Trustworthy Generative and Agentic AI for Manufacturing Transformation
A multidisciplinary workshop on dependable AI systems that can understand manufacturing intent, generate operational artifacts, collaborate with humans, and act safely in real industrial environments.
Artificial Intelligence is rapidly moving beyond information retrieval, text generation, and decision support toward systems capable of generating executable artifacts, interacting with physical environments, coordinating multiple agents, and operating autonomously under human supervision.
Manufacturing presents a particularly promising yet challenging environment for such systems. Unlike many traditional AI applications, manufacturing AI directly influences physical assets, industrial processes, production schedules, robots, operators, and safety-critical control systems. Consequently, reliability, predictability, explainability, and trustworthiness become first-class requirements rather than desirable properties.
Recent progress in large language models, multimodal foundation models, reasoning systems, code-generation models, and agentic AI has demonstrated remarkable capabilities in software development, robotics, planning, and process automation. However, transferring these advances into industrial environments remains a major scientific challenge. Manufacturing systems involve heterogeneous data sources, long equipment lifecycles, strict timing constraints, proprietary protocols, complex workflows, and the need for continuous human oversight.
The workshop seeks to establish a research agenda for the next generation of AI-enabled manufacturing systems by bringing together researchers from artificial intelligence, robotics, software engineering, industrial automation, cyber-physical systems, formal verification, and human-computer interaction.
A motivating example is the emerging problem of Natural-Language-to-PLC Code Generation (NL2PLC). In modern manufacturing projects, customer requirements are typically expressed through informal control narratives that describe expected system behavior, safety requirements, sequencing logic, and operational constraints. Human engineers translate these narratives into executable industrial control programs compliant with the IEC 61131-3 standard.
Recent generative AI systems have demonstrated the ability to automate portions of this translation process. However, generated control programs must satisfy considerably stronger correctness requirements than conventional software generation tasks because failures may result in production downtime, equipment damage, economic loss, or safety hazards.
This challenge provides a compelling research testbed for trustworthy AI, program synthesis, formal verification, explainability, human oversight, and deployment readiness.
The manufacturing sector is entering a period of profound technological transformation driven by advances in foundation models, generative AI, robotics, industrial Internet-of-Things technologies, digital twins, and autonomous agents. While these technologies have demonstrated remarkable success in enterprise software, software engineering, and consumer applications, their application to industrial environments remains in its early stages.
The emergence of large language models and code-generation systems has led researchers to explore whether operational artifacts such as control programs, robot instructions, operating procedures, and maintenance plans can be automatically generated from high-level human intent. At the same time, agentic AI systems are increasingly being proposed for engineering support, predictive maintenance, production optimization, quality assurance, and autonomous factory operations.
These developments raise important research questions regarding reliability, verification, uncertainty estimation, explainability, and governance. Existing benchmarks and evaluation methodologies developed for general-purpose AI systems are often insufficient because manufacturing environments impose hard operational constraints and involve direct interaction with the physical world.
Several communities are addressing parts of this challenge—industrial foundation models, robotics, software engineering, formal verification, manufacturing systems, and trustworthy AI—but their efforts remain fragmented. AI4Factory aims to create a common forum in which these communities can define shared problems, evaluation practices, and a forward-looking research agenda.
Manufacturing also provides an opportunity to move beyond conventional evaluations of generative AI. Industrial AI systems can be assessed using measurable operational outcomes such as functional correctness, safety properties, execution traces, timing constraints, robustness, energy efficiency, human effort, and deployment readiness.
We invite research contributions that advance trustworthy, reliable, deployable, and human-centered AI for manufacturing. The list below is illustrative rather than exhaustive.
Add paper categories, page limits, submission platform, review policy, archival status, presentation format, and camera-ready instructions here.
The final schedule can be inserted here once the AAAI 2027 workshop calendar and submission timeline are confirmed.
AI4Factory is organized by researchers spanning artificial intelligence, formal methods, software verification, software engineering, model-based engineering, and natural language processing.
TCS Research
Lead Scientist (Artificial Intelligence)
Ashoka University
Assistant Professor, Formal Methods • Software Verification • Generative AI
Telecom Paris, Hasso-Plattner-Institut
Associate Professor, Software Engineering • Model-based Engineering
Ashoka University
Assistant Professor, AI/NLP