Full workshop papers
High-quality submissions advancing grounded and faithful vision-language and vision-language-action models for real-world deployment.
NLP/CORE Unranked
VLM4RWD2026 2026
Grounded and Faithful Vision-Language Models for Real-World Deployment 2026
The VLM4RWD 2026 Workshop at NeurIPS focuses on grounded and faithful vision-language and vision-language-action models for real-world deployment, emphasizing reliable perception, reasoning, and decision-making in embodied and autonomous systems. It aims to advance methods for hallucination mitigation, causal reasoning, robust evaluation, and real-world reliability in multimodal AI.
Paper fit
Contribution paths
A strong submission should clearly identify its contribution and evaluate it appropriately.
Demo papers
Submissions presenting working demonstrations of systems or tools.
Extended abstracts
Concise summaries of ongoing or preliminary research.
Position papers
Opinionated or visionary perspectives on challenges and directions in grounded multimodal AI.
Datasets
New or curated datasets supporting grounding and faithfulness research.
Benchmarks
Evaluation frameworks or metrics for assessing grounded and faithful multimodal systems.
Emerging research ideas
Novel, early-stage concepts or exploratory work in the workshop's scope.
Research areas in scope
01
Grounded Understanding & Faithful Reasoning
Grounded perception and evidence localization for multimodal systemsFaithful reasoning and evidence-grounded decision-makingHallucination detection and mitigation in vision-language(-action) systemsSpatial, compositional, temporal, and relational reasoning in grounded multimodal systemsCausal reasoning, causal grounding, and counterfactual reasoning in multimodal systemsTraining paradigms and architectures for grounded VLMs and VLAs
02
Reliable Interaction & Decision-Making
Reliable decision-making for embodied agents under dynamic and uncertain conditionsOut-of-distribution detection and failure mode analysis for grounded multimodal pipelines
03
Evaluation & Deployment
Benchmarks, datasets, and evaluation methodologies for grounding and faithfulnessInterpretability and explainability of grounded multimodal reasoningEmerging directions in grounded multimodal intelligence: world models and agentic VLMs
Policies worth checking twice
- Submissions must undergo double-blind review and be fully anonymized.
- Papers may be up to 8 pages, excluding references and appendices, and must follow the NeurIPS 2026 format.
- Previously published work may be submitted for presentation but is not eligible for workshop awards.
- Workshop proceedings are non-archival; authors retain full copyright and may submit extended versions elsewhere.
- Submissions must be made through the OpenReview system.
Official sources
Compiled from the official call for papers. The organizers’ pages remain authoritative.
Last verified September 9, 2026