VeriDream · Project

AI-Powered Warehouse Robots That Learn and Adapt to Real-World Conditions

manufacturingTestedTRL 5

Imagine a warehouse robot that freezes every time a box is in the wrong spot or the lighting changes — that's the reality today. VeriDream took AI methods from two earlier EU projects and made them tougher and more flexible so robots can handle the messy, unpredictable reality of a real warehouse. Think of it like teaching a robot not just to follow a script, but to improvise when things go sideways. The team tested this at a real warehouse robotics startup called Magazino and also explored how smaller companies beyond logistics could use the same AI toolkit.

By the numbers

EUR 1,988,670

EU funding for AI robotics research

consortium partners across 4 countries

SMEs in the consortium

project deliverables produced

previous H2020 projects (DREAM and RobDREAM) that this builds on

The business problem

What needed solving

Warehouse robots today are brittle — they work fine in controlled setups but fail when real-world conditions change: different product shapes, unexpected obstacles, shifted inventory. Every failure means downtime, manual intervention, and lost productivity. Companies need robots that can handle surprises on their own, but current AI methods are too rigid for the variability of physical environments.

The solution

What was built

The project produced 8 deliverables including AI methods for continual learning, failure discovery and resolution, and state representation learning — all designed to make robots more adaptive. A documented demonstration showed the AI methods working in the Space Engineers video game environment. The warehouse logistics use cases at startup Magazino targeted high TRL applications for pick-and-place robotics.

Audience

Who needs this

E-commerce fulfillment centers with robotic picking systemsThird-party logistics providers automating warehouse operationsRobotics integrators customizing solutions for multiple clientsManufacturing SMEs adopting their first robotic automationAI startups building autonomous agents for physical environments

Business applications

Who can put this to work

Warehouse & Logistics

mid-size

Target: E-commerce fulfillment centers and third-party logistics operators

If you are a logistics operator dealing with robots that fail when products change shape, position, or packaging — this project developed AI methods for continual learning and failure recovery that let warehouse robots adapt without reprogramming. The work was tested at Magazino, a warehouse robotics startup among the 7 consortium partners, targeting high TRL readiness for pick-and-place operations.

Industrial Automation

SME

Target: Robotics integrators and manufacturing SMEs adopting automation

If you are a robotics integrator struggling to customize robot behavior for every new client environment — this project built generalized AI methods for state representation learning and failure discovery that reduce the need for case-by-case programming. The broad innovation strategy was specifically designed to help SMEs adopt these methods, with 3 SMEs among the 7 consortium partners.

Software & AI

SME

Target: AI startups building autonomous systems beyond robotics

If you are an AI company looking for proven methods to make autonomous agents more robust in unpredictable environments — this project developed transferable AI techniques including continual learning and failure resolution, demonstrated across multiple use cases. GoodAI even tested the methods inside a video game environment, showing the techniques work beyond physical robotics.

Frequently asked

Quick answers

What would it cost to implement these AI methods in our warehouse?

The project received EUR 1,988,670 in EU funding across 7 partners over 2.5 years. Implementation costs for your facility would depend on your existing robotics setup and integration complexity. Contact us for a tailored assessment based on your warehouse operations.

Can this scale to a full warehouse operation with hundreds of robots?

The deep innovation strategy targeted high TRL results specifically for warehouse logistics at Magazino, a real robotics startup. However, the project was a Research and Innovation Action (RIA), meaning full industrial-scale deployment would require additional engineering and validation beyond what was demonstrated.

Who owns the intellectual property and can we license it?

IP is shared among the 7 consortium partners across 4 countries (CZ, DE, FR, IT), coordinated by the German Aerospace Center (DLR). Licensing terms would need to be negotiated with the relevant partners. SciTransfer can facilitate introductions to the right contact.

How does this differ from off-the-shelf warehouse robotics AI?

Most commercial warehouse robots work well in controlled conditions but break down when the real world is messy. VeriDream specifically tackled failure discovery and resolution plus continual learning — meaning the robot identifies its own mistakes and improves over time without manual reprogramming.

What was actually demonstrated and tested?

The project produced 8 deliverables including a demonstration of AI methods within the Space Engineers video game environment by GoodAI. The warehouse logistics use cases at Magazino targeted high TRL. Based on available project data, detailed results from the Magazino demonstrations are not publicly summarized.

Is there regulatory risk in deploying AI-driven warehouse robots?

The project focused on AI robustness and failure handling, which actually supports compliance with emerging EU AI Act requirements for high-risk AI systems. Warehouse robotics falls under industrial safety regulations that vary by country, but improved failure detection is a positive factor.

Consortium

Who built it

The consortium of 7 partners across Germany, Czech Republic, France, and Italy blends research muscle with commercial intent. DLR (German Aerospace Center) coordinates, bringing world-class robotics expertise. With 3 SMEs and a 29% industry ratio, the project has genuine commercial pull — notably through Magazino, a warehouse robotics startup that served as the primary industrial testbed. Two universities and two research organizations provide the scientific foundation, while GoodAI and Synesis represent the broader AI startup ecosystem. The mix is well-suited for taking lab research into real-world testing, though the relatively low industry ratio suggests the technology still needs commercial partners to reach full deployment.

DEUTSCHES ZENTRUM FUR LUFT - UND RAUMFAHRT EVCoordinator · DE
SYNESIS-SOCIETA CONSORTILE A RESPONSABILITA LIMITATAparticipant · IT
GOODAI RESEARCH SROparticipant · CZ
ASSOCIATION POUR LA RECHERCHE ET LE DEVELOPPEMENT DES METHODES ET PROCESSUS INDUSTRIELSparticipant · FR
ECOLE NATIONALE SUPERIEURE DE TECHNIQUES AVANCEESthirdparty · FR
MAGAZINO GMBHparticipant · DE
SORBONNE UNIVERSITEparticipant · FR

How to reach the team

The coordinator is DLR (German Aerospace Center). SciTransfer can help you reach the right person on the VeriDream team.

Order a report on this consortium See DEUTSCHES ZENTRUM FUR LUFT - UND RAUMFAHRT EV profile →

Next steps

Talk to the team behind this work.

Want to explore how VeriDream's adaptive robotics AI could work in your warehouse or factory? Contact SciTransfer for a confidential briefing and introduction to the research team.

Order a report on this topic More in Manufacturing & Industry 4.0