Local intelligence for the batteries behind drones, mobile robots and field power. Understand what a battery can still store and how it responds under load.
From a signal to a decisionStarting with laboratory cells. Building toward military maintenance.
A full charge tells you little about how a battery has aged. Njal reads its electrical response to estimate what it can still store and how it handles demand.
For a drone, battery health matters to endurance and available power. For a maintainer, it informs what to inspect or replace. We are building that intelligence into the tools that keep military equipment running.
Watch the product film
Loading a real laboratory recording…
One shared model core · Recorded case demo
A controlled electrical probe reveals how a cell responds. One adapted foundation-model core turns that signal into three connected estimates of battery health.
Storage capacity. Recovery after demand. Resistance under load. The diagnostic engine runs locally, without cloud inference.
After a burst of power, how does the battery recover? Njal uses a lower-current recording to estimate its voltage response after a separate, heavier load.
Compare the prediction with an independent bench measurement. The model never sees that answer during inference.
Estimate capacity relative to the battery’s rating. A full charge indicator can conceal a battery that stores less than it used to.
Loading recorded capacity estimate…
The same recorded inputs produce a capacity estimate, checked against an independent discharge test. This secondary comparison remains statistically inconclusive.
A battery can hold charge and still struggle to supply power. Njal estimates resistance under heavier load, which determines how far voltage drops when current rises.
Capacity tells us about charge storage. Resistance and recovery describe the response to demand. One shared core estimates all three.
58% lower recovery error. 32% lower resistance error. Against conventional models selected before testing, on 15 battery units withheld from development.
Four commercial battery families. 920 paired laboratory tests. These gains measure diagnostic accuracy; equipment and field validation are the next step.
A separate 81-cell study also confirmed recovery and resistance gains against its selected conventional comparator. Both studies are detailed in Technology.
Explore the technology58% lower error
32% lower error
Capacity is the third output from the same core. Its benchmark comparison is secondary and remains inconclusive.
Explore saved model outputs for each laboratory recording, with its signal and all three estimates in the demonstration above.
Public NREL laboratory recordings, CC BY 4.0. TabICLv2 pretrained core, adapted by Njal. Equipment scenes illustrate the intended application; the measurements come from commercial cells and modules.
Our goal is to put this diagnostic core into chargers, test benches and maintenance tools for drones, mobile robots and field power. Equipment makers supply the measurements. Njal turns them into a clearer picture of equipment health.
The first integration starts with battery teams and equipment makers: evaluate their recordings, compare against bench tests, then measure the value in a real maintenance workflow.
Explore applications across air and seaBuilding equipment, maintaining a fleet, or exploring a partnership? Tell us what you want to understand. We’ll review your request and arrange a demonstration.
Each request is reviewed personally. See the working diagnostic engine, its measured results, and how we would evaluate your equipment.
rbokhari@usc.edu