DeepRAP: Deep Reasoning, Abstraction & Planning towards trustworthy Cognitive AI Systems

¿De qué trata esta convocatoria?

The challenge funds research on AI systems that reason, abstract and plan beyond today's pattern-matching models while remaining trustworthy. Projects aim to reach technology readiness level 4.

¿Qué financia esta convocatoria?

Scope

Current generative AI recognises patterns well but struggles with complex reasoning, understanding the real world, logic tasks and long-term planning, even when correct solutions exist. The challenge aims to go beyond current symbolic and neural approaches, including deep learning, large language models and reinforcement learning, and significantly improve the reasoning, abstraction and planning capabilities of AI systems. Proposals should explore new approaches, including combinations such as neuro-symbolic AI or entirely new frameworks inspired by fields such as neuroscience, biology, physics and philosophy.

Cognitive capabilities

Proposals should address one or more of:

• Deep reasoning: causal inference, logical reasoning and context-aware or commonsense decisions in complex, unstructured settings, with transparent and explainable rationales.
• Deep abstraction: generalising from limited data by forming and refining high-level concepts, analogies and internal world models that transfer across domains.
• Deep planning: robust, adaptive and scalable planning in open-world or uncertain real-time environments, including hierarchical planning, contingency planning and continual re-planning.

Expected outcomes

• Models or architectures that handle multimodal data, knowledge and uncertainty and can run with constrained computing resources.
• Provable trustworthiness mechanisms covering explainability, transparency, fairness, risk evaluation, security and alignment with ethical and legal standards, including fundamental rights and the EU AI Act.
• Capabilities integrated in a cognitive AI system at technology readiness level 4, performing complex real-world tasks such as scientific discovery, decision support and problem solving.
• New methods and metrics for evaluating and certifying reasoning, trustworthiness and computing resource use; FAIR data, models and results; and synergies with AI testing and experimentation facilities, EBRAINS, the Resource for AI Science in Europe, the AI-on-Demand Platform and the Quantum Flagship.

Portfolio approach

The portfolio should cover reasoning, abstraction and planning; a variety of technological approaches including neuro-symbolic AI, deep learning and reinforcement learning; and varied domains such as industry, mobility, civil security, scientific discovery, health, cybersecurity, justice and human-robot interaction. Selected projects take part in portfolio activities on interoperability, a shared benchmark and open evaluation platform, common pilots, multi-agent integration where feasible, application shaping, and ethical and societal alignment.

¿Cuándo termina el plazo?

28 de octubre de 2026

28 October 2026. This is a single-stage call.