JARVIS AI Agent Designs and Tests Circuit Boards

AI circuit board design agent JARVIS with robot arm and PCB printer

Researchers have built an AI agent that designs, fabricates and tests a circuit board by itself, starting from a single text instruction. The agent, called JARVIS, built a 60 Hz notch filter in 2 hours and 39 minutes with no human help and measured 50.6 dB of attenuation at 60 Hz. It is a working demonstration of an AI circuit board design loop that runs on ordinary desktop equipment.

From one instruction to a tested board

Developing new power electronics still depends on people to design, build, test and iterate. The team behind JARVIS, an embodied AI agent, set out to close that loop for circuit boards. Power electronics matters for the energy transition, which is why the researchers chose it.

The process starts with one line of text. From there JARVIS handles everything else.

  • It scans its workspace with a 3D scanner mounted on a robot arm.
  • It builds an executable digital twin of that workspace.
  • It rehearses robot motions, fabrication steps and circuit behavior in the twin before using any materials.
  • It runs the physical build.

The hardware it controls

JARVIS works with unmodified desktop equipment. It coordinates a robot arm, a PCB printer that deposits conductive silver ink and solder paste, and a pick-and-place machine. The agent keeps track of what each machine can and can’t do, and it checks its progress with cameras and sensors as the board takes shape.

Testing is part of the same loop. Once the circuit is built, JARVIS measures it with an OPAL-RT real-time simulator and an oscilloscope. If the readings differ from the design, it diagnoses the gap and writes a revised design in KiCad, the open-source circuit design tool, then rehearses that version again.

The notch filter demonstration

The test case was a 60 Hz notch filter for grid-connected converter sensing. Such a filter removes interference at the grid frequency from a signal. JARVIS finished the first-run experiment in 2 hours and 39 minutes and reached 50.6 dB of attenuation at 60 Hz.

The digital twin earned its keep. Rehearsing before fabrication, the agent found five issues, including a test criterion that would have rejected a board that actually worked. Fixing that in simulation saved material and avoided a false failure on the bench.

The agent also dealt with faults that came up during execution, without asking a person to step in.

Why the digital twin matters

Physical experiments cost time and materials. Silver ink, solder paste and components are consumed with every attempt, and a robot arm that misjudges a movement can waste a whole run. By practicing in a simulated copy of its own workspace first, JARVIS catches many mistakes at no material cost.

It also keeps the agent grounded in what its specific machines can do. A general design tool doesn’t know the limits of one particular printer or robot. JARVIS holds those constraints as part of its own knowledge of the lab.

What the claim covers

The authors say JARVIS is, to their knowledge, the first system in power electronics where an agent closes the full physical design, build, test and learn cycle for circuit boards. That is a claim from the researchers themselves, and it covers one demonstration on one kind of circuit.

A notch filter is a relatively simple circuit. Boards with many components, tighter tolerances or high-power parts would test the approach much harder. The results show an integrated workflow is possible. They don’t yet show how far it scales.

Where self-driving labs could go from here

The setup points toward labs where an engineer states a goal and an agent iterates on real hardware overnight. For power electronics teams, that could shorten the time between an idea and a measured prototype.

The next questions are practical. Can the same system handle more complex boards, and how often does it need help outside a clean demo? Answers to those will show whether agent-run hardware development stays a research result or becomes a normal part of electronics work.

Source: arXiv.org

Written by

Saksham Bhargava

Saksham Bhargava runs Alteroid and has been writing about consumer technology for over five years, covering iPhone, Android, Windows, AI tools and gaming. With an MCA and six years of hands-on web development experience, he tests the features he writes about on his own devices and focuses on clear, practical guides that solve real problems. Away from the screen, he enjoys travelling to places most tourists skip.