating embodied systems through the embodied context protocol: Motivation, progress, and directions.Research, 2025. doi: 10.34133/research.1047

work page doi:10.34133/research.1047 2025
[9]

Robotwin: Dual-arm robot benchmark with generative digital twins

Yao Mu, Tianxing Chen, Zanxin Chen, Shijia Peng, Zhiqian Lan, Zeyu Gao, Zhixuan Liang, Qiaojun Yu, Yude Zou, Mingkun Xu, et al. Robotwin: Dual-arm robot benchmark with generative digital twins. In Proceedings of the computer vision and pattern recognition conference, pages 27649–27660, 2025

2025
[10]

GenSim2: Scaling robot data generation with multi-modal and reasoning LLMs.arXiv preprint arXiv:2410.03645,

Pu Hua, Minghuan Liu, Annabella Macaluso, Yunfeng Lin, Weinan Zhang, Huazhe Xu, and Lirui Wang. Gensim2: Scaling robot data generation with multi-modal and reasoning llms.arXiv preprint arXiv:2410.03645, 2024

work page arXiv 2024
[11]

Robogen: Towards unleashing infinite data for automated robot learning via generative simulation.arXiv preprint arXiv:2311.01455, 2023

Yufei Wang, Zhou Xian, Feng Chen, Tsun-Hsuan Wang, Yian Wang, Katerina Fragkiadaki, Zackory Erickson, David Held, and Chuang Gan. RoboGen: Towards unleashing infinite data for automated robot learning via generative simulation.arXiv preprint arXiv:2311.01455, 2023

work page arXiv 2023
[12]

AI2-THOR: An Interactive 3D Environment for Visual AI

Eric Kolve, Roozbeh Mottaghi, Winson Han, Eli VanderBilt, Luca Weihs, Alvaro Herrasti, Daniel Gordon, Yuke Zhu, Abhinav Gupta, and Ali Farhadi. AI2-THOR: An interactive 3D environment for visual AI. arXiv preprint arXiv:1712.05474, 2017

work page internal anchor Pith review arXiv 2017
[13]

Habitat: A platform for embodied AI research

Manolis Savva, Abhishek Kadian, Oleksandr Maksymets, Yili Zhao, Erik Wijmans, Bhavana Jain, Julian Straub, Jia Liu, Vladlen Koltun, Jitendra Malik, Devi Parikh, and Dhruv Batra. Habitat: A platform for embodied AI research. InProceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 9339–9347, 2019

2019
[14]

Habitat 2.0: Training home assistants to rearrange their habitat

Andrew Szot, Alexander Clegg, Eric Undersander, Erik Wijmans, Yili Zhao, John Turner, Noah Maestre, Mustafa Mukadam, Devendra Singh Chaplot, Oleksandr Maksymets, Aaron Gokaslan, Vladimír V ondruš, Sameer Dharur, Franziska Meier, Wojciech Galuba, Angel X Chang, Zsolt Kira, Vladlen Koltun, Jitendra Malik, Manolis Savva, and Dhruv Batra. Habitat 2.0: Trainin...

2021
[15]

SAPIEN: A simulated part-based interactive environment

Fanbo Xiang, Yuzhe Qin, Kaichun Mo, Yikuan Xia, Hao Zhu, Fangchen Liu, Minghua Liu, Hanxiao Jiang, Yifu Yuan, He Wang, Li Yi, Angel X Chang, Leonidas J Guibas, and Hao Su. SAPIEN: A simulated part-based interactive environment. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 11097–11107, 2020

2020
[16]

Isaac gym: High performance GPU-based physics simulation for robot learning

Viktor Makoviychuk, Leszek Wawrzyniak, Yunrong Guo, Michelle Lu, Kier Storey, Miles Macklin, David Hoeller, Nikita Rudin, Arthur Allshire, Ankur Handa, and Gavriel State. Isaac gym: High performance GPU-based physics simulation for robot learning. InAdvances in Neural Information Processing Systems (NeurIPS), Datasets & Benchmarks Track, 2021

2021
[17]

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.arXiv preprint arXiv:2009.12293, 2020. 11

work page internal anchor Pith review arXiv 2009
[18]

Maniskill3: Gpu parallelized robotics simulation and rendering for generalizable embodied ai.arXiv preprint arXiv:2410.00425, 2024

Stone Tao, Fanbo Xiang, Arth Shukla, Yuzhe Qin, Xander Hinrichsen, Xiaodi Yuan, Chen Bao, Xinsong Lin, Yulin Liu, Tse-kai Chan, et al. ManiSkill3: GPU parallelized robotics simulation and rendering for generalizable embodied AI.arXiv preprint arXiv:2410.00425, 2024

work page arXiv 2024
[19]

ProcTHOR: Large-scale embodied AI using procedural generation

Matt Deitke, Eli VanderBilt, Alvaro Herrasti, Luca Weihs, Kiana Ehsani, Jordi Salvador, Winson Han, Eric Kolve, Aniruddha Kembhavi, and Roozbeh Mottaghi. ProcTHOR: Large-scale embodied AI using procedural generation. InAdvances in Neural Information Processing Systems (NeurIPS), 2022

2022
[20]

Holodeck: Language guided generation of 3D embodied AI environments

Yue Yang, Fan-Yun Sun, Luca Weihs, Eli VanderBilt, Alvaro Herrasti, Winson Han, Jiajun Wu, Nick Haber, Ranjay Krishna, Lingjie Liu, Chris Callison-Burch, Mark Yatskar, Aniruddha Kembhavi, and Christopher Clark. Holodeck: Language guided generation of 3D embodied AI environments. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recog...

2024
[21]

iGibson 2.0: Object-centric simulation for robot learning of everyday household tasks

Chengshu Li, Fei Xia, Roberto Martín-Martín, Michael Lingelbach, Sanjana Srivastava, Bhuvan Bhargava, Kim Leong, Feiran An, Carl Romero, Shyamal Shivkumar, et al. iGibson 2.0: Object-centric simulation for robot learning of everyday household tasks. InConference on Robot Learning (CoRL), 2022

2022
[22]

Bear, Dan Gutfreund, David Cox, Antonio Torralba, James J

Chuang Gan, Jeremy Schwartz, Seth Alter, Martin Schrimpf, James Traer, Julian De Freitas, Jonas Kubilius, Abhishek Bhandwaldar, Nick Haber, Megumi Sano, et al. ThreeDWorld: A platform for interactive multi- modal physical simulation.arXiv preprint arXiv:2007.04954, 2020

work page arXiv 2007
[23]

RoboCasa: Large-Scale Simulation of Everyday Tasks for Generalist Robots

Soroush Nasiriany, Abhiram Maddukuri, Lance Zhang, Adeet Parakh, Andrew Lo, Abhishek Joshi, Ajay Mandlekar, and Yuke Zhu. RoboCasa: Large-scale simulation of everyday tasks for generalist robots. arXiv preprint arXiv:2406.02523, 2024

work page internal anchor Pith review arXiv 2024
[24]

Imitating task and motion planning with visuomotor transformers

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, 2023

work page arXiv 2023
[25]

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.arXiv preprint arXiv:2310.17596, 2023

work page arXiv 2023
[26]

Maniskill2: A unified benchmark for generalizable manipulation skills.arXiv preprint arXiv:2302.04659, 2023

Jiayuan Gu, Fanbo Xiang, Xuanlin Li, Zhan Ling, Xiqiang Liu, Tongzhou Mu, Yihe Tang, Stone Tao, Xinyue Wei, Yunchao Yao, et al. Maniskill2: A unified benchmark for generalizable manipulation skills. arXiv preprint arXiv:2302.04659, 2023

work page arXiv 2023
[27]

Rlbench: The robot learning benchmark & learning environment.IEEE Robotics and Automation Letters, 5(2):3019–3026, 2020

Stephen James, Zicong Ma, David Rovick Arrojo, and Andrew J Davison. Rlbench: The robot learning benchmark & learning environment.IEEE Robotics and Automation Letters, 5(2):3019–3026, 2020

2020
[28]

Behavior-1k: A benchmark for embodied ai with 1,000 everyday activities and realistic simulation

Chengshu Li, Ruohan Zhang, Josiah Wong, Cem Gokmen, Sanjana Srivastava, Roberto Martín-Martín, Chen Wang, Gabrael Levine, Michael Lingelbach, Jiankai Sun, et al. Behavior-1k: A benchmark for embodied ai with 1,000 everyday activities and realistic simulation. InConference on Robot Learning, pages 80–93. PMLR, 2023

2023
[29]

Claude Code

Anthropic. Claude Code. https://claude.com/product/claude-code, 2026. Accessed: 2026-04- 01

2026
[30]

OpenAI Codex

OpenAI. OpenAI Codex. https://openai.com/index/openai-codex/, 2026. Accessed: 2026-04- 01

2026
[31]

Gemini CLI.https://geminicli.com/, 2026

Google. Gemini CLI.https://geminicli.com/, 2026. Accessed: 2026-04-01

2026
[32]

OpenClaw — personal AI assistant

OpenClaw Community. OpenClaw — personal AI assistant. https://openclaw.ai, 2026. Accessed: 2026-04-01

2026
[33]

Language models as zero-shot planners: Extracting actionable knowledge for embodied agents

Wenlong Huang, Pieter Abbeel, Deepak Pathak, and Igor Mordatch. Language models as zero-shot planners: Extracting actionable knowledge for embodied agents. 2022

2022
[34]

Code as policies: Language model programs for embodied control

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. 2023

2023
[35]

Voyager: An Open-Ended Embodied Agent with Large Language Models

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.arXiv preprint arXiv:2305.16291, 2023

work page internal anchor Pith review Pith/arXiv arXiv 2023
[36]

Danny Driess, Fei Xia, Mehdi S. M. Sajjadi, Corey Lynch, Aakanksha Chowdhery, Brian Ichter, Ayzaan Wahid, Jonathan Tompson, Quan Vuong, Tianhe Yu, et al. PaLM-E: An embodied multimodal language model. 2023. 12

2023
[37]

RT-2: Vision-language-action models transfer web knowledge to robotic control

Brianna Zitkovich, Tianhe Yu, Sichun Xu, Peng Xu, Ted Xiao, Fei Xia, Jialin Wu, Paul Wohlhart, Stefan Welker, Ayzaan Wahid, et al. RT-2: Vision-language-action models transfer web knowledge to robotic control. 2023

2023
[38]

Libero: Benchmarking knowledge transfer for lifelong robot learning.Advances in Neural Information Processing Systems, 36:44776–44791, 2023

Bo Liu et al. Libero: Benchmarking knowledge transfer for lifelong robot learning.Advances in Neural Information Processing Systems, 36:44776–44791, 2023

2023
[39]

Evaluating Real-World Robot Manipulation Policies in Simulation

Xuanlin Li et al. Evaluating real-world robot manipulation policies in simulation.arXiv preprint arXiv:2405.05941, 2024

work page internal anchor Pith review arXiv 2024
[40]

VirtualHome: Simulating household activities via programs

Xavier Puig, Kevin Ra, Marko Boben, Jiaman Li, Tingwu Wang, Sanja Fidler, and Antonio Torralba. VirtualHome: Simulating household activities via programs. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 8494–8502, 2018

2018
[41]

Foundation models in robotics: Applications, challenges, and the future.The International Journal of Robotics Research, 44(5):701–739, 2025

Roya Firoozi, Johnathan Tucker, Stephen Tian, Anirudha Majumdar, Jiankai Sun, Weiyu Liu, Yuke Zhu, Shuran Song, Ashish Kapoor, Karol Hausman, et al. Foundation models in robotics: Applications, challenges, and the future.The International Journal of Robotics Research, 44(5):701–739, 2025

2025
[42]

Generative artificial intelligence in robotic manipulation: A survey.arXiv preprint arXiv:2503.03464, 2025

Kun Zhang, Peng Yun, Jun Cen, Junhao Cai, Didi Zhu, Hangjie Yuan, Chao Zhao, Tao Feng, Michael Yu Wang, Qifeng Chen, et al. Generative artificial intelligence in robotic manipulation: A survey.arXiv preprint arXiv:2503.03464, 2025

work page arXiv 2025
[43]

Sage: Scalable agentic 3d scene generation for embodied ai, 2026

Hongchi Xia, Xuan Li, Zhaoshuo Li, Qianli Ma, Jiashu Xu, Ming-Yu Liu, Yin Cui, Tsung-Yi Lin, Wei-Chiu Ma, Shenlong Wang, Shuran Song, and Fangyin Wei. SAGE: Scalable agentic 3D scene generation for embodied AI.arXiv preprint arXiv:2602.10116, 2026

work page arXiv 2026
[44]

EmbodiedGen: Towards a generative 3D world engine for embodied intelligence, 2025

Xinjie Wang, Liu Liu, Yu Cao, Ruiqi Wu, Wenkang Qin, Dehui Wang, Wei Sui, and Zhizhong Su. EmbodiedGen: Towards a generative 3D world engine for embodied intelligence, 2025

2025
[45]

MuJoCo: A physics engine for model-based control

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

2012
[46]

M 2Diffuser: Diffusion-based trajectory optimization for mobile manipulation in 3D scenes.arXiv preprint, 2024

Zhou Xian et al. M 2Diffuser: Diffusion-based trajectory optimization for mobile manipulation in 3D scenes.arXiv preprint, 2024. Available athttps://m2diffuser.github.io

2024
[47]

arXiv preprint arXiv:2505.02836 (2025)

Lu Ling et al. Scenethesis: A language and vision agentic framework for 3D scene generation.arXiv preprint arXiv:2505.02836, 2025

work page arXiv 2025
[48]

Cutclaw: Agentic hours-long video editing via music synchronization.arXiv preprint arXiv:2603.29664, 2026

Shifang Zhao et al. Cutclaw: Agentic hours-long video editing via music synchronization.arXiv preprint arXiv:2603.29664, 2026

work page arXiv 2026
[49]

Autogen: Enabling next-gen llm applications via multi-agent conversations

Qingyun Wu, Gagan Bansal, Jieyu Zhang, Yiran Wu, Shaoshuai Li, Erkang Zhu, Li Jiang, Xiaohan Zhang, Chi Liu, Ahmed Hassan Awadallah, Jianfeng Wang, et al. Autogen: Enabling next-gen llm applications via multi-agent conversations. InFirst Conference on Language Modeling, 2024

2024
[50]

arXiv preprint arXiv:2404.11584 , year=

Tula Masterman, Sandi Besen, Mason Sawtell, and Alex Chao. The landscape of emerging ai agent architectures for reasoning, planning, and tool calling: A survey.arXiv preprint arXiv:2404.11584, 2024

work page arXiv 2024
[51]

A survey of ai agent protocols.arXiv preprint arXiv:2504.16736,

Yingxuan Yang, Huacan Chai, Yuanyi Song, Siyuan Qi, Muning Wen, Ning Li, Junwei Liao, Haoyi Hu, Jianghao Lin, Gaowei Chang, et al. A survey of ai agent protocols.arXiv preprint arXiv:2504.16736, 2025. 13

work page arXiv 2025