Research & Development
Pushing the frontier
A decade of delivered EU R&D — and a pipeline of new proposals advancing the frontier of Physical AI, coherent agentic systems, and trustworthy edge deployment.
Delivered projects
Horizon Europe and H2020 projects successfully completed.
Network of excellence building distributed, trustworthy, and scalable AI at the edge — connecting European research institutions and industry to advance real-world edge AI deployment.
PRAESC relevance: Physical AI architecture and edge inference pipelines validated at industrial scale — directly underpinning PRAESC's robotics and manufacturing deployment approach.
Initiative building a complete hardware-software stack for efficient edge AI on RISC-V processors — targeting performance and energy efficiency for embedded applications across autonomous inspection, industrial vision, and smart appliances.
PRAESC relevance: Hardware-software co-design methodology for resource-constrained embedded targets, directly inform PRAESC's model compression and edge deployment approach.
AI-as-a-Service platform for the deep edge, delivering modular services — model compression, optimisation, benchmarking, and deployment — to help European SMEs adopt AI on embedded hardware.
PRAESC relevance: End-to-end AI application deployment methodology from model to embedded target — now core to PRAESC's delivery model for industrial clients.
Open platform shifting AI development from cloud-centric to edge-device-centric, enabling European developers to build and deploy AI on embedded hardware through a collaborative marketplace.
PRAESC relevance: Neural architecture search and hardware-aware model optimisation techniques that underpin PRAESC's model compression and edge deployment work.
Matchmaking platform helping non-technical SMEs discover and access AI tools, experts, and compute resources through the European AI-on-Demand Platform.
PRAESC relevance: Deep understanding of SME AI adoption barriers — informing how PRAESC scopes and stages AI integration for industrial clients with limited AI maturity.
Coming up
Submitted proposals at the frontier of Physical AI and agentic systems — awaiting evaluation.
22-partner consortium addressing the core gap in today's AI deployments: agents that are isolated, stateless, and unaligned. CGA introduces Coherent GenAgentic AI — agents that are autonomous, stateful, aligned, and verifiably consistent across time and domains. PRAESC will lead the Coherent GenAgentic Mechanisms & Technologies and build the temporal consistency, memory, and closed-loop execution stack that feeds directly into PRAESC's industrial AI and robotics work.
A compact VLA combined with a physics-grounded two-tier Bayesian confidence loop, deployed entirely offline on Jetson Orin NX 16 GB. Targets autonomous valve, switch, and cabinet manipulation on offshore process platforms at SINTEF Trondheim — eliminating human exposure to hazardous in-zone tasks.
26-partner consortium building a federated, self-orchestrating AI service architecture where autonomous agents are composed, governed, and traded across organisations — with trust-by-design compliance (EU AI Act, GDPR, auditability) embedded at every layer. PRAESC will lead to the cognitive intelligence foundations: goal-driven reasoning, continual learning, and adaptive self-optimisation across the edge-cloud continuum.
Validating the concept that industrial humanoid robots can be programmed, taught, and operated entirely on-premise — with process knowledge and training data never leaving the factory. Partner: Swiss Smart Factory, Biel. PRAESC contributes the Physical AI architecture: an uncertainty-aware, edge-native agentic core combining a compact Vision-Language-Action model, Bayesian confidence loops, and operator-teachable skill creation.
Selected publications
- IWAI 2025 A Hardware-oriented Approach for Efficient Active Inference Computation and Deployment. [arXiv]
- HIPEAC 2025 Towards smart and adaptive agents for active sensing on edge devices. [arXiv]
- JEI 2024 Generic neural architecture search toolkit for efficient and real-world deployment of visual inspection CNNs in industry.
- IEEE 2025 Navigation under uncertainty: Trajectory prediction and occlusion reasoning with switching dynamical systems. [arXiv]