EIC Pathfinder

EIC Pathfinder Challenge: DeepRAP — Deep Reasoning, Abstraction and Planning towards trustworthy Cognitive AI

Supports cognitive AI development focused on trustworthy deep reasoning, abstraction, and planning capabilities.

Eligibility · EU member states + Horizon-associated

Generative AI can recognise patterns and generate accurate outputs but frequently fails on complex reasoning, long-term planning, and logic tasks where provably correct solutions exist. This Challenge funds approaches that go beyond traditional deep learning and reinforcement learning — including neuro-symbolic combinations and entirely new frameworks inspired by neuroscience, biology, physics, philosophy or other disciplines.

Proposals address one or more of three cognitive capabilities:

1. **Deep Reasoning** — causal inference and logical reasoning in unstructured environments, with transparent and explainable rationales, aligned with human values.

2. **Deep Abstraction** — generalising insights from limited data via high-level concepts, analogies, and internal world models that transfer across domains.

3. **Deep Planning** — robust adaptive planning in open-world agentic environments, including hierarchical planning across temporal levels, contingency planning, and continual re-planning.

Must reach TRL 4 with a cognitive AI system performing complex real-world tasks (scientific discovery, decision support, problem-solving) plus simulations at scale. Must follow FAIR principles and align with the EU AI Act.