Pith.

Robotics · arXiv:2610.20442 v1 · submitted 14 Oct 2026 · 300 references

Towards Agentic Robots: Model, Data, Environment, and Harness

Chonghan Qin, Dongming Wu, Kaituo Feng, Jiaxu Wang, Xinlei Yu, Peiwen Sun, Yijia Fan, Xilin He

show 8 more authors

Bingao Chen, Weitong Lian, Ryan Ji, Wenqi Ren, Xiaobin Hu, Dahua Lin, Yu-Gang Jiang, Xiangyu Yue

The paper says

“Foundation models are driving a paradigm shift in robotic intelligence, moving beyond low-level action prediction to goal-directed orchestration of data collection, model training, task planning, and real-world execution.”

more of the abstract

We characterize this emerging paradigm as an agentic robots, in which agents reason over human instructions, interact with robotic environments, rectify execution errors autonomously, and dynamically adjust operational strategies. Although this field develops fast, current studies are not well connected. Researchers have no consistent view of how to combine different agentic functions into a complete robot system. This paper provides a unified overview of existing research on agentic robots. We first trace the development from robot policies to agentic systems, covering vision-language-action models, world models, and world-action models. We then clarify the definition and scope of agentic robots and organize existing research into seventeen categories across four functional domains: Model, Data, Environment, and Harness. Across these domains, we examine how agents shape model learning and robot decisions, collect and curate robot data, construct worlds, tasks, and tests for learning and evaluation, and organize capabilities for robot use. Finally, we discuss open challenges and outline directions for future research. Looking forward, we hope this work contributes to a clearer understanding of agentic robots and supports new approaches to making them more capable, reliable, and adaptive.

from the abstract

Pith says

Machine review · deepseek-v4-flash · 15 Oct 2026 · what was read was not recorded · not checked by a person

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

Holds up, with conditions. Useful, carefully hedged taxonomy for a fragmented field; the main weakness is that the carving's reproducibility is asserted, not demonstrated. This line is wrong

3 major objections · 6 minor

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 300 references say.

The paper says, at Section 3.2; Abstract; Figure 3

The review did not quote this passage.

Pith says: Objection 1 of 3 · major

The paper claims to provide "a unified overview of existing research on agentic robots" (Abstract) and to organize existing work into seventeen categories (Section 3.2), but it never states how the curated corpus backing this claim was assembled. The repository announced in the abstract and visualized in Figure 3 contains roughly 470 works, yet no inclusion or exclusion criteria — source venues, time window, keyword or snowballing strategy, or definition-based screening — are given anywhere in the manuscript.

more of objection 1

Because the taxonomy's value depends on the corpus being representative of the field as a whole, and because the domain sizes are sharply uneven (Harness receives six categories, Data and Environment three each, Table 1), the reader cannot rule out that the corpus is skewed toward the authors' emphasis on runtime orchestration. I ask the authors to add a corpus-construction protocol to Section 3.2 and to report the distribution of the corpus across the seventeen categories.

The paper says, at Table 1; Sections 3.2, 5.1, 6.2

The review did not quote this passage.

Pith says: Objection 2 of 3 · major

The taxonomy's reproducibility is not evidenced. Section 3.2 states that each work is discussed "under the categories corresponding to its relevant functions," but what counts as the relevant function is settled by judgment, and the text shows the judgment doing real work: RATs and Teach and Grow are moved out of Data Collection because they "primarily retain programs, lessons, or …

more of objection 2

… behavior-library entries" (Section 5.1), and EmbodiedSmith's demonstrations are assigned to data generation because "generating a solution does not necessarily generate a new task" (Section 6.2). These are reasonable calls, and the text gives reasons for them, but Table 1 supplies no operational rule that makes them checkable, and no inter-annotator agreement or held-out coverage test is reported. Since the taxonomy is the paper's central contribution, the authors should either formalize the assignment criteria or report a reproducibility check, for example independent classification of a held-out sample of recent agentic-robotics papers into the seventeen categories.

The paper says, at Section 3.3

The review did not quote this passage.

Pith says: Objection 3 of 3 · major

The choice of the four-domain, seventeen-category carving is asserted rather than argued. Section 3.3 disposes of five alternative framings — PAPAV's Perceive-Anticipate-Plan-Act-Verify lifecycle, model–robot integration via protocol interfaces [287], robot-learning surveys [129], executable agent software [250], and verification-signal analysis [331] — in a single paragraph, after which the four domains are taken as given for the rest of the paper.

more of objection 3

A reader who believes the field is better carved by lifecycle stages, by integration layers, or by the distinction between learned and programmed behavior has no stated grounds to prefer the proposed taxonomy. The authors should expand Section 3.3 to articulate what the four functional domains explain that the alternatives do not, and where the seventeen categories deliberately cut across them.

6 minor objections

The paper says, at Abstract

Pith says: minor objection 1 of 6

"We characterize this emerging paradigm as an agentic robots" — the article "an" does not agree with the plural "robots"; the phrase should read "as agentic robots" or "as an agentic robot."

The paper says, at Section 2.4

Pith says: minor objection 2 of 6

"Following section examine four functional domains" should read "The following section examines" or "The following sections examine."

show 4 more

The paper says, at Tables 10 and 12

Pith says: minor objection 3 of 6

The entries "/commen◎-do◎s" and "/da◎abase" appear to be rendering artifacts of icon characters; the intended symbols should be typeset properly.

The paper says, at Table 17 caption; Section 7.5

Pith says: minor objection 4 of 6

The admission that the Infrastructure category "includes supporting tools and deployment systems, which need not be autonomous agents" is a scope decision that deserves discussion in the main text of Section 7.5, since the paper defines the field by agentic behavior (Section 3.1).

The paper says, at Sections 3.1 and 3.3

Pith says: minor objection 5 of 6

The Section 3.1 definition of agentic robots as systems in which agents "autonomously plan and coordinate task execution" sits in some tension with the later clarification that agency "does not imply complete independence from human guidance"; the definitional wording should accommodate the clarification more directly.

The paper says, at Reference [235]

Pith says: minor objection 6 of 6

Reference [235] appears to be used for two distinct works: BTA in Table 4 and MetaTool in Section 7.2 and Table 14. Please verify the citation numbering.

What would settle it

Draw a stratified random sample of about 50 papers from the curated repository, plus about 50 recent arXiv cs.RO papers matching 'agentic robot' (2025-2026) that the authors did not curate. Give two independent annotators only Section 3.2 and Table 1, blinded to the paper's own assignments, and ask them to mark every category that applies. Compute per-category Krippendorff's alpha and coverage, defined as the fraction of papers assigned at least one category. If alpha falls below 0.6 on the Model, Data, or Environment categories, or if a substantial share of held-out papers fits no category, the central organizing claim needs revision; reporting the resulting per-domain distribution would also test whether Harness's six categories reflect the field or the authors' weighting.

from the skeptic's reading

How the review went

4 readings by deepseek-v4-flash · 15 Oct 2026

  1. Reader holds up, with conditions confidence moderate
  2. Skeptic agrees 1 min 9 s · 48,540 tokens read, 6,928 written
  3. Editor wrote the verdict above 36 s · 49,519 tokens read, 3,725 written
  4. Referee major revision: 3 major objections, 6 minor 2 min 15 s · 48,814 tokens read, 17,788 written

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

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

The review in full

The review in brief, machine-written

Best part of the paper is the MDEH taxonomy. It gives a fragmented field a four-question structure that maps well onto how the literature is actually written, and the seventeen categories are specific enough to help without being too fine-grained. The System 2/1/0 discussion in Section 7.6 is the sharpest piece: it connects current VLA hierarchies to older control stacks and cleanly separates planner, learned policy, and analytic controller. The authors also hedge honestly. They state that simulation procedures establish feasibility, not real-world validity, and that adaptive evaluation depends on test selection. There is no derivation to audit, so soundness concerns come down to coverage and assignment, not math.

The soft spots are real but not fatal. The carving is asserted, not demonstrated. Section 3.2 says each paper is discussed under the categories corresponding to its relevant functions, but no operational rule says how to assign a multi-function paper. The text shows judgment calls: RATs and Teach and Grow are moved to Section 7.6, and EmbodiedSmith's demonstrations are assigned to data generation rather than task generation. Each call is reasonable, but none is checkable. The repository has no stated inclusion criteria, and the citations are dominated by 2025-2026 preprints, so the corpus cannot be audited. I do not see evidence of a bad skew; the Environment and Infrastructure sections cover simulation, benchmarks, serving, and safety. The problem is that a skeptical reader cannot rule out skew, and the paper's central claim is that this is the right map of the field. Taxonomies can be proposed without inter-annotator agreement, but a selection protocol and a category-assignment rule would make the claim testable. The imbalance of six Harness categories versus three each for Data and Environment is defensible if the work clusters there; the paper just does not show it.

This paper is for people entering agentic robotics and for anyone who needs a shared vocabulary to compare systems. I would cite it, and I would bring it to a reading group. It deserves a serious referee. My recommendation: engage with it, and ask for the assignment protocol and corpus selection criteria in revision. The taxonomy will be useful either way, but those additions would take it from a useful map to a verifiable one.

The strongest objection

The central claim is that Model/Data/Environment/Harness with seventeen categories is the right map of agentic robotics. For that claim to hold, the map must be a stable, reproducible carving: independent readers using Section 3.2 and Table 1 should place the field's work in the same categories, and the cited corpus should be representative. The paper supplies no evidence for either. Section 3.2 says works are discussed 'under the categories corresponding to their relevant functions,' but what counts as relevant is left to judgment, and the text shows the judgment doing real work: RATs and Teach and Grow are moved out of Data Collection into Section 7.6 because they 'primarily retain programs, lessons, or behavior-library entries' (Section 5.1), and EmbodiedSmith's demonstrations are assigned to data generation rather than Task Generation because 'generating a solution does not necessarily generate a new task' (Section 6.2). These are reasonable calls, but Table 1 gives no operational rule that makes them checkable. The curated repository has no stated inclusion or exclusion criteria, so the roughly 470-work corpus may be skewed toward the authors' emphasis; Harness receives six categories, including Infrastructure, whose own table caption admits some works 'need not be autonomous agents' (Table 17), while Data and Environment receive three each. If another annotator would distribute the corpus differently, or if a held-out set of recent agentic-robotics papers fits no category, the taxonomy's organizing value is not established. This is a falsifiable method gap, not an accusation of misclassification.

The paper's central contribution is an organizing map, so the load-bearing condition is that the map is stable and representative. The reader's weakest-assumption analysis identifies exactly this condition: the definition in Section 3.1 and the informal corpus selection in Section 3.2 fix the taxonomy's shape without an empirical check, and there is no inter-annotator agreement test. My reading of the full text supports that concern and adds a concrete way to see it at work: the paper's own boundary decisions, such as moving RATs and Teach and Grow from Data Collection to Section 7.6, or classifying EmbodiedSmith's demonstrations as data generation rather than task generation, are reasonable but unconstrained by any stated rubric. I also note a supporting tension the reader did not emphasize: Harness.Infrastructure is defined to include systems that, by the paper's own Table 17 caption, 'need not be autonomous agents,' which sits uneasily with the stated principle that the taxonomy traces model-mediated decisions. The authors acknowledge the breadth of 'harness' explicitly, so I treat this as further evidence of the need for a category-assignment rule rather than as a fatal internal contradiction. I would not move the verdict. The paper is a survey and position piece, not a methods paper; it is carefully hedged, openly lists limitations such as simulation-only evidence and protocol-dependent benchmark scores, and all quantitative claims are quoted from cited sources. A taxonomy proposal can legitimately be accepted conditionally without a formal validation study. But because the central claim is empirical in character, the lack of a selection protocol and of a stability check justifies keeping the reader's CONDITIONAL verdict rather than upgrading to full acceptance. The proposed inter-annotator and coverage test is the single check that would settle whether the four-domain, seventeen-category carving actually organizes the field or mostly organizes the authors' reading of it.

The referee's report

If the taxonomy holds, this paper gives a fast-moving and fragmented field a shared vocabulary for locating where agent decisions act and how those decisions connect across a robotic system. The paper's strengths include the operational definitions in Table 1; the comparison tables (Tables 2–18), which report executor, feedback, and evaluation-protocol structure rather than only task success; a curated GitHub repository of roughly 470 works; and unusually careful hedging, including explicit statements that simulation procedures establish feasibility rather than real-world validity (Section 6.1), that adaptive-evaluation conclusions depend on test selection and score validation (Section 6.3), and that retained context does not imply persistent capability acquisition (Section 4.5). The central weakness is evidentiary: the taxonomy's reliability as a map of the field is not established, because the corpus selection protocol is unstated and category assignment is not validated for stability. The paper's value as a reference map is real, but its central organizing claim currently rests on the authors' judgment rather than on a checkable procedure.

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

The paper's central claim is that the emerging field of agentic robotics is best understood through a single taxonomy: four functional domains — Model, Data, Environment, and Harness — that span seventeen overlapping categories of research. It defines an agentic robot as a system in which an agent autonomously plans and coordinates task execution using environmental observations and execution feedback, going beyond action prediction. The paper argues this map matters because current research is fragmented: researchers study model learning, data pipelines, world construction, and runtime software separately, with no consistent view of how agentic functions combine into a complete robot system. It traces the evolution from task-specific policies through vision-language-action models and world models to agentic systems, then uses the taxonomy to review representative methods and to frame a research agenda on fair benchmarking, real-robot deployment, efficient robot use, and cross-embodiment generalization. If the taxonomy is right, researchers gain a shared way to locate where agent decisions act, how those decisions connect across a system, and which open problems block reliable deployment.

What carries the argument. The load-bearing machinery is the taxonomy itself: Table 1's four functional domains and seventeen categories, each defined by the part of the robotic system shaped by model-mediated decisions. Each domain is organized by a headline question — Model asks what the agent should learn, understand, decide, and control; Data asks how it should collect, annotate, and refine robot data; Environment asks what worlds, tasks, and tests it should build; and Harness asks how capabilities should be orchestrated, recovered, and evolved. The taxonomy does the argument's work by turning a sprawling, fragmented literature into a coordinate system: any agentic function can be located by domain and category, and cross-domain dependencies (data refinement feeding adaptation, monitoring feeding skill evolution) become visible as connections between categories. The secondary machinery is the evolution narrative in Section 2 — task-specific policies, then vision-language-action models, then world models and world-action models, then agentic systems — which supplies the contrast showing what agentic decisions add beyond action prediction.

Core claim. The paper's discovery is organizational: it claims that the work of agentic robots can be divided into seventeen non-exclusive functional categories across four domains, and that this division identifies where model-mediated decisions act in a robotic system. The Model domain covers what the agent learns, perceives, plans, controls, and adapts; the Data domain covers how the agent collects, annotates, and refines robot experience; the Environment domain covers how the agent reconstructs scenes, generates tasks, and builds tests; and the Harness domain covers how capabilities are orchestrated, remembered, monitored, recovered, packaged as skills, served, and layered. A direct corollary stated throughout the review is that a benchmark score depends on the full execution contract — agent decision, executor, feedback path, observation access, and interaction budget — so task success alone cannot reveal what the agent contributed. The taxonomy is functional rather than exclusive: a single paper can appear in multiple categories when it makes contributions to more than one.

The premise it rests on. The taxonomy's shape rests on the paper's own definition of agentic robots (Section 3.1), which sets what counts as in scope, and on an informal, criteria-free selection of literature behind the curated repository, so if real research does not cluster along Model, Data, Environment, and Harness the organizing claim weakens.

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

Researchers gain a shared vocabulary for locating where agent decisions act in a robot system and how those decisions connect across model learning, data, environments, and runtime software.

Evaluation practice would change: a benchmark score becomes meaningful only when reported with its full execution contract — executor, observation access, feedback path, and interaction budget — since the same task measures different abilities under different contracts.

Agent contribution becomes experimentally isolable: comparing a system with and without a given component (memory, monitoring, replanning, skill evolution) under fixed executors reveals what the agent actually adds.

Progress stops being a purely model-scaling story: efficient robot use is shown to depend on when models are called and where computation runs, not only on how fast inference is.

The four outlook directions — fair benchmarking, real-robot agentic robotics, efficient robot use, and cross-embodiment generalization — mark the open problems that currently block reliable deployment.

A direct test of the taxonomy's organizing claim would be an inter-annotator study: if independent researchers cannot reliably assign papers to the seventeen categories, the claim that the field clusters this way is weakened (the paper itself provides no such agreement test).

The "agent contribution" framing suggests a reporting standard the paper leaves implicit: publish agent score, executor baseline, and resource budget as a tuple, so the marginal value of agentic decisions is measurable across papers.

The paper's layered-systems analysis (slow planner, reactive policy, analytic controller) supports a design principle it states only implicitly: authority should track operating rate, with slow layers setting goals and fast layers executing within checked bounds.

An immediate extension would be a benchmark that varies exactly one domain while holding the other three fixed — for instance, identical Model and Data with different Harness components — to quantify where agentic gains actually originate.

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

Foundation models are driving a paradigm shift in robotic intelligence, moving beyond low-level action prediction to goal-directed orchestration of data collection, model training, task planning, and real-world execution. We characterize this emerging paradigm as an agentic robots, in which agents reason over human instructions, interact with robotic environments, rectify execution errors autonomously, and dynamically adjust operational strategies. Although this field develops fast, current studies are not well connected. Researchers have no consistent view of how to combine different agentic functions into a complete robot system. This paper provides a unified overview of existing research on agentic robots. We first trace the development from robot policies to agentic systems, covering vision-language-action models, world models, and world-action models. We then clarify the definition and scope of agentic robots and organize existing research into seventeen categories across four functional domains: Model, Data, Environment, and Harness. Across these domains, we examine how agents shape model learning and robot decisions, collect and curate robot data, construct worlds, tasks, and tests for learning and evaluation, and organize capabilities for robot use. Finally, we discuss open challenges and outline directions for future research. Looking forward, we hope this work contributes to a clearer understanding of agentic robots and supports new approaches to making them more capable, reliable, and adaptive.

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 · 2026-10-15

The authors' main claim, as the classifier picked it out 0.52

“Looking forward, we hope this work contributes to a clearer understanding of agentic robots and supports new approaches to making them more capable, reliable, and adaptive.”

How the main claim is stated: hedged possibility 0.80

What the abstract itself states, as probabilities: claims first 0.07 · states limitation 0.11 · human participants 0.08 · controlled comparison 0.02 · randomized assignment 0.01 · states quantitative result 0.72 · compares with prior results 0.05 · mentions code or data availability 0.03

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
Foundation models are driving a paradigm shift in robotic intelligence, moving beyond low-level action prediction to goal-directed orchestration of data collection, model training, task planning, and real-world execution. background 1.00 0.06 0.06 asserted 1.00
We characterize this emerging paradigm as an agentic robots, in which agents reason over human instructions, interact with robotic environments, rectify execution errors autonomously, and dynamically adjust operational strategies. background 0.66 0.13 0.09 asserted 1.00
Although this field develops fast, current studies are not well connected. gap 0.90 0.14 0.03 asserted 1.00
Researchers have no consistent view of how to combine different agentic functions into a complete robot system. gap 0.98 0.14 0.03 asserted 1.00
This paper provides a unified overview of existing research on agentic robots. contribution 1.00 0.85 0.10 asserted 0.99
We first trace the development from robot policies to agentic systems, covering vision-language-action models, world models, and world-action models. contribution 0.80 0.67 0.04 asserted 0.95
We then clarify the definition and scope of agentic robots and organize existing research into seventeen categories across four functional domains: Model, Data, Environment, and Harness. contribution 0.89 0.92 0.13 asserted 1.00
Across these domains, we examine how agents shape model learning and robot decisions, collect and curate robot data, construct worlds, tasks, and tests for learning and evaluation, and organize capabilities for robot use. contribution 0.84 0.61 0.08 asserted 0.93
Finally, we discuss open challenges and outline directions for future research. future work 0.86 0.65 0.08 asserted 0.70
Looking forward, we hope this work contributes to a clearer understanding of agentic robots and supports new approaches to making them more capable, reliable, and adaptive. future work 0.93 0.08 0.17 hedged possibility 0.80

What kind of paper the classifier reports this is: review 1.00 · evidence from literature 1.00

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, 2026-10-15: 1 request, 7,468 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 (7)
Figure 1 of the paper
Figure 1. Robot intelligence across four paradigms: task-specific policies for direct control, vision-language-action models grounded in language model knowledge, imagination-centric world models or world-action models, and agentic decision-making that reasons beyond learned policies. †Core Supervisors 1
Figure 2 of the paper
Figure 2. Overview of agentic robots across Model, Data, Environment, and Harness. The four domains connect model decisions, robot execution, environmental interaction, and learning from experience. 1 Introduction Over the past few years, foundation models including large language models (LLMs) and multimodal large language models (MLLMs) have greatly advanced robotic policy learning. They endow robots with novel capabilities that traditional task-specific models lack, especially in scene and task understanding and cross-scene generalization, by scaling robot datasets and model size [20,…
Figure 3 of the paper
Figure 3. Taxonomy and literature landscape of agentic robotic systems. Four functional domains cover seventeen non-exclusive categories: Model covers learning and robot-facing decisions. Data covers the acquisition and curation of data. Environment covers worlds, tasks, and evaluation. Harness covers how robotic capabilities are coordinated, reused, and improved to support agentic task execution. 3 Foundations of Agentic Robots The approaches discussed in Section 2 provide capabilities for perception, reasoning, action prediction, and world modeling. Building agentic robots requires…
Figure 4 of the paper
Figure 4. Five model-side interventions illustrated by a drawer-retrieval task. Training shapes the learning process; Perception interprets task-relevant evidence; Planning organizes task structure; Robot Control specifies robot-facing commands; and Adaptation changes model parameters or task-specific conditioning using experience. 4 Model: What Should the Agent Learn, Understand, Decide, and Control?
Figure 5 of the paper
Figure 5. Three data-side interventions over shared robot experience, illustrated by a drawer interaction with failure and recovery. Collection acquires or generates experience; Annotation adds task meaning, temporal structure, and supervision; and Refinement selects useful records or segments and revises unreliable supervision. 5.1 Collection Agent-based data collection aims to generate new robot trajectories or demonstrations through autonomous agents. We group these methods into three categories. Physical Data Collection gathers data from real-robot interaction and human demonstration.…
Figure 6 of the paper
Figure 6. Three ways to build and test an environment for one robot task. Reconstruction recovers an executable scene from observations. Task Generation defines goals, constraints, and success conditions. Benchmarking fixes test conditions, reset protocols, metrics, and evaluators for success, safety, efficiency, and robustness. distinguish progress and recovery from idle or failure-inducing behavior, while ARS uses agents to call video-inspection tools and verify task events before progress estimates guide segment selection. Zero2Skill [334] and Reflective Test-Time Planning [105] revisit…
Figure 7 of the paper
Figure 7. Harness functions that organize agentic capabilities around robot execution. The central row connects task goals and evidence to decision and dispatch, exposed executors, robot execution, and returned feedback. Orchestration governs calls and handoffs; Memory supplies retained information; Monitoring and Recovery assess execution and respond to failure; Skills package reusable behavior; Infrastructure exposes and gates callable capabilities; and Layered Systems divide authority and operating rates across planners, learned policies, and analytic control. 7.1 Orchestration…

Figures are the authors', reproduced from the paper.

References (300)
  1. [287] SaharSalimpour,LeiFu,KajetanRachwał,PascalBertrand,KevinO’Sullivan,RobertJakob,FarhadKeramat, Leonardo Militano, Giovanni Toffetti, Harry Edelman, and Jorge Peña Queralta. Towards Embodied Agentic AI: Review and Classification of LLM- and VLM-Driven Robot Autonomy and Interact... on PitharXiv
  2. [129] Gaytri Jena, Kapil Wanaskar, Vinija Jain, Aman Chadha, Vasu Sharma, and Amitava Das. Weights or Skills? A Survey of Robot-Learning Techniques: from Action-Predicting Weights to Robots that Write their Own Skills.arXiv preprint arXiv:2608.01851, 2026 on PithPith's reviewarXiv
  3. [250] Xuying Ning, Katherine Tieu, Dongqi Fu, Tianxin Wei, Zihao Li, Yuanchen Bei, Jiaru Zou, Mengting Ai, Zhining Liu, Ting-Wei Li, Lingjie Chen, Yanjun Zhao, Ke Yang, Bingxuan Li, Cheng Qian, Gaotang Li, Xiao Lin, Zhichen Zeng, Ruizhong Qiu, Sirui Chen, Yifan Sun, Xiyuan Yang, Rui... on PithPith's reviewarXiv
  4. [1] Dionysios Adamopoulos, Nattapol Chanpaisit, Basel Fakhri, and Christina Giannoula. Efficient Vision- Language-Action Management and Serving for Robot Factories.arXiv preprint arXiv:2609.12075, 2026 on PithPith's reviewarXiv
  5. [2] AgiBot Research Team, Renhang Liu, Wenzhi Zhao, Zhuo Yang, Liliang Chen, Pengfei Zhou, Shengcong Chen, Guanghui Ren, Youlun Peng, Rongjun Jin, Nan Wang, Sukai Wang, Xindong He, Jinyuan Feng, Ziyu Xiong, Linqing Zhong, Yifei Wei, Feng Han, Long Zhang, Da Huang, Nanshu Zhao, Che... on PithPith's reviewarXiv
  6. [3] Michael Ahn, Debidatta Dwibedi, Chelsea Finn, et al. AutoRT: Embodied Foundation Models for Large Scale Orchestration of Robotic Agents.arXiv preprint arXiv:2401.12963, 2024 on PitharXiv
  7. [4] Alibaba Cloud. Qwen-Max, 2026. URL https://www.alibabacloud.com/help/en/model-studio/qwen-max on Pith
  8. [5] Mido Assran, Adrien Bardes, David Fan, Quentin Garrido, Russell Howes, Mojtaba Komeili, Matthew Muckley, Ammar Rizvi, Claire Roberts, Koustuv Sinha, Artem Zholus, Sergio Arnaud, Abha Gejji, Ada Martin, Francois Robert Hogan, Daniel Dugas, Piotr Bojanowski, Vasil Khalidov, Patr... on PithPith's reviewarXiv
  9. [6] Awomo-PhysicalRSI Team, Danjiao Ma, Enhui Ma, Haohan Liu, Heng Jia, Hui Shan, Jianhua Xu, Jiahuan Zhang, Jiangdi Xu, Kaiwen Guo, Kaicheng Yu, Linwei Zhang, Liyang Jin, Maochun Luo, Pengyao Niu, Shiwen Li, Shuangyu Feng, Tong Zhang, Tianheng Wang, Xin Wang, Xiangru Huang, Yongq... on PithPith's reviewarXiv
  10. [7] Rohan Bansal, David He, Nadun Ranawaka Arachchige, Zhenyang Chen, Soobum Kim, Kexin Rong, and Danfei Xu. Action Chunk Scheduling for Batched Robot Policy Serving.arXiv preprint arXiv:2608.00337, 2026 on PithPith's reviewarXiv
  11. [8] Anuva Banwasi, William Muckelroy III, Priya Sundaresan, Linfeng Zhao, Jeannette Bohg, and Cherie Ho. MessyMem: Learning-from-Doing Memory for Mobile Manipulation.arXiv preprint arXiv:2609.15976, 2026 on PithPith's reviewarXiv
  12. [9] Nicolas Baumann, Liam Boyle, Pu Deng, Edoardo Ghignone, Boyang Sun, Marc Pollefeys, Luca Benini, and Michele Magno. LocalNav: Distilling Frontier VLMs and Embodied RL for On-Device Object Goal Navigation.arXiv preprint arXiv:2606.27871, 2026 on PithPith's reviewarXiv
  13. [10] Suneel Belkhale, Tianli Ding, Ted Xiao, Pierre Sermanet, Quon Vuong, Jonathan Tompson, Yevgen Chebotar, Debidatta Dwibedi, and Dorsa Sadigh. RT-H: Action Hierarchies Using Language.arXiv preprint arXiv:2403.01823, 2024 on PithPith's reviewarXiv
  14. [11] Vineet Bhat, Mikaela Angelina Uy, Siyi Chen, Alex Zook, Xuning Yang, Stan Birchfield, Valts Blukis, and Jonathan Tremblay. MG-VQA: Manipulation Grounded Visual Question Answering with VLMs.arXiv preprint arXiv:2608.17129, 2026 on PithPith's reviewarXiv
  15. [12] Hongzhe Bi, Hengkai Tan, Shenghao Xie, Zeyuan Wang, Shuhe Huang, Haitian Liu, Ruowen Zhao, Yao Feng, Chendong Xiang, Yinze Rong, Hongyan Zhao, Hanyu Liu, Zhizhong Su, Lei Ma, Hang Su, and Jun Zhu. Motus: A Unified Latent Action World Model.arXiv preprint arXiv:2512.13030, 2025 on PithPith's reviewarXiv
  16. [13] Kevin Black, Noah Brown, Danny Driess, Adnan Esmail, Michael 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 Shi, Jam... on PithPith's reviewarXiv
  17. [14] KevinBlack,ManuelY.Galliker,andSergeyLevine. Real-TimeExecutionofActionChunkingFlowPolicies. arXiv preprint arXiv:2506.07339, 2025. 49 on PithPith's reviewarXiv
  18. [15] KevinBlack,AllenZ.Ren,MichaelEqui,andSergeyLevine. Training-TimeActionConditioningforEfficient Real-Time Chunking.arXiv preprint arXiv:2512.05964, 2025 on PitharXiv
  19. [16] Nils Blank, Moritz Reuss, Marcel Rühle, Ömer Erdinç Yağmurlu, Fabian Wenzel, Oier Mees, and Rudolf Lioutikov. Scaling Robot Policy Learning via Zero-Shot Labeling with Foundation Models. InProceedings of The 8th Conference on Robot Learning, volume 270, pages 4158–4187. PMLR, 2025 on Pith
  20. [17] Nils Blank, Paul Mattes, Maximilian Xiling Li, Jakub Suliga, Thomas Roth, Moritz Reuss, Pankhuri Vanjani, and Rudolf Lioutikov. SPARC: Reliable Spatial Annotations from Robot Demonstrations at Scale.arXiv preprint arXiv:2606.13497, 2026 on PithPith's reviewarXiv
show 80 more
  1. [18] Konstantinos Bousmalis, Giulia Vezzani, Dushyant Rao, Coline Devin, Alex X. Lee, Maria Bauza, Todor Davchev, Yuxiang Zhou, Agrim Gupta, Akhil Raju, Antoine Laurens, Claudio Fantacci, Valentin Dalibard, Martina Zambelli, Murilo Martins, Rugile Pevceviciute, Michiel Blokzijl, Mi... on Pith
  2. [19] Anthony Brohan, Noah Brown, Justice Carbajal, Yevgen Chebotar, Xi Chen, Krzysztof Choromanski, Tianli Ding, Danny Driess, Avinava Dubey, Chelsea Finn, Pete Florence, Chuyuan Fu, Montse Gonzalez Arenas, Keerthana Gopalakrishnan, Kehang Han, Karol Hausman, Alexander Herzog, Jasm... on PithPith's reviewarXiv
  3. [20] Anthony Brohan, Noah Brown, Justice Carbajal, Yevgen Chebotar, Joseph Dabis, Chelsea Finn, Keerthana Gopalakrishnan, Karol Hausman, Alex Herzog, Jasmine Hsu, Julian Ibarz, Brian Ichter, Alex Irpan, Tomas Jackson, Sally Jesmonth, Nikhil J Joshi, Ryan Julian, Dmitry Kalashnikov,... on PithPith's reviewarXiv
  4. [21] Qingwen Bu, Hongyang Li, Li Chen, Jisong Cai, Jia Zeng, Heming Cui, Maoqing Yao, and Yu Qiao. Towards Synergistic, Generalized, and Efficient Dual-System for Robotic Manipulation.arXiv preprint arXiv:2410.08001, 2024 on PithPith's reviewarXiv
  5. [22] Bytedance Seed. Seed1.8 Model Card: Towards Generalized Real-World Agency.arXiv preprint arXiv:2603.20633, 2026 on PithPith's reviewarXiv
  6. [23] Remi Cadene, Simon Aliberts, Francesco Capuano, Michel Aractingi, Adil Zouitine, Pepijn Kooijmans, Jade Choghari, Martino Russi, Caroline Pascal, Steven Palma, et al. LeRobot: An Open-Source Library for End-to-End Robot Learning.arXiv preprint arXiv:2602.22818, 2026 on PitharXiv
  7. [24] XiaoweiCai,YunuoCai,BingaoChen,JingxiaoChen,ZhiChen,SiyuanFeng,TengyuHou,JingshunHuang, Han Jiang, Runkun Ju, Dong Li, Mingxiang Li, Shaowei Li, Xinchen Li, Yifan Li, Yi Liu, Zhongyuan Liu, Jianlan Luo, Junwen Miao, Ruiqi Ni, Buqing Nie, Mingjie Pan, Xinlin Ren, Jianheng Song,... on PithPith's reviewarXiv
  8. [25] Jiahang Cao, Qiang Zhang, Jingkai Sun, Jiaxu Wang, Hao Cheng, Yulin Li, Jun Ma, Kun Wu, Zhiyuan Xu, Yecheng Shao, Wen Zhao, Gang Han, Yijie Guo, and Renjing Xu. Mamba Policy: Towards Efficient 3D DiffusionPolicywithHybridSelectiveStateModels.In2025IEEE/RSJInternationalConferen... on Pithdoi:10.1109/iros60139.2025.11247625
  9. [26] Irvin Steve Cardenas, Marcus Anthony Arnett, Natalie Catherine Yeo, Lucky Sah, and Jong-Hoon Kim. ROSClaw: An OpenClaw ROS 2 Framework for Agentic Robot Control and Interaction.arXiv preprint arXiv:2603.26997, 2026 on PitharXiv
  10. [27] Nicolas Carion, Laura Gustafson, Yuan-Ting Hu, Shoubhik Debnath, Ronghang Hu, Didac Suris, Chaitanya Ryali, Kalyan Vasudev Alwala, Haitham Khedr, Andrew Huang, et al. SAM 3: Segment Anything with Concepts.arXiv preprint arXiv:2511.16719, 2025 on PithPith's reviewarXiv
  11. [28] Jun Cen, Siteng Huang, Yuqian Yuan, Kehan Li, Hangjie Yuan, Chaohui Yu, Yuming Jiang, Jiayan Guo, Xin Li, Hao Luo, Fan Wang, Deli Zhao, and Hao Chen. RynnVLA-002: A Unified Vision-Language-Action and World Model.arXiv preprint arXiv:2511.17502, 2025 on PithPith's reviewarXiv
  12. [29] Jun Cen, Chaohui Yu, Hangjie Yuan, Yuming Jiang, Siteng Huang, Jiayan Guo, Xin Li, Yibing Song, Hao Luo, Fan Wang, Deli Zhao, and Hao Chen. WorldVLA: Towards Autoregressive Action World Model.arXiv preprint arXiv:2506.21539, 2025 on PithPith's reviewarXiv
  13. [30] Nga Teng Chan, Yi Zhang, Yechi Liu, Renwen Cui, Fanhu Zeng, Zeyuan Ding, Xiancong Ren, Zhang Zhang, Qifeng Chen, Jian Liu, Yong Dai, and Xiaozhu Ju. Robo-Cortex: A Self-Evolving Embodied Agent via Dual-Grain Cognitive Memory and Autonomous Knowledge Induction.arXiv preprint ar... on PithPith's reviewarXiv
  14. [31] Matthew Chang, Gunjan Chhablani, Alexander Clegg, Mikael Dallaire Cote, Ruta Desai, Michal Hlavac, Vladimir Karashchuk, Jacob Krantz, Roozbeh Mottaghi, Priyam Parashar, Siddharth Patki, Ishita Prasad, Xavier Puig, Akshara Rai, Ram Ramrakhya, Daniel Tran, Joanne Truong, John M.... on Pith
  15. [32] 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
  16. [33] BoyuanChen,ZhuoXu,SeanKirmani,BrianIchter,DorsaSadigh,LeonidasGuibas,andFeiXia.SpatialVLM: Endowing Vision-Language Models with Spatial Reasoning Capabilities. InProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 14455–14465, 2024 on Pith
  17. [34] Daphne Chen, Archit Ritesh Jain, Eric Goossen, Emma Romig, Michael Murray, Nick Walker, and Maya Cakmak. A Few Words Go a Long Way: Language Guided Robot Policy Synthesis.arXiv preprint arXiv:2607.23784, 2026 on PithPith's reviewarXiv
  18. [35] Guanxiong Chen, Yiduo Qu, Qianjun Xia, Pengyu Jing, Yixian Cheng, Bole Ma, Pengzhi Yang, Bingyang Zhou, Ziming Li, Shashwat Suri, Gongbo Sun, Chao Liu, Peter Yichen Chen, Ziqiu Zeng, and Fan Shi. DiagGen: Agentic Generation of Deformable Assets with Sim-based Diagnostics for R... on PithPith's reviewarXiv
  19. [36] Guanxiong Chen, Qianjun Xia, Jiawei Peng, Heng Zhang, Pengyu Jing, Bole Ma, Justin Qian, Yixian Cheng, Ziyi Jiao, Bingyang Zhou, Yiduo Qu, Luoxin Ye, Kaifeng Zhang, Kunyi Wang, Weijia Zeng, Yunuo Chen, Pengzhi Yang, Ziqiu Zeng, Siyuan Luo, Huamin Wang, Chao Liu, Alan Yuille, F... on PithPith's reviewarXiv
  20. [37] Junting Chen, Checheng Yu, Xunzhe Zhou, Tianqi Xu, Yao Mu, Mengkang Hu, Wenqi Shao, Yikai Wang, Guohao Li, and Lin Shao. EMOS: Embodiment-aware Heterogeneous Multi-robot Operating System with LLM Agents. InInternational Conference on Learning Representations, 2025 on Pith
  21. [38] Kaiyuan Chen, Shuangyu Xie, Letian Fu, Justin Yu, William Pacini, Sandeep Bajamahal, Hudson Kim, Jaimyn Drake, Daehwa Kim, Haoru Xue, Jonathan Francis, Christian Juette, Peter Schaldenbrand, Muham- metYunusSeker,RuwanWickramarachchi,UksangYoo,GuanzhiWang,AdithyavairavanMurali,... on PithPith's reviewarXiv
  22. [39] SitongChen,FatemehZargarbashi,JinCheng,TianxuAn,andStelianCoros. KINO:AKeyframeInterfacefor VLMPlanningandWhole-BodyControlinHumanoidLoco-Manipulation.arXivpreprintarXiv:2609.18869, 2026. 51 on PithPith's reviewarXiv
  23. [40] Siyi Chen, Hugo Hadfield, Alex Zook, Mikaela Angelina Uy, Chan Hee Song, Erwin Coumans, Xuning Yang, Faisal Ladhak, Qing Qu, Stan Birchfield, Jonathan Tremblay, and Valts Blukis. VoLo: A Physical Orchestrator for Open-Vocabulary Long-Horizon Manipulation.arXiv preprint arXiv:2... on PithPith's reviewarXiv
  24. [41] Tianxing Chen, Yue Chen, Zixuan Li, Junyuan Tang, Kailun Su, Haoran Lu, Weijie Wan, Baijun Chen, Songling Liu, Haowen Yan, Honghao Su, Zhiyang Dou, Kaixuan Wang, Dandan Zhang, Yunze Liu, Yan Qin, Qiwei Liang, Qiwei Wu, Zijian Lin, Wenwei Lin, Yuran Wang, Minghua He, Tianshu Wu... on PithPith's reviewarXiv
  25. [42] William Chen, Jagdeep Singh Bhatia, Catherine Glossop, Nikhil Mathihalli, Ria Doshi, Andy Tang, Danny Driess,KarlPertsch,andSergeyLevine. SteerableVision-Language-ActionPoliciesforEmbodiedReasoning and Hierarchical Control.arXiv preprint arXiv:2602.13193, 2026 on PithPith's reviewarXiv
  26. [43] Xinyi Chen, Yilun Chen, Yanwei Fu, Ning Gao, Jiaya Jia, Weiyang Jin, Hao Li, Yao Mu, Jiangmiao Pang, Yu Qiao, Yang Tian, Bin Wang, Bolun Wang, Fangjing Wang, Hanqing Wang, Tai Wang, Ziqin Wang, XueyuanWei,ChaoWu,ShuaiYang,JinhuiYe,JunqiuYu,JiaZeng,JingjingZhang,JinyuZhang,ShiZ... on PithPith's reviewarXiv
  27. [44] Yang Chen, Lirong Che, Zhenyu Huang, Wenbo Fu, Chuang Wang, Xu Cao, Daqi Liu, Yuzhe Yang, Jian Su, and Lan-Zhe Guo. HarnessVLN: Unifying Training-Free Embodied Navigation through an Agent Harness. arXiv preprint arXiv:2609.15195, 2026 on PithPith's reviewarXiv
  28. [45] Yangtao Chen, Zixuan Chen, Peiyang Wang, Yong-Lu Li, Jing Huo, Jieqi Shi, and Yang Gao. Wh0: Generative World Models as Scalable Sources of Egocentric Human Hand Manipulation Data.arXiv preprint arXiv:2606.22136, 2026 on PithPith's reviewarXiv
  29. [46] Yanzhe Chen, Zechen Bai, Zhijun Cao, Wenzheng Zeng, Kevin Qinghong Lin, Yiqi Lin, Guoqiang Liang, Kevin Yuchen Ma, Qiming Huang, and Mike Zheng Shou. Show-Harness: Just a VLM Agent Can Play Robots.arXiv preprint arXiv:2609.10522, 2026 on PithPith's reviewarXiv
  30. [47] Yanzhe Chen, Ziyi Yang, Jifeng Zhu, Qiming Huang, Ruihe An, Peiyao Xu, Hesen Yang, Runda Liu, Chang Gong, Zhijun Cao, Zechen Bai, Wenzheng Zeng, Yiqi Lin, Guoqiang Liang, Kevin Yuchen Ma, Kevin Qinghong Lin, and Mike Zheng Shou. Survey on Multimodal Embodied Agents: A Unified ... on Pith
  31. [48] YitengChen,ZheCao,HongjiaRen,ChenjieYang,WenboLi,ShiyiWang,YeminWang,LiZhang,Yanming Shao, Zhenjun Zhao, Huiping Zhuang, and Qingyao Wu. RoboRouter: Training-Free Policy Routing for Robotic Manipulation.arXiv preprint arXiv:2603.07892, 2026 on PithPith's reviewarXiv
  32. [49] Yiteng Chen, Huiping Zhuang, Wenbo Li, Shiyi Wang, Xiangyu Zhao, and Qingyao Wu. ManipEvalAgent: Promptable and Efficient Evaluation Framework for Robotic Manipulation Policies. InInternational Conference on Learning Representations, volume 2026, pages 84574–84598, 2026 on Pith
  33. [50] Yitong Chen, Zezheng Huai, Sixian Li, Yubang Wang, Haozhe Zhang, Yifei Zhang, Hechang Chen, Jingjing Gong, Yu-Gang Jiang, and Xipeng Qiu. ETA: A New Agentic Paradigm for Embodied Tasks.arXiv preprint arXiv:2608.03924, 2026 on PithPith's reviewarXiv
  34. [51] Yuzhen Chen and KC Zhou. WCM: World-Cognition Model for Generalizable Human-Robot Interaction. arXiv preprint arXiv:2607.22999, 2026 on PithPith's reviewarXiv
  35. [52] An-ChiehCheng,YandongJi,ZhaojingYang,ZaitianGongye,XueyanZou,JanKautz,ErdemBıyık,Hongxu Yin, Sifei Liu, and Xiaolong Wang. NaVILA: Legged Robot Vision-Language-Action Model for Navigation. arXiv preprint arXiv:2412.04453, 2024 on PithPith's reviewarXiv
  36. [53] Dongzhou Cheng, Taoran Yi, Ye Fang, Xingwu Zhang, Fan Feng, Yixuan Li, Gengxiong Zhuang, Rongze Wang, Shuai Yang, WeiSong, WeizhiXue, Minyan Wu, JieGui, JiaqiWang, andTongWu. In-ContextRobot Learning with VLM Agents.arXiv preprint arXiv:2609.19138, 2026. 52 on PithPith's reviewarXiv
  37. [54] Ho Kei Cheng, Seoung Wug Oh, Brian Price, Alexander Schwing, and Joon-Young Lee. Tracking Anything withDecoupledVideoSegmentation. InProceedingsoftheIEEE/CVFInternationalConferenceonComputer Vision (ICCV), pages 1316–1326, 2023 on Pith
  38. [55] 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.arXiv preprint arXiv:2303.04137, 2024 on PithPith's reviewarXiv
  39. [56] Gheorghe Comanici, Eric Bieber, Mike Schaekermann, Ice Pasupat, Noveen Sachdeva, et al. Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities.arXiv preprint arXiv:2507.06261, 2025 on PithPith's reviewarXiv
  40. [57] Nitish Dashora, Douglas Chen, Idan Shenfeld, John Marangola, Pulkit Agrawal, and Max Simchowitz. Workspace Models: Lightweight Robotic Memory via Saliency-Driven Supervision.arXiv preprint arXiv:2609.20820, 2026 on PithPith's reviewarXiv
  41. [58] Matt Deitke, Christopher Clark, Sangho Lee, Rohun Tripathi, Yue Yang, Jae Sung Park, Mohammadreza Salehi, Niklas Muennighoff, Kyle Lo, Luca Soldaini, et al. Molmo and PixMo: Open Weights and Open Data for State-of-the-Art Vision-Language Models.arXiv preprint arXiv:2409.17146, 2024 on PithPith's reviewarXiv
  42. [59] Dexmal. OpenDM: DM0.5 for Open-World Robot Control, 2026. URL https://github.com/dexmal/opendm on Pith
  43. [60] Xin Ding, Liang Mi, Mingzhe Huang, Zixuan Wang, Chao Zhang, Zixu Hao, Fu Chen, Xiangyu Li, Yikai Zheng, Yaoyu Guo, Weijun Wang, Kun Li, Hao Wu, Yunxin Liu, and Ting Cao. Zetta𝜁: An Efficient Closed-Loop Embodied Harness for Self-Evolving Physical Intelligence.arXiv preprint ar... on PithPith's reviewarXiv
  44. [61] 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
  45. [62] Yilun Du, Mengjiao Yang, Bo Dai, Hanjun Dai, Ofir Nachum, Joshua B. Tenenbaum, Dale Schuurmans, and Pieter Abbeel. Learning Universal Policies via Text-Guided Video Generation.arXiv preprint arXiv:2302.00111, 2023 on PithPith's reviewarXiv
  46. [63] Jiafei Duan, Samson Yu, Hui Li Tan, Hongyuan Zhu, and Cheston Tan. A Survey of Embodied AI: From Simulators to Research Tasks.IEEE Transactions on Emerging Topics in Computational Intelligence, 6(2): 230–244, 2022. doi: 10.1109/TETCI.2022.3141105 on Pithdoi:10.1109/tetci.2022.3141105
  47. [64] Jiafei Duan, Wilbert Pumacay, Nishanth Kumar, Yi Ru Wang, Shulin Tian, Wentao Yuan, Ranjay Krishna, Dieter Fox, Ajay Mandlekar, and Yijie Guo. AHA: A Vision-Language-Model for Detecting and Reasoning Over Failures in Robotic Manipulation. InInternational Conference on Learning... on Pith
  48. [65] Karim Elmaaroufi, Justin Svegliato, Sarunas Kalade, Graham Schelle, Sanjit A. Seshia, and Matei Zaharia. RHO: Your Coding Agent is Secretly a Roboticist.arXiv preprint arXiv:2606.16458, 2026 on PitharXiv
  49. [66] YijiaFan, ZonglinDi, ZimoWen, YifanYang, MingxiCheng, QiDai, BeiLiu, KaiQiu, YueDong, JiLi, and Chong Luo. Resource2Skill: Distilling Executable Agent Skills from Human-Created Multimodal Resources. arXiv preprint arXiv:2606.29538, 2026 on PithPith's reviewarXiv
  50. [67] Hao-Shu Fang, Chenxi Wang, Hongjie Fang, Minghao Gou, Jirong Liu, Hengxu Yan, Wenhai Liu, Yichen Xie, and Cewu Lu. AnyGrasp: Robust and Efficient Grasp Perception in Spatial and Temporal Domains. arXiv preprint arXiv:2212.08333, 2022 on PithPith's reviewarXiv
  51. [68] YuanFang,ZechuLi,HaoleiTong,PuzeLiu,andGeorgiaChalvatzaki.FindSomethingYouCan’tDo: Agentic Real-World Reinforcement Learning for Self-Improving VLA Models.arXiv preprint arXiv:2609.32069, 2026 on PithPith's reviewarXiv
  52. [69] FBddcz. EmbodiedJev: A Browser-Based Workbench for Language-Model Decisions in Robot Manipulation, on Pith
  53. [70] Youhe Feng, Hansen Shi, Haoyang Li, Xinlei Guo, Yang Wang, Chengyang Zhang, Jinkai Zhang, Xiaohan Zhang, Jie Tang, and Jing Zhang. ProcVLM: Learning Procedure-Grounded Progress Rewards for Robotic Manipulation.arXiv preprint arXiv:2605.08774, 2026 on PithPith's reviewarXiv
  54. [71] YunhaiFeng,JiamingHan,ZhuoranYang,XiangyuYue,SergeyLevine,andJianlanLuo. ReflectivePlanning: Vision-Language Models for Multi-Stage Long-Horizon Robotic Manipulation. InProceedings of The 9th ConferenceonRobotLearning,volume305ofProceedingsofMachineLearningResearch,pages2038–2... on Pith
  55. [72] Figure AI. Helix: A Vision-Language-Action Model for Generalist Humanoid Control, 2025. URL https://www.figure.ai/news/helix on Pith
  56. [73] Chelsea Finn and Sergey Levine. Deep Visual Foresight for Planning Robot Motion.arXiv preprint arXiv:1610.00696, 2017 on PithPith's reviewarXiv
  57. [74] Letian Fu, Justin Yu, Karim El-Refai, Ethan Kou, Haoru Xue, Huang Huang, Wenli Xiao, Guanzhi Wang, Dantong Niu, Fei-Fei Li, Guanya Shi, Jiajun Wu, Shankar Sastry, Yuke Zhu, Ken Goldberg, and Linxi "Jim" Fan. CaP-X: A Framework for Benchmarking and Improving Coding Agents for R... on PithPith's reviewarXiv
  58. [75] Liane Galanti, Dhruv Shah, and Tri Dao. Addressing the Orchestration Gap in Generalist Robots via Physical Agency.arXiv preprint arXiv:2607.21725, 2026 on PithPith's reviewarXiv
  59. [76] Galaxea Team. Galaxea G0.5: One Autoregressive Stream for Robot Reasoning and Action, 2026 on Pith
  60. [77] Chongkai Gao, Zixuan Liu, Zhenghao Chi, Junshan Huang, Xin Fei, Yiwen Hou, Yuxuan Zhang, Yudi Lin, Zhirui Fang, Zeyu Jiang, et al. VLA-OS: Structuring and Dissecting Planning Representations and Paradigms in Vision-Language-Action Models.arXiv preprint arXiv:2506.17561, 2025 on PithPith's reviewarXiv
  61. [78] Shenyuan Gao, William Liang, Kaiyuan Zheng, Ayaan Malik, Seonghyeon Ye, Sihyun Yu, Wei-Cheng Tseng, Yuzhu Dong, Kaichun Mo, Chen-Hsuan Lin, Qianli Ma, Seungjun Nah, Loic Magne, Jiannan Xiang, Yuqi Xie, Ruijie Zheng, Dantong Niu, You Liang Tan, K. R. Zentner, George Kurian, Sun... on PithPith's reviewarXiv
  62. [79] Jingzhan Ge, Ruimin Chen, Azadeh Haghighi, Jiong Tang, and Farhad Imani. Kinematics-Grounded Agentic AI for Robotic Additive Manufacturing Process Planning.arXiv preprint arXiv:2609.19347, 2026 on PithPith's reviewarXiv
  63. [80] Yuying Ge, Annabella Macaluso, Li Erran Li, Ping Luo, and Xiaolong Wang. Policy Adaptation from Foundation Model Feedback.arXiv preprint arXiv:2212.07398, 2023 on PithPith's reviewarXiv
  64. [81] Gemini Robotics Team et al. Gemini Robotics: Bringing AI into the Physical World.arXiv preprint arXiv:2503.20020, 2025 on PithPith's reviewarXiv
  65. [82] Genesis Authors. Genesis: A Generative and Universal Physics Engine for Robotics and Beyond, December on Pith
  66. [83] Theophile Gervet, Zhou Xian, Nikolaos Gkanatsios, and Katerina Fragkiadaki. Act3D: 3D Feature Field TransformersforMulti-TaskRoboticManipulation. InProceedingsofThe7thConferenceonRobotLearning, volume 229 ofProceedings of Machine Learning Research, pages 3949–3965. PMLR, 2023 on Pith
  67. [84] Dorian Benhamou Goldfajn, Mason Nakamura, Saaduddin Mahmud, Justin Svegliato, Kyle H. Wray, and Shlomo Zilberstein. RoboTalk: Learning Multi-Robot Communication and Coordination from Multimodal Demonstrations.arXiv preprint arXiv:2609.23997, 2026 on PithPith's reviewarXiv
  68. [85] Litian Gong, Fatemeh Bahrani, Yutai Zhou, Amin Banayeeanzade, Jiachen Li, and Erdem Bıyık. AutoFocus- IL:VLM-basedSaliencyMapsforData-EfficientVisualImitationLearningwithoutExtraHumanAnnotations. arXiv preprint arXiv:2511.18617, 2025 on PitharXiv
  69. [86] Ruiyan Gong, Yingnan Guo, Junjun Hu, Jintao Kong, Xiaoxu Leng, Tianlun Li, Weize Li, Fei Liu, Zhicheng Liu, Jia Lu, Minghua Luo, Chenlin Ming, Yanfen Shen, Jiyue Tao, Zhengbo Wang, Mingyang Yin, Minqi Gu, Zihao Guan, Wei Guo, Guoqing Liu, Huachong Pang, Menglin Yang, Zeqian Ye... on PithPith's reviewarXiv
  70. [87] GoogleDeepMind. Gemini4Argon: ModelCard, 2026. URLhttps://deepmind.google/models/model-cards/ gemini-4-argon/. Accessed October 14, 2026 on Pith
  71. [88] Ankit Goyal, Jie Xu, Yijie Guo, Valts Blukis, Yu-Wei Chao, and Dieter Fox. RVT: Robotic View Transformer for 3D Object Manipulation.arXiv preprint arXiv:2306.14896, 2023 on PithPith's reviewarXiv
  72. [89] Zhaopeng Gu, Bingke Zhu, Tianxi Lin, Guibo Zhu, Yingying Chen, Kai Wang, Tingyu Yuan, 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
  73. [90] WeifanGuan,QinghaoHu,HuasenXi,ChenxiaoZhang,AoshengLi,andJianCheng.RoboNeuron: AMiddle- Layer Infrastructure for Agent-Driven Orchestration in Embodied AI.arXiv preprint arXiv:2512.10394, 2025 on PitharXiv
  74. [91] Xinyuan Guan, Feifan Chen, Xinyu Zhan, Fu-Cheng Zhang, Cewu Lu, and Lixin Yang. EgoAfford: Task- Oriented Affordance Grounding via Egocentric Referring Segmentation.arXiv preprint arXiv:2608.04533, 2026 on PithPith's reviewarXiv
  75. [92] Christian Gumbsch, Leonardo Barcellona, Lennard Schünemann, Platon Karageorgis, Andrii Zadaianchuk, ZehaoWang,SergeyZakharov,FabienDespinoy,RahafAljundi,andEfstratiosGavves. FromDemonstrations to Rewards: Test-Time Prompt Optimization for VLM Reward Models.arXiv preprint arXiv... on PithPith's reviewarXiv
  76. [93] Meng-Hao Guo, Zhe-Han Mo, Jia-Jun Wang, Yi Zhang, Kejin Wang, Yi-Xuan Deng, Jia-Peng Zhang, Yongming Rao, and Shi-Min Hu. Transferring the Intelligence of VLMs to Robotic Control.arXiv preprint arXiv:2609.22966, 2026 on PithPith's reviewarXiv
  77. [94] Yanjiang Guo, Yen-Jen Wang, Lihan Zha, and Jianyu Chen. DoReMi: Grounding Language Model by Detecting and Recovering from Plan-Execution Misalignment. In2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2024 on Pith
  78. [95] Yanjiang Guo, Tony Lee, Lucy Xiaoyang Shi, Jianyu Chen, Percy Liang, and Chelsea Finn. VLAW: Iterative Co-Improvement of Vision-Language-Action Policy and World Model.arXiv preprint arXiv:2602.12063, 2026 on PitharXiv
  79. [96] DanijarHafner, TimothyLillicrap, IanFischer, RubenVillegas, DavidHa, HonglakLee, andJamesDavidson. Learning Latent Dynamics for Planning from Pixels. InProceedings of the 36th International Conference on Machine Learning, volume 97 ofProceedings of Machine Learning Research, p... on Pith
  80. [97] MinkyuHam,DonghoKim,ChanLee,MinJunKim,YixiZhang,JiayiWang,andHanLiu. SPINE:Bridging the Cyber-Physical Gap with Agentic AI.arXiv preprint arXiv:2607.13049, 2026 on PithPith's reviewarXiv

The first 100 of 300 references are listed here. All of them are in this paper's citation record.

Cited by 3 Pith papers

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