A New Wave in Battery AI — Foundation Models and Agentic Workflows

As the EV and energy storage markets expand rapidly, the complexity of battery management is accelerating at an unprecedented pace. Vehicle development cycles have shortened to as little as 24 months, while battery chemistries continue to diversify — LFP, NMC, sodium-ion, solid-state, and the list keeps growing. Running months of characterization tests from scratch every time a new cell appears — this conventional approach can no longer keep up with the speed of development.

Eatron Technologies (UK), for whom we serve as a tech partner handling all aspects of business development in Japan, presented a new approach to this challenge at the SDV Automotive Expo Tokyo, held in February 2026. In a talk titled "Battery AI — Foundation Models and Agentic Workflows," the company's Dr. Ugur Yavas took the stage to share the cutting edge of battery AI.

Foundation Models: From Specific Solutions to General Solutions

Conventional battery AI has been "task-specific." A dedicated model was built for SOC (State of Charge) estimation, another for SOH (State of Health), each requiring its own data collection and training cycle every time a new cell or chemistry appeared.

The Battery Foundation Model that Eatron is developing fundamentally changes this paradigm. It is a pre-trained AI model that learns universal electrochemical patterns from large-scale, diverse battery data spanning cells, packs, chemistries, vehicles, and operating conditions. In essence, it is a general-purpose intelligence platform that has learned the "laws of physics" of batteries from data.

The true value of this approach lies in "zero-shot estimation" for unseen cells. During the presentation, results were shown where the pre-trained model was applied to LFP (lithium iron phosphate) cells it had never encountered before, performing SOC estimation without any additional tuning. Tested across diverse conditions — including constant-current capacity mapping, various energy storage system operating patterns, and real-world EV driving cycles — the model maintained consistent performance across multiple temperature ranges and even end-of-life cells.

What is particularly noteworthy is that the model corrects SOC based on physically consistent voltage behavior rather than heuristics. In an experiment where a deliberate 5% SOC initialization error was introduced, the model self-corrected over time and converged on the true SOC trajectory. In other words, rather than being a black box, it performs estimation with a genuine understanding of the battery's physical behavior.

Even when improved accuracy is needed under specific conditions such as cold environments, rapid correction is possible with just a small amount of fine-tuning data. This is where the potential lies to compress months of cell characterization testing into just days.

Extending to Battery Degradation Prediction (RUL)

The same foundation model approach has been extended to battery Remaining Useful Life (RUL) prediction. Conventional RUL prediction relied on limited lab test data or simple extrapolation of current SOH trajectories, but in real-world operating environments, this alone is insufficient for warranty management.

Eatron's AI-RUL model has learned degradation behavior that generalizes across operating conditions, uncertainties, and domain shifts. The presentation showcased degradation simulations under varying conditions of temperature, C-rate, and duty cycle, demonstrating that the model can visualize in advance how much operational cost each option would entail in response to "what if?" questions.

Agentic Workflows: From Numerical Output to Decision Support

The latter half of the presentation addressed an even more forward-looking vision. Conventional battery management systems stop at outputting SOC, SOH, and RUL values, but Eatron's BOS (Battery Optimization Suite) Agent goes further — into "action."

The BOS Agent employs a multi-agent workflow. Triggered by structured events from the BMS or cloud — such as imbalance alerts, anomalous trends, or threshold exceedances — a Root Cause Analysis (RCA) agent identifies context from battery specifications and historical operating data. Rather than issuing a single rule-based response, it evaluates multiple balancing strategies by referencing manuals, constraints, and past outcomes. Finally, a planning agent converts the selected strategy into an actionable intervention plan that considers timeframes, SOC targets, and operational costs.

Reasoning through "when, how, and how much each option costs" to make decisions — this is an industry-leading initiative that applies agentic AI specifically to battery management.

What Sets Eatron Apart from Other Battery AI Companies

While multiple players exist in the battery AI market, Eatron's approach stands out for reaching "general solution" level rather than remaining at the "specific solution" level. Instead of building individual models for each task, they adapt from a foundation model that has learned universal physical patterns of batteries, requiring minimal additional training for each task. This technical depth has been demonstrated through zero-shot application to unseen cells and physically consistent estimation.

Additionally, the integration of agentic workflows that autonomously convert AI model outputs into action plans sets the company apart as more than just a data analytics tool.

Eatron currently monitors over 30,000 batteries worldwide and counts global companies such as Oshkosh, LG, ABB, VinFast, and Knorr-Bremse among its partners.

Envital's mission is "Bringing edge-cloud intelligence to the real world." Eatron's BOS integrates real-time inference at the edge with learning and optimization in the cloud, and further realizes autonomous decision-making through agentic workflows. It is a concrete implementation of our vision.

We will continue to serve as a bridge connecting overseas companies with cutting-edge edge-cloud technologies, including battery AI, to the Japanese market, driving business development in the mobility and energy sectors.

If you are interested in Eatron's technology or its business development in Japan, please do not hesitate to contact us.