Robotic Foundation Models for Industrial Control: A Comprehensive Survey and Readiness Assessment Framework

David Kube; Simon Hadwiger; Tobias Meisen

cs.ROMar 6, 2026

Robotic Foundation Models for Industrial Control: A Comprehensive Survey and Readiness Assessment Framework

Authors: David Kube, Simon Hadwiger, Tobias Meisen

Organizations: Factory Automation, Siemens AG, Gleiwitzer Straße 555, Nuremberg, 90475, Germany. · Institute for Technologies and Management of Digital Transformation, Bergische Universität Wuppertal, Gaußstraße 20, Wuppertal, 42119, Germany.

Abstract

Robotic foundation models (RFMs) are emerging as a promising route towards flexible, instruction- and demonstration-driven robot control, however, a critical investigation of their industrial applicability is still lacking. This survey gives an extensive overview over the RFM-landscape and analyses, driven by concrete implications, how industrial domains and use cases shape the requirements of RFMs, with particular focus on collaborative robot platforms, heterogeneous sensing and actuation, edge-computing constraints, and safety-critical operation. We synthesise industrial deployment perspectives into eleven interdependent implications and operationalise them into an assessment framework comprising a catalogue of 149 concrete criteria, spanning both model capabilities and ecosystem requirements. Using this framework, we evaluate 324 manipulation-capable RFMs via 48,276 criterion-level decisions obtained via a conservative LLM-assisted evaluation pipeline, validated against expert judgements. The results indicate that industrial maturity is limited and uneven: even the highest-rated models satisfy only a fraction of criteria and typically exhibit narrow implication-specific peaks rather than integrated coverage. We conclude that progress towards industry-grade RFMs depends less on isolated benchmark successes than on systematic incorporation of safety, real-time feasibility, robust perception, interaction, and cost-effective system integration into auditable deployment stacks.

Figures & tables

Figure 1: Robotic control-method hierarchy: Each level increases flexibility, intelligence and generalisation capabilities, while reducing manual engineering effort and required expertise for operation and adaptation.

Figure 2: Publication increase. Visualising the amount of publications gathered per category.

Figure 3: Table of Contents. Highlighting preliminary sections in red, background in purple and the main review in green.

Figure 4: Overview of all utilised corpora and number of considered publications, as presented within this section.

Figure 5: General Literature Acquisition Pipeline Procedure: Manual user involvement is required only for the Manual Refinement steps (highlighted in pink); all other processes execute automatically based on user-input. Only the Query Proposal and Filter Process modules leverage LLMs, while the DB-Search is fully deterministic.

Figure 6: Literature Acquisition Pipeline: Process of gathering the RFM-related corpora (see Figure 7 ). Results yielded by manual cross-search are included for simplicity.

Figure 7: Main RFM-Related Corpus Composition

Figure 8: Literature Acquisition Pipeline: Process of gathering the industrial implication-related corpus.

Figure 9: Identified Related Works (Survey corpus) composition: GEA=general embodied AI, +RFMs=including RFMs, IND=industrial-focus.

Ref	Focus Area	Scope	Models	Span	Broad	Year
Ma et al. (2025b)	VLAs for embodied AI	Architectures, challenges, components, control, datasets, objectives, resources, simulators, tasks, task planning, taxonomy	Encoder, LLM, VLA, VLM	14-24	✓	2025
Zhong et al. (2025a)	Action tokenisation in VLAs	Action token categorisation: (affordance, code, goal state, language, latent, raw action, reasoning, trajectory), datasets, FM evolution	LFM, VFM, VLA, VLM	17-25	✓	2025
Xu et al. (2025a)	Challenges and roadmap in VLAs	Applications, challenges: (dataset and evaluation, execution, generalization, representation, safety), evolution, roadmap, milestones, modules, opportunities	VLA	22-25	✓	2025
Zhou et al. (2025b)	Language-conditioned robot manipulation	Auxiliary tasks, challenges, datasets, evaluation, environments, learning paradigms, limitations, neuro-symbolic, planning, reasoning, reward shaping, semantics extraction, task representation	FM, LLM, VLA, VLM	18-24	✓	2025
Shao et al. (2025)	Large VLM-based VLAs	Architectures (hierarchical, monolithic), benchmarks, characteristics, datasets, task planning, learning paradigms, memory mechanisms, multi-agents, operational strengths, taxonomy	VLA	23-25	✓	2025
Cui et al. (2025)	Dual-System VLAs	Architectures, definition, design-elements, empirical evaluation, training strategies	VLA	24-25	✗	2025

Table 1: RFM: Overview of most relevant survey/ review papers with RFM focus. Reporting the focus area, the review’s main considerations (Scope), the types of targeted model families (Models), the timespan of reviewed papers (Span), whether the survey is considered concise or extensive (Broad), and the survey’s publication year.

Ref	Focus Area	Scope	Models	Span	Broad	Year
Fan et al. (2025c)	VLMs in HRC for smart manufacturing	Architectures, human-robot skill transfer, limitations, manipulation, navigation, pretraining, task planning	LLM, VLM	20-24	✓	2025
Ren et al. (2024a)	Embodied intelligence in smart manufacturing	Applications, architecture, challenges, characteristics, capabilities, limitations	N/A	N/A	✗	2024

Table 2: IND: Overview of most relevant survey/ review papers with industrial focus.

Type	Config	Comments
Delta	(PPP) Parallel	• Several symmetric kinematic chains • High speed & acceleration • Redundancy increases accuracy
Cartesian	(PPP) Serial	• Perpendicular prismatic joints • Smooth linear movement • Rectangular workspace
Cylindrical	(RPP) Serial	• Rotational base combined with two prismatic joints • Cylindrical workspace
SCARA	(RRP) Serial	• Both revolute joints along the same axis • High accuracy & repeatability • Planar positioning
Spherical Arm	(RRP/RRR) Serial	• Revolute axes intersect at a common center point • Simplifies collision & reachability analysis • Spherical workspace
Articulated	(RRR) Serial	• Revolute joints’ axes typically offset • Similar to human arm • Flexible 3D workspace

Table 3: Main industrial robot configuration categories. Amiri et al. (2024) ; Brinker et al. (2017) ; Autsou et al. (2022) ; Collins and Yim (2016) ; Suri et al. (2018) , ( Khatib and Siciliano, 2016 , pp. 72, 222, 452, 1393)

$n=180$	DoF	Payload	Reach	Accuracy
		(kg)	(mm)	(mm)
Min	4	0.3	280	0.01
Max	7	170	2790	1
Mean	6.09	8.56	970.12	0.09
Median	6	5.5	910	0.05

Table 4: Composed cobot configuration and capability statistics. Taesi et al. (2023) Values are rounded to two decimal places.

$n=63$	Payload	TCP Velocity	Power Consumption
	(kg)	(m/s)	(kW)
Min	0.5	0.2	0.06
Max	20	6	0.9
Mean	6.69	1.68	0.31
Median	5	1.1	0.25

Table 5: Composed peak cobot TCP velocity and power consumption statistics. Taesi et al. (2023) Showing only values for typical cobots with a payload between

0.5

and

20

kg as in the wattage comparison of Taesi et al. Taesi et al. (2023) . Values are rounded to two decimal places.

Jetson Device	Price ( $)$ \downarrow$	TOPS (INT8) $\uparrow$	Memory (GB) $\uparrow$	Power (W) $\downarrow$	GOPS /\uparrow$	TOPS /W $\uparrow$	Memory (MB)/\uparrow$	$/year$ \downarrow$	TOPS /y\uparrow$
Orin Nano 4GB	229	34	4	25	148.47	1.36	17.47	54.75	0.62
Orin Nano 8GB	249	67	8	25	269.08	2.68	32.13	54.75	1.22
Orin Nano SuDK	249	67	8	25	269.08	2.68	32.13	54.75	1.22
Orin NX 8GB	479	117	8	40	244.26	2.93	16.7	87.6	1.34
Orin NX 16GB	699	157	16	40	224.61	3.93	22.89	87.6	1.79
Orin AGX 32GB	999	200	32	40	200.2	5	32.03	87.6	2.28

Table 6: Comparison of NVIDIA edge devices’ AI performance in INT8 precision (dense) BuyLatestJetson2025 (2025) ; JetsonT5000Modules2025 (2025) ; NVIDIARTXBlackwell2025 (2025) . Prices for all Jetson devices correspond to single-unit purchases from the same NVIDIA-specified retailer ArrowElectronicsConnect2025 (2025) , collected on the same date. Operational costs are estimated for 24/7 usage with 0.25$/kWh. All values are rounded to two decimal places. Su = Super, DK = Developer Kit, IND = Industrial

(1) Hardware	(13) Security	(25) Actions/ Outputs	(37) Precision & Accuracy
(2) Tasks	(14) Detection, Recognition & Understanding	(26) Skills	(38) Trust
(3) Environment	(15) Multimodal Inputs/ Outputs	(27) Observation/ Input	(39) Goal Specification & Interpretation
(4) Objects/ Materials	(16) Integration	(28) External Influences/ Disturbances	(40) Solution Strategy
(5) Instructability/ Commanding	(17) Safety	(29) Reaction	(41) Offline Adaptation (Finetuning/ Retraining)
(6) Distractors/ Distractions	(18) Compliance	(30) Validation/ Benchmark/ Evaluation	(42) Energy
(7) Adaptability/ Generalisation/ Flexibility	(19) Supervision	(31) Realtime Performance & Inference Frequency	(43) Data

Table 7: Consolidated list of the 47 attributes considered for each implication.

	Gemini Robotics 1.5 Team et al. (2025a)		OpenVLA Kim et al. (2024a)		$\pi_{0}$ Black et al. (2024)
Implication	Expert	LLM	Expert	LLM	Expert	LLM
I1	9	8	3	4	1	1
I2	1	1	0	0	0	0
I3	1	1	0	0	0	0
I4	1	1	1	1	0	0
I5	0	0	0	0	0	0

Table 8: Comparison between expert- and LLM-based evaluation results per implication for three representative RFMs.

Setting	$N$	Acy	F1	$\kappa$	FPR/FNR
All	447	0.966	0.762	0.744	1.9%/22.6%
InfoFound	304	0.951	0.762	0.734	2.9%/22.6%

Table 9: LLM-expert agreement on criterion fulfillment. Reporting the number of N evaluations considered, the overall Accuracy (Acy),

F_{1}

-score, Cohen’s kappa agreement (

\kappa

) and false positive/ negative rates (FPR/FNR).

Paper	Pos%	Acy	F1	$\kappa$	$J^{+}$
Gemini 1.5 Team et al. (2025a)	0.121	0.960	0.833	0.810	0.714
OpenVLA Kim et al. (2024a)	0.060	0.973	0.800	0.786	0.667
$\pi_{0}$ Black et al. (2024)	0.027	0.966	0.286	0.269	0.167

Table 10: Per-paper agreement. Pos% reports the expert positive rate and

J^{+}

the Jaccard similarity index, computed only for the positive class to ignore the dominating negatives.

Ref	Model	Total $\downarrow$	I1	I2	I3	I4	I5	I6	I7	I8	I9	I10	I11
Team et al. (2025a)	Gemini Robotics 1.5	0.12	0.44	0.06	0.07	0.07	0.00	0.00	0.00	0.07	0.14	0.23	0.09
Sun et al. (2025c)	Collab VLA	0.11	0.28	0.00	0.20	0.07	0.08	0.11	0.00	0.07	0.21	0.08	0.09
Li et al. (2025m)	GF-VLA	0.11	0.28	0.00	0.00	0.14	0.08	0.00	0.00	0.07	0.14	0.23	0.27
Lin et al. (2025a)	OneTwo VLA	0.11	0.33	0.00	0.20	0.14	0.00	0.11	0.10	0.07	0.14	0.00	0.09
Li et al. (2025b)	SC-VLA	0.11	0.33	0.06	0.00	0.21	0.23	0.11	0.00	0.00	0.00	0.15	0.09

Table 11: Overview of the five highest rated RFMs ( total ) of the

324

evaluated. Reporting their proportional industrial applicability ratings for each individual implication, and a total score over all implications. Per-criterion ratings are created utilising the publication’s context, by leveraging our LLM-based pipeline as described in Section 8.1 . Highest values within the five RFMs for each implication are shown in green, lowest in red. Values are rounded to two decimal places.

Ref	Model $\uparrow$	Total	I1	I2	I3	I4	I5	I6	I7	I8	I9	I10	I11
Sun et al. (2025c)	Collab VLA	0.11	0.28	0.00	0.20	0.07	0.08	0.11	0.00	0.07	0.21	0.08	0.09
Jones et al. (2025b)	FuSe	0.05	0.06	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.21	0.00	0.27
Team et al. (2025a)	Gemini Robotics 1.5	0.12	0.44	0.06	0.07	0.07	0.00	0.00	0.00	0.07	0.14	0.23	0.09
Huang et al. (2025c)	GraphCoT VLA	0.05	0.06	0.00	0.00	0.00	0.00	0.00	0.00	0.21	0.14	0.00	0.09
Li et al. (2025p)	HAMSTER	0.07	0.11	0.00	0.00	0.00	0.00	0.00	0.10	0.00	0.07	0.23	0.36
Wu et al. (2025a)	HiBer NAC	0.10	0.22	0.11	0.00	0.00	0.08	0.11	0.10	0.00	0.21	0.15	0.09

Table 12: Overview of the highest rated RFMs per implication of the

324

total evaluated. Per-criterion ratings are created utilising the publication’s context, by leveraging our LLM-based pipeline as described in Section 8.1 . Highest values within the respective implication are highlighted in green. Values are rounded to two decimal places. The entry ”Multiple” represents the fact that several works have the similar highest score of 0.11 for I6, which are thus consolidated in one row.

Figure 10: Proportional implication fulfillment. Depicting the five overall highest-rated RFMs (see Table 11 for detailed values).

Figure 11: Normalised per-implication criteria coverage. Each value represents the sum over all criteria fulfilled within the given implication & year, divided by the total number of papers published in the respective year.

Panel of Figure 12 — Figure 12: Implication-specific evaluation coverage across the full set of 324 RFMs (Full Corpus) and the Top-50 subset.

Metric	Corpus	I1	I2	I3	I4	I5	I6	I7	I8	I9	I10	I11
1+	Full	0.756	0.034	0.052	0.172	0.167	0.049	0.071	0.219	0.213	0.753	0.707
1+	Top 50	1.0	0.22	0.34	1.0	1.0	0.32	0.46	1.0	1.0	1.0	1.0
ACC	Full	0.089	0.003	0.007	0.016	0.014	0.005	0.008	0.017	0.019	0.099	0.097
ACC	Top 50	0.238	0.019	0.043	0.096	0.086	0.036	0.05	0.081	0.097	0.232	0.227

Table 13: Detailed comparison of

1_{+}

-Coverage and ACC between the full evaluated corpus (

324

) and the Top 50: For visualisation, see Figure 12 . Values are rounded to three decimal places.

Criterion	Impl	Fulfilled	Rate $\downarrow$	Expl(fulf)	Evid	ConfFul
Multi-Task Generalisation	I1	160	0.4938	0.2812	0.9938	0.8605
Flexible Data Type Acceptance	I11	158	0.4877	0.3671	0.9907	0.8693
Repeatability & Reproducibility	I10	145	0.4475	0.2483	0.9815	0.8474
Low-Data Adaptation & Few-Shot Requirements	I11	136	0.4198	0.8088	0.9907	0.9102
Module & Skill-Specific Benchmarking	I10	129	0.3981	0.0698	0.9938	0.8517
Generalisation to Environmental Variation	I1	101	0.3117	0.5446	0.9815	0.8739

Table 14: Top 10 most frequently fulfilled criteria across all papers. Reporting each criterion’s implication (I1-I11), absolute count and rate of fulfillment, explicitness among fulfilled cases (Expl(fulf)), evidence availability (Evid), and mean confidence when fulfilled (ConfFul). Values are rounded to four decimal places.

Impl $\uparrow$	Criterion	$g_{c}$	EvidRate $\downarrow$	FulfillRate
I1	Multi-Task Generalisation	0.5000	0.9938	0.4938
I2	Reactive, Real-Time Safety	0.8086	0.8117	0.0031
I3	Collaborative Adaptation	0.8735	0.8981	0.0247
I4	Environmental Robustness & Adaptation	0.9475	0.9907	0.0432
I5	Consistency Over Time & Contexts	0.8796	0.9969	0.1173
I6	Goal-Responsive Solution Strategies	0.9352	0.9722	0.0370

Table 15: Evidence-Fulfilled-Gap analysis: Reporting the gap score

g_{c}

alongside its evidence availability (EvidRate) and full-corpus fulfillment rate (FulfillRate) for each implication’s criterion with the highest evidence rate. Values are rounded to four decimal places.

Appendix figures & tables14 assets

Supplementary material from the paper’s appendix.

Appendix

Ref	Title	Year $\downarrow$
Sun et al. (2025a)	A Review of Embodied Grasping	2025
Yuan et al. (2025a)	A Survey of Behavior Foundation Model: Next-Generation Whole-Body Control System of Humanoid Robots	2025
Ma et al. (2025b)	A Survey on Vision-Language-Action Models for Embodied AI	2025
Zhong et al. (2025a)	A Survey on Vision-Language-Action Models: An Action Tokenization Perspective	2025
Gao et al. (2025b)	A Taxonomy for Evaluating Generalist Robot Policies	2025
Liu et al. (2025i)	Aligning Cyber Space with Physical World: A Comprehensive Survey on Embodied AI	2025

Appendix

Table 16: Full list of identified related works/ surveys. Total 55 publications from survey corpus as depicted in Figure 9 .

Implication	Category/ Criterion
I1:	A. Multi-Hardware and Configuration Adaptability
Adaptability & Flexibility	1. Multi-Hardware Support: Operates with different kinematics, sensors, and end-effectors without retraining.
	2. Hot-Swappable & Dimensionality Adaptation: Sensors/actuators can be replaced, added, or removed during operation. Automatically detects and configures input/output channels at runtime according to connected hardware and environmental context: Flexibly adapts the dimensionality, type and interpretation without manual intervention or system restart.
	B. Task Adaptability and Strategy Flexibility
	3. Multi-Task Generalisation: Handles multiple task types/categories, including unseen variations thereof, without retraining. Adapts strategies dynamically for task variations and can switch between tasks online.
	4. Restricted Autonomous Task Switching: Only switches between distinct tasks upon explicit user instruction: autonomy remains bounded, so adherence to the operator’s goal stays primary (may autonomously adapt solution approaches/ strategies).

Appendix

Table 19: Full criteria catalogue for industrial applicability rating as discussed in Section 7.3 . Derived from the general industrial implications (see Section 7.1 )

Ref	Model	Total $\downarrow$	I1	I2	I3	I4	I5	I6	I7	I8	I9	I10	I11
Team et al. (2025a)	Gemini Robotics 1.5	0.12	0.44	0.06	0.07	0.07	0.00	0.00	0.00	0.07	0.14	0.23	0.09
Sun et al. (2025c)	Collab VLA	0.11	0.28	0.00	0.20	0.07	0.08	0.11	0.00	0.07	0.21	0.08	0.09
Li et al. (2025m)	GF-VLA	0.11	0.28	0.00	0.00	0.14	0.08	0.00	0.00	0.07	0.14	0.23	0.27
Lin et al. (2025a)	OneTwo VLA	0.11	0.33	0.00	0.20	0.14	0.00	0.11	0.10	0.07	0.14	0.00	0.09
Li et al. (2025b)	SC-VLA	0.11	0.33	0.06	0.00	0.21	0.23	0.11	0.00	0.00	0.00	0.15	0.09
Wu et al. (2025a)	HiBer NAC	0.10	0.22	0.11	0.00	0.00	0.08	0.11	0.10	0.00	0.21	0.15	0.09

Appendix

Table 20: Overview of all

324

identified RFMs that include manipulation capabilities (for control & integrated; see Section 3.2 for composition details). Reporting their proportional industrial applicability ratings for each individual implication, and a total score over all implications. Per-criterion ratings are created utilising the publication’s context, by leveraging our LLM-based pipeline as described in Section 8.1 . Values are rounded to two decimal places.

Criterion	$\sum\downarrow$	% fulfilled
Multi-Task Generalisation	160	49.38%
Generalisation to Environmental Variation	101	31.17%
Autonomous Handling of Distractors/Obstacles	87	26.85%
Online/Mid-Execution Adaptation	49	15.12%
Autonomous Incorporation of Skills	22	6.79%
Online Environment Adaptation and Response	21	6.48%

Appendix

Table 21: Criterion fulfilment for I1 – Adaptability & Flexibility.

Criterion	$\sum\downarrow$	% fulfilled
Redundant, Independent Safety Mechanisms	5	1.54%
Command Compliance with Online-Safety	4	1.23%
Proactive, Plan-Level Safety	3	0.93%
Learning from Safety Events	2	0.62%
Decoupled, Real-Time Safety Control	1	0.31%
Reactive, Real-Time Safety	1	0.31%

Appendix

Table 22: Criterion fulfilment for I2 – Safety & Compliance.

Criterion	$\sum\downarrow$	% fulfilled
Intuitive Feedback Acceptance	13	4.01%
Collaborative Adaptation	8	2.47%
Transparent Reasoning Explanation	5	1.54%
Real-time Interruptibility	4	1.23%
Continuous Responsiveness	2	0.62%
Active Operator Exclusivity	0	0.00%

Appendix

Table 23: Criterion fulfilment for I3 – HRI & HRC.

Criterion	$\sum\downarrow$	% fulfilled
Disturbance Resilience & Continuous Operation	44	13.58%
Environmental Robustness & Adaptation	14	4.32%
Graceful Degradation & Strategy Adaptation	6	1.85%
Reproducible & Benchmarked Fault Diagnosis	5	1.54%
Adaptive Error Learning & Memorisation	3	0.93%
Logging, Explanation & Operator Guidance	1	0.31%

Appendix

Table 24: Criterion fulfilment for I4 – Robustness & Reliability.

Criterion	$\sum\downarrow$	% fulfilled
Consistency Over Time & Contexts	38	11.73%
Skill Integration with Known v/P&A	9	2.78%
v/P&A Learning from Experience & Feedback	6	1.85%
HRI/HRC Integration	3	0.93%
User-Defined Trade-offs	2	0.62%
Disturbance & Uncertainty Management	1	0.31%

Appendix

Table 25: Criterion fulfilment for I5 – Precision & Accuracy.

Criterion	$\sum\downarrow$	% fulfilled
Goal-Responsive Solution Strategies	12	3.70%
Historical Data Utilisation	2	0.62%
System-wide Real-time Compliance	1	0.31%
Task, Input & Environment Independence	1	0.31%
Autonomous Adaptation & Safety Response	0	0.00%
Dynamic Resource Management & Determinism	0	0.00%

Appendix

Table 26: Criterion fulfilment for I6 – Real-time Performance.

Criterion	$\sum\downarrow$	% fulfilled
Dynamic Resource Allocation	16	4.94%
Low-Cost Hardware Readiness	8	2.47%
Comprehensive Integration Documentation	1	0.31%
Autonomous Fault Handling & Feedback	0	0.00%
Cost-Reduction Memory & Adaptive Optimisation	0	0.00%
Flexible Input/Output Adaptation	0	0.00%

Appendix

Table 27: Criterion fulfilment for I7 – Cost-effectiveness & Integration Capabilities.

Criterion	$\sum\downarrow$	% fulfilled
Performance Benchmarking & OOD Feedback	63	19.44%
Explains Influences & Disturbances	4	1.23%
Reason Logging & Accessible Justification	3	0.93%
Task Feasibility & Capability Explanation	3	0.93%
Error, Anomaly & Safeguard Explanations	2	0.62%
User Feedback & Adaptive Improvement	2	0.62%

Appendix

Table 28: Criterion fulfilment for I8 – Explainability & Trust.

Criterion	$\sum\downarrow$	% fulfilled
Perceptual Feature Exposure & Modularity	32	9.88%
Comprehensive Scene Understanding via Fusion	29	8.95%
Multimodal Sensor Acceptance & Fusion	11	3.40%
Flexible Input Combination & Goal Specification	10	3.09%
Dynamic Sensor Prioritisation	2	0.62%
Peer-to-Peer Perception Knowledge Exchange	1	0.31%

Appendix

Table 29: Criterion fulfilment for I9 – Sensor Fusion & Perception.

Criterion	$\sum\downarrow$	% fulfilled
Repeatability & Reproducibility	145	44.75%
Module & Skill-Specific Benchmarking	129	39.81%
Generalisation & Adaptation Assessment	90	27.78%
Benchmark Diversity & Scope	47	14.51%
Instruction, Command, & Modality Coverage	3	0.93%
Explainability, Reasoning, & Feedback Benchmarks	1	0.31%

Appendix

Table 30: Criterion fulfilment for I10 – Standardised Benchmarking & Evaluation.

Criterion	$\sum\downarrow$	% fulfilled
Flexible Data Type Acceptance	158	48.77%
Low-Data Adaptation & Few-Shot Requirements	136	41.98%
Documented Data Composition & Simulation-Real Transfer	33	10.19%
Efficient Task, Environment, & Modality Adaptation	14	4.32%
Resilience Through Redundancy & Fault Tolerance	2	0.62%
Self-Assessment to Reduce Input & Modalities	1	0.31%

Appendix

Table 31: Criterion fulfilment for I11 – Data Requirements & Usage.

Robot Foundation Models

Robot Foundation Models are pretrained systems that combine perception, reasoning, and action to support diverse tasks and robot embodiments.

See all 135 papers

Foundation-Model-Based Agents in Industrial Automation: Purposes, Capabilities, and Open Challenges
May 4, 2026Vincent Henkel, Felix Gehlhoff, David Kube +13Industrial Automation AI Agent Evaluation
Institute of Automation Technology, Helmut Schmidt University / University of the Federal Armed Forces Hamburg, Hamburg, Germany. · Siemens AG, Nuremberg, Germany; Institute for Technologies and Management of Digital Transformation, Bergische Universität Wuppertal, Germany. · Artiquare GmbH, Ingolstadt, Germany. +11
How Should I Pick a Foundation Model for My Robot? In Favor of a Community Evaluation Framework for Social Robots
Aug 7, 2026Eric Nichols, Alva Markelius, Hatice GunesRobot Foundation Models Human-Robot Interaction
Honda Research Institute Japan, Wako, Japan. · Affective Intelligence and Robotics Lab, University of Cambridge, UK.
Verifiable Foundation Models for Robot Safety
Jun 22, 2026Davide Corsi, Kyungmin Kim, Roy FoxRobot Policy Learning Robot Foundation Models
University of California, Irvine

May 4, 2026J/K move · Enter open · S save

Foundation-Model-Based Agents in Industrial Automation: Purposes, Capabilities, and Open Challenges

Vincent Henkel, Felix Gehlhoff, David Kube, Asaad Almutareb+12

Institute of Automation Technology, Helmut Schmidt University / University of the Federal Armed Forces Hamburg, Hamburg, Germany. · Siemens AG, Nuremberg, Germany; Institute for Technologies and Management of Digital Transformation, Bergische Universität Wuppertal, Germany. · Artiquare GmbH, Ingolstadt, Germany.+11

Industrial Automation AI Agent Evaluation LLM Agent Reliability

Open paper AskPDF arXiv

Foundation models, particularly large language models, are increasingly integrated into agent architectures for industrial tasks such as decision support, process monitoring, and engineering automation. Yet evidence on their purposes, capabilities, and limitations remains fragmented across domains. This work examines how mature foundation-model-based agent systems are in industrial contexts, how their functional profile differs from conventional agent systems, and which limitations persist. A systematic literature survey following the PRISMA 2020 guideline is presented, screening 2,341 publications and synthesising a corpus of 88 publications through a structured coding scheme. The results show that reported systems are predominantly at prototype and early validation stages (75.0% at TRL 4-6), with deployment-oriented evidence remaining rare (9.1%). Operational goals are most frequently positioned in user assistance, monitoring, and process optimisation, while conventional production-control purposes such as planning and scheduling are less prominent. Compared with an established baseline for industrial agent systems, the capability profile reveals substantial gains in human interaction (+37%) and dealing with uncertainty (+35%), but a pronounced deficit in negotiation (-39%). The most widely reported limitations concern lack of generalization, hallucination and output instability, data scarcity, and inference latency. A working definition of foundation-model-based industrial agents is also proposed, bridging conventional agent theory, automation-engineering standards, and the foundation-model paradigm.

Abstract

Robot Foundation Models

Explore similar work

Foundation-Model-Based Agents in Industrial Automation: Purposes, Capabilities, and Open Challenges

How Should I Pick a Foundation Model for My Robot? In Favor of a Community Evaluation Framework for Social Robots

Verifiable Foundation Models for Robot Safety