SC1: Safe Co-existence of Automated and Manual Transport at Industrial Sites
Go to SC1.1
SC2: Search & Rescue (SAR) and Emergency Response for Civil Safety
SC5: Quantum Sensor Multi-Modal, Multi-Physical Sensing at Highest Precision
SC7: Cooperative Multi-Agent Systems (Decentralized AI for Emergent Industrial Solutions)
SC1.1
Demo leader: Unikie
Partners involved: Mantsinen, Metsä Group, VTT
As part of the project, several different use cases were implemented. This case focuses on Metsä Group’s Rauma sawmill, where the goal was to explore autonomous log transportation in an industrial setting. The aim was to develop a solution that could automatically transport logs from the train unloading area to the sawmill’s feed table.
The solution is based on Unikie’s autonomous driving system, which uses sensors installed around the factory area to create a real-time digital twin of the environment. This system controls a ground vehicle developed by Mantsinen, capable of transporting log cages while monitoring surrounding activity. Operational safety was a key research focus due to the semi-open nature of the environment, where other vehicles and pedestrians may be present.
After three years of research and development, the concept was successfully tested in October 2025 at the Rauma sawmill by Unikie and Mantsinen. The site was equipped with sensors, cabling, and a server that enabled the ground vehicle to drive from behind the drying building to the sawmill’s feed table. A real-world use case was tested—albeit on a smaller scale—by transporting full log cages to the unloading area and returning empty cages to the loading point. Safety was also tested by walking in front of the vehicle, and driving precision was validated with sub-10 cm accuracy when positioning under the log cages.
All tests were successful, despite installation challenges due to a tight schedule. The ground vehicle was able to repeatedly position itself under the log cages and transport full loads along a predefined route. It stopped reliably—whether empty or loaded—when a person entered its safety zone. The concept worked as intended, and no obstacles were identified that would prevent commercial deployment.
We are proud to say that this is one of the world’s first solutions where a heavy ground vehicle can move autonomously and safely among people and manually operated machines in an industrial environment.
Kalmar Reach Stacker with autonomous loading and unloading of containers.
SC1.2 Rauma
Demo leader: Unikie
Partners involved: Mantsinen, Metsä Group, VTT
As part of the project, several different use cases were implemented. This case focuses on Metsä Group’s Rauma sawmill, where the goal was to explore autonomous log transportation in an industrial setting. The aim was to develop a solution that could automatically transport logs from the train unloading area to the sawmill’s feed table.
The solution is based on Unikie’s autonomous driving system, which uses sensors installed around the factory area to create a real-time digital twin of the environment. This system controls a ground vehicle developed by Mantsinen, capable of transporting log cages while monitoring surrounding activity. Operational safety was a key research focus due to the semi-open nature of the environment, where other vehicles and pedestrians may be present.
After three years of research and development, the concept was successfully tested in October 2025 at the Rauma sawmill by Unikie and Mantsinen. The site was equipped with sensors, cabling, and a server that enabled the ground vehicle to drive from behind the drying building to the sawmill’s feed table. A real-world use case was tested—albeit on a smaller scale—by transporting full log cages to the unloading area and returning empty cages to the loading point. Safety was also tested by walking in front of the vehicle, and driving precision was validated with sub-10 cm accuracy when positioning under the log cages.
All tests were successful, despite installation challenges due to a tight schedule. The ground vehicle was able to repeatedly position itself under the log cages and transport full loads along a predefined route. It stopped reliably—whether empty or loaded—when a person entered its safety zone. The concept worked as intended, and no obstacles were identified that would prevent commercial deployment.
We are proud to say that this is one of the world’s first solutions where a heavy ground vehicle can move autonomously and safely among people and manually operated machines in an industrial environment.
Testing envrionment at Metsä Group’s Rauma sawmill.
Testing operational safety with full log cage.
SC2: Search & Rescue (SAR) and Emergency Response for Civil Safety
SC2.1 Single-robot
Demo leader: Virtual Vehicle Research
Partners involved: AVL, HUAWEI Technologies Sweden AB, Pumacy, Universidad de Alcala, Montanuniversitaet Leoben, Ostbayerische Technische Hochschuleamberg-Weiden, AIT Austrian Institute of Technology, Bundesministerium fuer Landesverteidigung und Sport, EDI - Institute of Electronics and Computer Science, Laabmayr
Demonstrator SC2.1 represents a robotic platform capable of supporting Search and Rescue (SAR) operations in subterranean and GNSS-denied environments. As part of SC2, its purpose is to validate autonomous localization, mapping, and survivor detection capabilities under real-world underground conditions at the tunnel facilities at Zentrum am Berg (ZaB). The platform combines robust hardware, advanced perception technologies, and AI-driven algorithms to enable autonomous mission execution in challenging, GPS-inaccessible tunnel systems.
SC2.1 is an Unmanned Ground Vehicle (UGV) equipped with a suite of specialized sensors (e.g. LiDAR, thermal camera) and an onboard computing platform tailored to execute enhanced algorithms enabling autonomous navigation and AI-based survivor detection for complex underground missions.
Due to an extensive simulation study, robust open-source SLAM algorithms dealing well with featureless environments were selected to be implemented on the UGV. The Survivor Detection pipeline aims to detect and estimate the poses of potential human survivors near the robot platform.
Two approaches are introduced, an onboard method utilizing thermal images and LiDAR pointcloud, and an offboard method leveraging only thermal images. Data communication interfaces are established to transmit survivor positions and mission updates to emergency personnel. A wireless system simulator provides information about relevant communication attributes within the tunnels.
Additionally, SC2.1 supports a complementary feasibility study of magnetic field-based mapping and self-localization, evaluating the potential of magnetometer signals as an alternative positioning method in underground settings. This approach could, in the future, significantly enhance localization robustness where other signals are unreliable.
SC2.1 combines multiple innovative elements that push the state of the art in subterranean SAR robotics:
-
Autonomous operation in GNSS-denied subterranean environments: SC2.1 successfully performs localization and supports navigation without GNSS which is an essential requirement for tunnel missions.
-
AI-supported survivor detection: Integrated perception algorithms allow the UGV to identify survivors and provide their estimated positions to human responders, enabling faster and safer rescue operations.
-
Quantum sensing supported localization (future integration): In the context of SC2.1 the incorporation of a quantum sensor from SC5 will further elevate localization precision under highly degraded conditions.
-
Real-world validation: SC2.1 is deployed at the ZaB facilities, a unique European-scale experimental site offering realistic tunnel geometries and material properties.
These innovations allow SC2.1 to act as a reference platform for safe, autonomous, and efficient underground operations. The capabilities of SC2.1 are directly applicable to multiple real-world scenarios where human access is dangerous, slow, or impossible, such as SAR missions in collapsed e.g. tunnels, exploration and inspection tasks in GNSS-denied environments, safety assessment afterdisasters.
SAR real-world demonstrator vehicle at ZaB.
Visualization of the position of the SPIDER and the video stream from the UGV visualized in the XR COMMAND tool (provided by LAAB).
SC2: Search & Rescue (SAR) and Emergency Response for Civil Safety
SC2.2 Multi-robot
Demo leader: Virtual Vehicle Research
Partners involved: Universidad de Alcala, Montanuniversitaet Leoben, Ostbayerische Technische Hochschuleamberg-Weiden, AIT Austrian Institute of Technology, Bundesministerium fuer Landesverteidigung und Sport
Demonstrator SC2.2 entirely operates in simulation representing a simulation-based robotic platform and the operation environment within SC2 focusing on cooperative multi-robot system. A high-fidelity simulation model of ZaB and SC2.1 were developed enabling two or more UGVs to collaboratively localize and map a subterranean environment. The virtual attribute of the demo offers scalable and flexible development opportunities for testing advanced multi-agent strategies and communication-oriented behaviours without the logistical constraints and safety risks of physical deployment. The Demo comprises a complete simulation environment that models underground tunnel conditions, UGV operation, and communication effects. A detailed model of the ZaB tunnel facilities based on LiDAR measurements was developed within SC2.2, representing connected highway and railway tunnel tubes. Two UGV models are available in simulation, a Turtlebot4 platform and the SPIDER vehicle, including designated sensor characteristics. The multi-robot localisation and mapping modules allow at least two agents to build a shared representation of the tunnel environment. The effects of communication on multi-robot SLAM systems are covered in a conducted simulation study focusing on communication delay, de-synchronization and packet loss. Mitigating communication link failures between robots and/or a communication centre, an AI-based Quality of Service (QoS) prediction model is developed, providing forecasts of communication performance and blind spots at different locations.
SC2.2 introduces several innovations that advance the state of the art in subterranean SAR robotics:
-
Cooperative multi-robot SLAM in GNSS-denied environments: SC2.2 demonstrates how multiple UGVs jointly localize and map the ZaB tunnels without GNSS.
-
Communication-aware decision-making: A future integration of the QoS prediction into the navigation and coordination modules of the UGV will allow the agents to adjust their strategy to maintain reliable data exchange.
-
Simulation framework for subsurface SAR scenarios: SC2.2 provides a flexible and scalable validation and testing environment for evaluating algorithms and models before deploying them onto real systems, reducing development risks and costs.
The SC2.2 simulation-based demonstrator offers broad potential for applicability reaching from SAR research and robotics development over to safety-critical planning, such as research on GNSS-denied localisation and sensor fusion; pre-mission planning and optimisation before first responders enter hazardous environments; training and validation of AI-driven multi-agent coordination algorithms; etc.
SPIDER operating in one highway tunnel tube executing a SLAM algorithm (Fast-LIO2).
Example of reconstructed map using multi-robot SLAM with a fleet of 4 robots.
SC5: Quantum Sensor Multi-Modal, Multi-Physical Sensing at Highest Precision
SC5.1 Quantum sensing using NV-Centres in diamond
Quantum technologies are commonly grouped into three major domains: quantum computing, quantum communication, and quantum sensing. While quantum computing promises transformative capabilities for solving complex computational problems and quantum communication aims to provide fundamentally secure information transfer, quantum sensing is currently the most mature of the quantum technology domains and the closest to broad practical adoption. A-IQ Ready brings quantum sensing from laboratory prototypes to industrial and commercial applications. Specifically, we focus on Nitrogen Vacancy (NV) centre defects in diamond, which can be used to develop ultra-precise magnetometer sensors, and have the unique capability to operate at room temperatures. We join our interdisciplinary expertise to:
-
improve “out-of-the-box” usability of the sensor,
-
reduce its size and costs,
-
and achieve the world's fastest readout (while preserving advantageous performance).
A-IQ Ready’s quantum sensor system and sensor miniaturised fibre-based sensor head design.
Measurements for underground/indoor localisation
In our recent indoor localisation tests, we explored the potential of quantum sensor magnetometers for precise
spatial measurements. The study was conducted over two iterations to assess the reliability and accuracy of the
technology in dynamic environments. For the second iteration, we took a step further by automating the process, leveraging a ground-truth OptiTrack positioning system alongside a robot to ensure consistent and repeatable measurements. This enhanced approach allowed us to refine our data collection methods and minimise human error. In the data, you can observe a magnetic field anomaly, likely caused by a structural element of the building, which highlights the sensitivity of the quantum sensor in detecting even the smallest variations in the environment. This anomaly is an interesting finding, demonstrating the complexities involved in indoor localisation and the power of quantum sensors in uncovering hidden details within physical spaces.
Setups for initial and automated magnetic field mapping tests. (a) Initial tests at the premises of VIF; (b) Sensor deployment on automated robot platform; (c) Indoors magnetic field mapping at EDI’s automated testing rig.
Measurements for motor control
This motor control use case presents a fascinating challenge, as it involves a quantum sensor that requires an exceptionally high framerate for accurate measurements. To tackle this, we integrated the fibre-based sensor head of the quantum sensor directly inside a motor, a complex task that required precise engineering. During our tests, we gathered data that demonstrated the sensor's ability to track and determine the motor's phase. This breakthrough is significant, as it opens up new possibilities for using quantum sensors in real-time motor diagnostics and control systems.
Measurement setup for motor control tests using a quantum magnetometer. (a) Measurement setup; (b) Measurement data illustrating the motor’s rotation phase; (c) Fibre-based sensor head concept.
Next steps
The advancements achieved in our quantum sensor development already provide a highly valuable input for the continued evolution of the technology, particularly within Chips-JU Horizon Europe initiatives such as ARCHIMEDES, Cynergy4MIE, and MOSAIC. Insights gained from these projects have directly informed the refinement of both system design and implementation strategies. Notably, the analogue electronics circuitry we developed is now undergoing miniaturisation, marking an important step toward more compact and scalable solutions. In parallel, we have introduced an optimised event-driven control hardware architecture, designed to enhance efficiency and responsiveness. Together, these developments represent a significant stride toward practical, deployable quantum sensing systems that can meet the demands of real-world applications.
Miniaturisation effort of quantum sensor’s analogue electronics.
SC7: Cooperative Multi-Agent Systems (Decentralized AI for Emergent Industrial Solutions)
SC7.1 AGV system
Demo leader: Rosenheim University of Applied Sciences
Partners involved: Safelog, ScaliRo
The technical system of Demonstrator 7.1 has been successfully implemented in the laboratory environment of
Proto_lab. In close collaboration between Safelog and THRO, the automated guided vehicles (AGVs) and the necessary peripheral components, such as charging stations and actuators for automated gate control, have been installed and tested. This step represents a significant advancement in the development of the AGV system. As part of the implementation, a shift from WIFI to 5G communication infrastructure was carried out. This measure aims to substantially improve the performance of data transmission between various hardware components and the control system. Tests have demonstrated that 5G technology offers stable connections and faster data rates, which are crucial for the efficiency of the overall system.
The cloud-based control system was established by our project partner ScaliRo. In this context, THRO programmed the travel paths, routes, and internal processes to compare them with the reference use case FT06. Continuous testing during the implementation phase allowed for further optimization of the overall system performance.
A standout result of this phase is the reduction of the throughput time using a standard controller to manage a FT06 pass. With a seamless operation, the throughput time was optimized to approximately 1 hour and 50 minutes. This represents a significant improvement in efficiency. Additionally, the use of the controller from WP 4, which includes a newly developed AI controller, was also successfully tested in the simulation environment of the ScaliRo Cloud. The performance of this new technology was slightly better in terms of throughput time compared to the previously used standard controllers from ScaliRo.
Overall, the current status of Demonstrator 7.1 shows that all technical components are working seamlessly
together, and the set goals regarding system optimization have largely been achieved. The close collaboration of project partners has proven vital for the success of this implementation. The next steps will focus on integrating the insights gained into the further development of the AGV system and conducting additional tests to validate stability and performance under real-world conditions.
Setup of the training environment.
Technical implemented fleet of AGVs in the laboratory environment.
SC7: Cooperative Multi-Agent Systems (Decentralized AI for Emergent Industrial Solutions)
SC7.2 Virtual Training Platform for AMR Navigation in Unstructured Warehouses
Overview:
We deliver a modular virtual training platform that trains, evaluates, and verifies reinforcement-learning (RL)
navigation policies for autonomous mobile robots in challenging warehouse environments. The toolchain
integrates a Unity-based digital twin, Nav2-inspired model-based controllers, specification-driven reward
shaping, and ontology-based scenario generation into a reproducible pipeline for offline-to-online RL and
automated validation.
Developments:
• Simulation & tooling (TUM): Procedural Unity environments with configurable physics, lighting, and
appearance randomization; multi-modal sensors (RGB BEV/front, LiDAR, IMU, experimental quantum
magnetometer model); deterministic annotations and a gRPC API for state/action exchange. Packaged as
Docker images for on-premises use.
-
Control & data pipeline (TUM): ROS2-decoupled Nav2-inspired controllers (A*/local planner/DWA) for
baseline behavior and supervised trajectory logging to bootstrap offline RL (TD3-BC) and subsequent online
fine-tuning (PPO/SAC/online-TD3).
-
Specification & monitoring (AIT): Hierarchical potential-based reward shaping (HPRS) and hybrid runtime
monitoring (approximate + exact) that translate formal temporal requirements into dense rewards and
automated oracles for safety checks.
-
Ontology & test generation (TUG): Browser app to author ontologies, convert constraints into parameter sets,
combinatorial generation (pairwise/3-wise), OpenSCENARIO bundles for compact, reproducible test suites
and curriculum progression.
Key innovations:
-
Structured realism: semantically constrained domain randomization (color palette spacing,
probability-weighted lighting, physically coherent parameter ranges) to produce realistic, diverse training
distributions that improve transfer.
-
Disturbance & delay awareness: explicit disturbance channels and delay hooks in simulation, plus
disturbance-aware representations in policies for more robust, delay-resolved control.
-
Spec-to-reward integration: automated pipeline from formal requirements to prioritized shaped rewards,
reducing manual reward engineering and enabling verified training.
-
Reproducible, semantic testing: ontology-driven scenario generation yields machine-checkable regression
suites and critical-scenario libraries for safety validation.
Applications:
-
Intralogistics: AGV/AMR navigation, routing, docking, congestion management, and multi-robot coordination.
-
Research & education: reproducible RL benchmarks, lab courses, and student projects.
-
Industrial pre-validation: scenario-based risk assessment and pre-deployment testing for system integrators and operators.
Partner contributions for SC7 demo. 2.
Scalable simulations with canonical colour randomization for robust agent training.
Integration flow of components developed by each partner as a sequence diagram.