Automate What
You Can See

Teach the cobot by hand, as you would any cobot. The camera memorises the taught path and finds it again on the next part — no 3D laser scanner, no robotics engineer.

Path adaptation . Real-time correction . Claude Code integration . One-day training

Rocket Welder Automation Platform — camera preview with vision overlay, robot status, PLC tags, and program workspaces on a single screen

The Camera Finds the Geometry.
Your Program Computes the Path.
The Robot Follows.

The camera is the sensor. The vision model extracts geometry from every frame. Your program turns that geometry into robot waypoints or real-time corrections. The robot executes. The camera sees the result. Loop repeats.

1

Camera

Camera frame capturing the welding process

Captures the process. Every frame.

Welding weld joint
Inspection part surface
Measurement component dimensions
2

Vision Model

Camera frame with keypoints and polygons overlaid by vision model

Extracts geometry of recognized objects — keypoints, edges, dimensions, polygons — as numbers your program can use.

Welding seam edges, gap 2.3mm
Inspection defect at 3 o'clock
Measurement width 45.2mm +/-0.1
3

Your Program

IDE screenshot with program code computing robot waypoints

Keypoints become 3D points in robot frame. You compute waypoints — or push real-time corrections.

Welding follow the real seam
Inspection trigger reject on defect
Measurement flag out-of-tolerance
4

Robot / PLC Executes

Robot and PLC executing computed path with live dashboard

Robot follows the computed path — or PLC triggers based on what the vision model measured.

Welding robot corrects 1.7mm
Inspection PLC diverts to reject
Measurement log result, stop line if OOT

Two Ways to Close the Loop

Path Adaptation

Teach -> Find -> Recompute

You teach the path by hand, as with any cobot. The camera memorises how the taught path looks. On the next part the cobot makes the same sweep, the camera finds the taught points, and the program is recomputed.

Use case: 2–5 mm scatter from part to part within a series, on bigger parts — without a 3D laser scanner.

Real-time correction

Real-Time Correction

The robot is already welding; the camera keeps watching and your program corrects the torch 10 to 25 times a second.

Use case: the part distorts under heat, or moves during welding.

Don't Take Our Word for It

Real welding, real solutions. We can't name every customer, but we can show how it could work on your materials. Send us your parts — we can schedule a demo on-site or remote.

Two sets — two computers

Both sets use the Rocket Retina camera and a monitor at the station. The computer decides what the set can do: the Rocket Neuron runs path adaptation and, as an option, real-time correction; the camera controller shows and records the weld.

Rocket Neuron — the AI computer of the RocketWelder Set

Rocket Neuron — the RocketWelder Set

$2,399

The AI computer of the RocketWelder Set. It runs path adaptation on every part and, as an option, real-time correction while the torch welds.

Cameras 1-2Correction 10-25 HzAI (TOPS) 100+Mount DIN/19-inch rack
  • ✓ 1-2 camera inputs (GigE Vision)
  • ✓ Path adaptation
  • ✓ Real-time correction
  • ✓ Claude Code integration (MCP connector)
  • ✓ Cobot + robot controller integration
Contact sales
Camera controller — the Intel-based mini PC of the vision-only set

Camera controller — the vision-only set

$1,899

An Intel-based mini PC. It runs the live preview and recording of the Rocket Retina camera and drives the monitor. No AI: no path adaptation, no real-time correction.

  • ✓ Live preview on the monitor or in a browser
  • ✓ Recording
  • ✓ No AI computer
Contact sales

Need the camera, the controller, a screen at the station and commissioning?

See the Rocket Retina lines and prices →

Runs on Neuron — at the Machine

Compact. DIN-rail mountable. Air-gapped capable.

Neuron device, DIN-rail mounted, cables connected

One box. Plug in the camera, connect the robot, wire the PLC. Everything runs here — the vision processing, your program, the robot commands. No cloud, no server room, no IT department.

✓

Camera, robot, PLC — plug in via Ethernet. GigE Vision, ABB EGM, Fairino, Modbus TCP.

✓

Vision model on-device — GPU always included. NVIDIA for high-performance inference at 25+ FPS. No cloud round-trip.

✓

Works air-gapped — no internet required for operation. Connect when you need it — remote support same day, over-the-air updates.

✓

Open a browser — access the full platform from any device on the network. Program, monitor, record.

Works With

ABB IRB robot arm

ABB

EGM, 250 Hz feedback

Now
Fairino FR5 cobot

Fairino

Direct TCP control

Now
Universal Robots cobot

Universal Robots

URScript integration

Coming soon
KUKA robot

KUKA

RSI interface

Coming soon

for any PLC . Any GigE Vision / USB camera . Rocket Retina welding camera

Price your welding set

Choose the camera, the arm and the power source, and get a specification and a net price without talking to sales. The configuration is a link you can save and send back to us.

Net prices from the published price list. Items we cannot price yet are named, not hidden.

Configure a welding set →

Frequently Asked Questions

Answers for the welding engineer who will teach the cobot.

No. Welding is our first and deepest vertical. But the platform is general-purpose. Any workflow that involves a camera, a vision model, and industrial hardware (robot, PLC, or both) can run on Rocket Welder. Quality inspection, measurement, assembly verification — the same platform, same tools.
A compact edge computer with GPU that runs the Rocket Welder platform. Mounts on DIN rail at the machine. Industrial-grade, air-gapped capable. NVIDIA GPU available for high-performance vision inference at 25+ FPS. Camera input, vision processing, your program, robot and PLC control — all run on this one device.
ABB IRB series via EGM (250 Hz feedback) and Fairino cobots (direct TCP control). Both accessed through a unified IRobot interface — swap robot brands without rewriting your program. Universal Robots and KUKA support is planned.
No. You teach the cobot by hand, as you would any cobot, and you need to know your welding process. For the program logic you work with an assistant: Claude, connected through our MCP connector. It writes programs, edits them and generally helps. The programs are C# underneath, and you can read and change them. Operator training takes one day.
You teach the path by hand. On each new part the camera finds the taught points; with a laser distance sensor our SDK returns them directly in robot coordinates — no manual calibration math. The program is recomputed from those points: that is path adaptation. As an option, real-time correction runs the vision model and your program on every frame while the robot moves and pushes corrections continuously.
Yes — at runtime. Everything runs on the Rocket Neuron at the station: vision processing, recording, robot and PLC control, all on-premises, no cloud required. The only piece that needs an internet connection is the assistant: Claude works online. Once your program is deployed, the shop-floor stack runs fully offline. Updates can be pushed via USB or local network.
Claude is the assistant; we provide the MCP connector that links it to your station. Through it Claude lists, reads and writes the station's programs, and reads the robot's position and joints, the camera image, the connected devices and the PLC tags. You ask; it writes, edits and helps. It never runs the robot — that is always you, at the station.
One day. We train your operators at your station, as part of installation and commissioning.