ATOM · Physical AI Ops platform
ATOM turns a plain-language description of your workcell into a physics-accurate NVIDIA Isaac Sim model, then generates controller code from the same validated spec. Prove out the cell in hours, before anything is purchased.
What the numbers are
Hours-not-months is a target. A documented cell with baseline hours against ATOM hours is still to come. The other two figures are from the simulations below.
How ATOM works
Most of the cost of a robot cell is the months of engineering before it runs. ATOM compresses that front end.
Write the application in plain language: parts, fixtures, conveyor, robot and task. ATOM turns it into a structured workcell specification.
A deterministic assembler builds the Isaac Sim model from real, dimensionally correct equipment models.
Check reach, collisions and cycle time in simulation before anything is bought.
Produce controller motion and logic from the same validated spec, so the cell you approved is the cell you get.
The language model interprets what you asked for. It never generates geometry, coordinates or motion. Every dimension comes from real asset data, so there are no invented reach envelopes.
It does not replace a controls engineer or a safety assessment. It is not an autonomous robot brain: the output is a conventional, readable robot program. It is not a runtime, because once the code is on the controller it is your controller's program.
Multi-agent orchestration
ATOM uses a multi-agent AI pipeline to turn your plain-language cell description into a structured simulation. Each agent handles one job — parsing the request, selecting equipment, laying out the cell, generating motion — so the system is auditable and deterministic where it matters.
Generates Isaac Sim scene code from the workcell specification. No manual scripting.
Optimises paths using cuMotion GPU planning. Measures cycle time and joint loads from the physics engine.
Produces vendor-specific robot programs (ABB RAPID, FANUC TP/Karel, KUKA KRL) via ROS2 Bridge from the same validated spec.
Simulation library
Each entry is a simulation run with its own video. New simulations are added here as they are completed.
Three M-710iC/50 robots on Electron risers, a stretch wrapper, pallet dispenser and forklift in one plant. Includes a robot swap to the R-2000iC/210F and a fault caught before build.
122 placements measured in the swap comparison
An ABB robot dispenses paste onto 32 pad targets on a circuit board, ahead of a reflow station. Four virtual cameras show the cell, the robot, the paste and the result.
32 pad targets per board
Next
Additional cells and applications will appear here, each with its video and the robots used.
Case study · Palletizing line
Everything below comes from one simulated palletizing line: three FANUC cells, a wrapper, a dispenser and a forklift.
Caught in simulation · Palletizing line
A three-robot FANUC M-710iC palletizing cell on Electron risers, running in ATOM. On pallet two, layer 3, one carton could not be placed.
Bench run, 2026-09-30. Every carton, robot and second is logged in a run record.
Swap a robot, keep the cell · Palletizing line
Same conveyor, same pallet pattern, two robots. ATOM ran a FANUC M-710iC/50 and a FANUC R-2000iC/210F through the cell and measured every placement, so the choice is made from data before anyone orders a robot.
| FANUC M-710iC/50 | FANUC R-2000iC/210F | |
|---|---|---|
| Electron riser height | 0.9 m | 0.6 m |
| Median cycle, carton to carton | 15.0 s | 13.8 s |
| Median placement offset from slot | 1.9 mm | 3.1 mm |
From the Round 2 video: 122 placements across two takes, medians only. Faster and less precise is a trade-off the layout has to accept or fix. A longer run will tighten these numbers.
The whole plant · Palletizing line
ATOM simulates the line around the robot too, so handoffs and queues show up in the results.
Every run is logged: each joint's angle and torque 20 times a second, every stop and fault coded, every machine's state.
Case study · Solder paste dispensing
An ABB IRB 1200 places solder paste on a circuit board that then moves to a reflow station. The same run is filmed from four virtual cameras: cell overview, robot side, paste microscope and paste inspection.
Uncalibrated simulation, dispensing excerpt. Paste shape and timing are not yet calibrated to a real process, and no process results are claimed.
Robots & platforms
| Robot brand | Controller language | Bridge |
|---|---|---|
| ABB | RAPID | ROS2 |
| FANUC | TP / Karel | ROS2 |
| KUKA | KRL | ROS2 |
| Universal Robots | URScript | ROS2 |
Controller code is generated via the ROS2 Bridge from the same validated simulation spec. Before launch, list only the brands where generated code runs end to end on request.
Isaac Sim, Isaac Lab, Replicator and cuMotion GPU motion planning, unified behind one pipeline. ATOM is an Isaac Sim extension, not an Isaac Sim competitor.
Manufacturing analytics are built in: cycle time, success and failure rates, reachability, singularity detection, layout optimization.
Who uses it
Automation as a service
Don't have an automation team? We design, simulate and deliver production-ready robot cells, lines and complete factories — from concept through commissioning.
Single-station workcells — pick-and-place, palletizing, dispensing, welding — designed in ATOM, validated in simulation, deployed on your floor.
Multi-cell lines with conveyors, buffers and coordinated motion, proven end-to-end in simulation before hardware is ordered.
Full-facility automation planning — robot placement, material flow, throughput modeling and safety zoning across your entire operation.
Every engagement starts with an ATOM simulation. You see the system running before a single dollar is committed to hardware.
Enterprise security
ATOM is designed for environments where workcell layouts, cycle times and process parameters are competitive advantages.
On-prem, hybrid or private cloud — deploy where your policies require.
Complete isolation by customer with no cross-tenant data access.
Fine-grained data-sharing rules for system integrators and partners on your terms.
No forced model training on your data. Your workcell designs stay yours.
Team
Decades of hardware development, industrial robotics integration and 35+ years of manufacturing experience meets AI.
Co-founder Breeze Test Instruments (sold to Tektronix). CEO & President, Axiom Electronics, Inc. 35+ years in electronic manufacturing and product development.
U of Colorado Denver · Tektronix
M.S. Mechatronics & Robotics Engineering. Robotics Researcher at AI4CE Labs, NYU.
NYU
MOTH CTO and serial entrepreneur. Building the first consumer quantum applications.
Cornell · Dartmouth · Yale
Careers
We are a small, technical team solving hard problems at the intersection of robotics, simulation and AI. If you want to ship real tools that put robots to work in factories, we want to hear from you.
Work on the ATOM transpiler — translating natural-language workcell descriptions into validated Isaac Sim scenes and deterministic robot programs.
Portland, OR · Remote considered
Own the cloud pipeline that runs physics-accurate simulation at scale — GPU orchestration, CI for robot code, and deployment tooling.
Portland, OR · Remote considered
Develop and validate robot motion planning, URDF/USD asset pipelines and sim-to-real workflows for industrial manipulators.
Portland, OR · Remote considered
To apply, email your résumé and a brief note on what interests you to robert@electron-robotics.ai.
Privacy Policy
Effective October 2026
When you contact us via email, phone or our website contact form, we collect the information you provide — typically your name, email address, company and the content of your message. We do not use tracking cookies or third-party analytics on this site beyond what SquareSpace provides as our hosting platform.
We use contact information solely to respond to your inquiry, schedule demonstrations and communicate about Electron Robotics products and services. We do not sell, rent or share your personal information with third parties for their marketing purposes.
Workcell designs, simulation configurations and robot programs created through ATOM belong to the customer. We do not use customer workcell data to train AI models unless a customer explicitly opts in. See our product terms for details.
This website is hosted by SquareSpace. Video content is embedded from YouTube. These services have their own privacy policies. We use NVIDIA Isaac Sim and Omniverse infrastructure — data processed through those platforms is subject to NVIDIA's terms.
You may request access to, correction of, or deletion of any personal information we hold about you by emailing robert@electron-robotics.ai.
Questions about this policy: robert@electron-robotics.ai
electr0n, Inc. · 2130 SW 5th Ave, Suite 101, Portland, OR 97201
Contact
We are running controlled early evaluations. Tell us the robot, the task and the layout, and we will show you what ATOM finds.
