LAISA: Low-Latency AI based Situation Awareness on 5G Data Streaming
Duration:
03/2026
—
02/2028
Total Time:
2 Years
The Project
In the LAISA-5G project, we are researching an ‘inference-as-a-service’ approach to AI-based perception for mobile robots. Cloud-based Visual Language Foundation Models (VLFMs) are integrated into edge robot systems via a low-latency 5G data connection. Our aim is to achieve efficient, scalable and near-real-time perception that reliably captures complex environments and enables the flexible integration of modern AI models into mobile robot units.
Our activities within the project
We are responsible for the development and implementation of cloud-based AI perception within the project. This includes the selection, adaptation and validation of vision-language models, the establishment of data and inference pipelines on HPC infrastructure, and their integration via 5G. We also develop the interfaces to the robot systems, contribute to demonstrations and evaluations, and oversee the scientific analysis and dissemination of the results.
Research groups
Downloads
Sponsors
FFG
Project Partners
ARTI - Autonomous Robot Technology GmbH
Project Details
The LAISA-5G project addresses key challenges in mobile robotics, in particular autonomous navigation and the semantic understanding of dynamic, unknown environments. Existing inspection robots often rely on structured operating environments, complex pre-mapping or specialised training data, and demonstrate limited robustness in the face of unforeseen situations. At the same time, there is a growing need for flexible, autonomous inspection solutions in sectors such as energy infrastructure, industrial facilities, and disaster and emergency response scenarios.
LAISA-5G therefore adopts an ‘inference-as-a-service’ approach, in which powerful, cloud-based Visual Language Foundation Models (VLFMs) are integrated into mobile robotic systems via low-latency 5G networks. Computationally intensive inference processes are offloaded from the edge to a central high-performance infrastructure. The robots continue to perform time-critical functions such as navigation and sensor data acquisition, whilst the AI models semantically analyse the incoming data streams and provide context-related information in near real time.
The 5G connection enables reliable, low-latency communication between robots and inference services, thereby laying the foundation for the scalable integration of modern AI functionalities. Sensor data, such as camera and LiDAR information, is continuously transmitted, processed and made available to the robots to inform their decision-making.
The key innovation lies in flexible, language-based perception: by using VLFMs, robots can recognise and interpret objects and structures in a context-dependent manner without relying on extensive, task-specific training datasets. This enables significantly greater adaptability in variable operating environments.
The approaches developed are evaluated in realistic demonstration scenarios and assessed for performance, latency behaviour and reliability in 5G standalone networks. LAISA-5G thus makes a significant contribution to the further development of scalable, AI-supported inspection systems and enhances efficiency, safety and sustainability in demanding application areas.
Project Participants
