The first European foundation model for safe, adaptable and explainable industrial robotics

GRAIL project – Generative Robotics & AI for EU Industrial Leadership – is a Horizon Europe project developing a new generation foundational model of AI technologies for robotics in manufacturing.

Its goal is to transfer advances in generative AI to the physical world, enabling robots to perceive, reason, learn new skills, adapt to changing situations and act safely in real environments.

At the core of the project is the GRAIL Foundational Robotics Model, a modular and agentic system designed to unify robot planning, grounding and control. Instead of developing a black-box model for a single robot or task, GRAIL builds a transferable technological foundation that can support different robotic embodiments and industrial applications.

A modular AI system that helps robots plan tasks, understand their surroundings and control their movements

GRAIL Foundational Model is based on three connected agents that work together through a shared view of the situation.

This helps robots adapt to different tasks, explain their actions, recover from problems and operate safely across different robots and industrial applications.

€40M

project budget

5

years
(60 months)

June 2026
to May 2031

5

industrial applications

across strategic manufacturing sectors
+10

robot embodiments

for cross-platform validations
+25
validation setups
3
Open Calls

funding +40
third-party projects

Project phases

A 5-year roadmap from lab prototypes to industrial pilots

GRAIL project follows an agile methodology based on short learning cycles and continuous evaluation. The project advances through three phases, each consolidating previous results and adding new capabilities, supporting the progression from TRL2 laboratory prototypes to TRL6 industrial pilots.

Phase 1

Contact GRAIL

Genesis

June 2026 – January 2028

TRL3 → TRL4

Genesis builds the first version of the GRAIL AI solution. This phase focuses on developing the initial model, creating synthetic data, testing the technology in the lab and setting up the project’s governance framework.

Phase 2

Ascent

February 2028 – September 2029

TRL3 → TRL5

Ascent improves the reliability of the model and prepares it for industrial testing. This phase will develop a more advanced version of the system, add safety mechanisms, protect sensitive data and test the technology in controlled industrial environments.

Phase 3

Summit

October 2029 – May 2031

TRL5 → TRL6

Summit brings the GRAIL model closer to real industrial use. This phase will focus on explainability, human supervision, large-scale industrial validation and pathways for future adoption by companies.

Pillars

Eight pillars for real-world robot intelligence

GRAIL project is structured around eight interconnected pillars that link Generative AI, robotics and society into one coherent framework.

These pillars cover the full path from data-efficient foundation models and world models to planning, grounding, control, human-centric adaptation, safety and cybersecurity.

Pillar 1

Data-efficient, foundational models

Defines the GRAIL architecture, including the agentic model, shared situational model and collaboration and training protocols.

Pillar 2

Real-world data collection

Provides multimodal industrial data to train and validate the model across tasks, sectors and robotic platforms.

Pillar 3

Simulation & predictive world models

Develops simulation, augmentation and predictive rollouts to improve learning, safety and transfer from virtual to real environments.

Pillar 4

Autonomous decision-making

Supports high-level reasoning, task planning and decision-making. Implements the Abstract Planning Agent.

Pillar 5

Embodiment and context grounding

Connects abstract plans with the physical world, adapting them to perception, embodiment and context. Implements the Situational Grounding Agent.

Pillar 6

Physical world interaction

Executes grounded plans through real-time control, reflexive policies and safe interaction with the environment. Implements the Physical Control Agent.

Pillar 7

Human-centric adaptation

Ensures that the system remains understandable, usable and aligned with human supervision, interaction and trust, both during the training phase and the deployment phase.

Pillar 8

Reliability, safety and cybersecurity

Defines the operational envelope for reliable, secure and certifiable deployment in industrial environments.

Get updated on GRAIL news

Subscribe to the GRAIL newsletter to receive updates on project events, milestones and advancements

Subscribe to updates