SciTransfer
STAR · Project

Trustworthy AI That Makes Factory Robots Safer for Workers and Harder to Hack

manufacturingTestedTRL 6
star-ai.eu

Imagine putting a smart robot on your factory floor, but you can't fully trust it — it might make a dangerous move near a worker, or a hacker could trick it into producing defective parts. STAR built a toolkit of AI safety features: systems that explain why the robot made a decision, digital copies of production lines to test scenarios without risk, and cyber-defence shields that stop attackers from corrupting the AI's brain. Think of it as a seatbelt, airbag, and anti-theft system rolled into one — but for industrial AI.

By the numbers
17
consortium partners
11
countries represented
12
working prototype deliverables
47
total project deliverables
71%
industry partners in consortium
4
SMEs in consortium
The business problem

What needed solving

Manufacturers deploying AI on their production lines face three interlinked risks: the AI makes unsafe decisions near human workers, nobody can explain why the AI did what it did (creating liability and compliance headaches), and sophisticated cyberattacks can corrupt AI models to cause quality failures or safety incidents. These risks are blocking adoption of advanced AI in factories where it could deliver the most value.

The solution

What was built

STAR delivered 12 working prototypes: an integrated AI platform, reinforcement learning for autonomous mobile robots, a human-robot collaboration knowledge base, active learning systems, digital twins for security and safety, a market platform with Virtual Digital Innovation Hub beta, cyber-defence mechanisms against AI poisoning and evasion attacks, visual safety zone detection, simulated reality for human-robot collaboration, an Explainable AI algorithm library, security and data governance infrastructure, and decentralized reliability for industrial data analytics.

Audience

Who needs this

Automotive and aerospace manufacturers deploying cobots on assembly linesElectronics manufacturers using AI-driven visual quality inspectionWarehouse and logistics operators with autonomous mobile robot fleetsIndustrial plant managers responsible for worker safety around robotic systemsManufacturing IT security teams protecting AI systems from cyberattacks
Business applications

Who can put this to work

Automotive Manufacturing
enterprise
Target: Car or auto-parts manufacturers running collaborative robots on assembly lines

If you are an automotive manufacturer dealing with safety incidents around cobots on the shop floor — STAR developed a visual safety zone detection system and human-centric digital twins that monitor workers in real time. These prototypes were validated in quality management and human-robot collaboration scenarios across a 17-partner consortium with 71% industry participation.

Electronics & Precision Manufacturing
mid-size
Target: Electronics manufacturers using AI-driven quality inspection

If you are a precision manufacturer worried that your AI quality-inspection system could be fooled by adversarial attacks or produce unexplainable reject decisions — STAR built a library of Explainable AI algorithms and cyber-defence mechanisms against poisoning and evasion attacks on deep neural networks. These were delivered as working prototypes tested in real manufacturing conditions.

Warehouse & Logistics Automation
any
Target: Logistics companies deploying autonomous mobile robots in warehouses

If you are a logistics operator struggling with autonomous mobile robots that navigate unpredictably in busy warehouse environments — STAR developed reinforcement learning techniques specifically for autonomous mobile robot navigation and safety zone detection. The integrated STAR platform combines these with active learning to accelerate safe decision-making in dynamic settings.

Frequently asked

Quick answers

What would it cost to implement STAR technologies in our plant?

STAR was a publicly funded research project, not a commercial product with set pricing. The technologies exist as 12 working prototypes. Licensing or integration costs would depend on negotiations with individual consortium partners — contact the coordinator or relevant technology providers within the 17-partner consortium.

Can these AI safety tools scale to a full production line, not just a lab demo?

STAR validated its technologies in manufacturing line scenarios covering quality management, human-robot collaboration, and agile manufacturing. The consortium included 12 industry partners (71% of the consortium), suggesting the tools were designed with industrial scale in mind. However, moving from validated prototypes to full-scale deployment would require additional integration work.

Who owns the IP and can we license specific components?

IP is distributed across the 17 consortium partners from 11 countries, governed by the EU grant agreement. Key components like the XAI library, digital twins, and cyber-defence mechanisms each have specific partner owners. Licensing discussions would go through the coordinating partner, Netcompany SA (Belgium).

How does the Explainable AI component actually work in practice?

STAR delivered a library of XAI algorithms as a final prototype. These make AI decisions in manufacturing transparent — so when an AI system flags a quality defect or navigates a robot, operators can see why that decision was made. This directly addresses EU AI Act requirements for transparency in high-risk AI systems.

Is this compliant with the EU AI Act?

STAR was explicitly designed around safe, trusted, and human-centric AI principles. Its Explainable AI library, safety zone detection, and human-centric digital twins directly address EU AI Act requirements for high-risk industrial AI systems. The project aligned with EFFRA and AI4EU initiatives for standard-based deployment.

What is the timeline from first contact to a working pilot?

The core technologies exist as final-version prototypes delivered by end of 2023. Based on available project data, an integration pilot could build on the existing STAR platform, which already combines multiple AI modules. Timeline would depend on your specific manufacturing setup and which of the 12 prototype components you need.

Consortium

Who built it

This is a large, industry-heavy consortium: 12 out of 17 partners (71%) come from industry, with only 3 universities and 2 research organizations. That ratio is unusually high for an EU research project and signals that the technologies were built with real factory floors in mind, not just academic papers. The consortium spans 11 countries across Europe, coordinated by Netcompany SA from Belgium — a systems integration company, not a university, which further tilts the project toward practical implementation. Four SMEs participated, bringing agility and niche expertise alongside the larger industrial players.

How to reach the team

Netcompany SA (Belgium) — a systems integration company coordinating the 17-partner consortium. Contact through the project website or CORDIS portal.

Order a report on this consortiumSee NETCOMPANY SA profile →
Next steps

Talk to the team behind this work.

Want to know which STAR partner built the component you need? SciTransfer can identify the right contact and arrange an introduction.

Order a report on this topicMore in Manufacturing & Industry 4.0
More in Manufacturing & Industry 4.0
See all Manufacturing & Industry 4.0 projects →
Prediction Software That Tells You How Polymers Behave Before You Make Them
PolySolv
Open →
Unbreakable Pivot Pins That Eliminate Heavy Machinery Downtime and Joint Failures
BTPin
Open →
First Large-Scale Plant Turning Wood Pulp Into High-Performance Bio-Based Material
EXILVA
Open →