Battery innovation is constrained not only by the size of the chemical design space, but also by the complex and often poorly connected chain linking material composition, synthesis history, electrode manufacturing, microstructure, interfaces and cell performance. Overcoming this fragmentation requires more than faster individual experiments or simulations: it requires an integrated discovery system capable of reasoning across scales, selecting the next most informative action and translating target performance into experimentally realizable materials and processes.

This contribution presents the emerging FULL-MAP concept for an AI-driven, chemistry-agnostic materials acceleration platform for sustainable batteries. The framework connects high-throughput synthesis and characterization with multiscale and multiphysics modelling, manufacturing simulation, cell-level performance prediction and sustainability assessment. Agentic AI provides the orchestration layer: it decomposes design objectives into linked computational and experimental tasks, selects suitable models and data sources, coordinates autonomous workflows, evaluates uncertainty and provenance, and iteratively proposes the next simulations or experiments. In this way, the platform advances from conventional forward prediction towards multi-objective inverse design—identifying combinations of chemistry, synthesis route, processing conditions and electrode architecture that can deliver required performance, lifetime, manufacturability and sustainability.

A first methodological frontier is the use of machine-learned interatomic potentials (MLIPs) to model synthesis-relevant phenomena at atomistic resolution and at time and length scales inaccessible to conventional first-principles calculations. Beyond predicting equilibrium structures and properties, emerging reactive, charge-aware and chemically transferable MLIPs can address nucleation, phase formation, surface and interphase evolution, defect generation and temperature- or environment-dependent transformations. They also provide physically grounded descriptors and parameter surfaces for coarse-grained mesoscale models, enabling information to propagate from atomistic synthesis mechanisms towards microstructure and electrode behaviour. A second frontier concerns electrode processing. Particle-based and other microstructure-resolved methods can represent slurry constituents, aggregation, drying, binder migration, compaction and particle rearrangement, while continuum manufacturing and electrochemical models connect the resulting structure to transport and performance. Because direct high-fidelity simulation is computationally demanding, AI surrogates and multifidelity learning are required to explore processing windows, quantify trade-offs and enable rapid inverse mapping from a desired electrode structure or performance to feasible manufacturing parameters.

Together, these developments redefine modelling as an active component of autonomous discovery. By coupling novel physics-based and ML-assisted models with agentic decision-making and automated laboratories, FULL-MAP aims to establish closed-loop workflows spanning materials discovery, synthesis, manufacturing and validation—and thereby open new routes for faster, more systematic and more sustainable battery innovation.