ESAB introduces AI cobot for adaptive MIG and TIG welding

Key Takeaways

• ROBBI AI is ES AB’s first collaborative robot that embeds artificial‑intelligence directly in the welding arc.
• The system combines a 6‑axis, 30 kg payload cobot with a 2 MP vision camera and a 1 µm‑resolution laser scanner to sense joint geometry in real time.
• Unlike traditional robotic welders that follow a pre‑programmed path, ROBBI AI creates its own weld trajectory and continuously adjusts current, voltage, travel speed, and oscillation to compensate for gaps, mis‑alignments, or material variations.
• The AI learns from every weld, storing video and process data for defect analysis and continuous improvement.
• Users can switch to “manual‑cobot” mode, disabling AI‑driven decisions while retaining the robot’s repeatability.

Introduction: A New Breed of Welding Cobot

On 22 Oct 2023, ES AB announced ROBBI AI, a collaborative robot that fuses AI, vision, and laser‑based sensing to deliver adaptive MIG and TIG welding. Targeted at low‑volume, high‑mix production—such as pipe spools, structural frames, and custom‑fit joints—the system promises to automate tasks that have traditionally required skilled human welders.

How ROBBI AI Works

Integrated Sensing Suite

The camera and laser continuously feed data to the on‑board AI engine. Before each run, operators input the welding procedure specification (WPS)—including consumable type, base material (e.g., 304 SS, 6061‑Al), and desired weld class. The AI then maps the joint, predicts the optimal path, and begins welding.

Real‑Time Adaptation

During the arc, ROBBI AI monitors:

• Joint gap changes (±0.5 mm tolerance)
• Oscillation amplitude (adjustable 0‑3 mm)
• Travel speed (range 10‑200 mm/min)
• Current/voltage (dynamic scaling ±15 % of setpoint)

If the gap widens, the system slows travel and increases filler feed; if it closes, the robot shortens the torch‑to‑work distance and reduces current to avoid burn‑through. All adjustments occur within the pre‑approved WPS envelope, eliminating the need for re‑programming.

Dual‑Mode Operation

Robots can run in Full‑AI mode (auto‑path generation) or Cobot‑Only mode, where the operator commands the arm while the AI remains dormant. This flexibility lets shops transition gradually, preserving existing programming investments.

Learning Loop: From Data to Defect Prevention

Before field deployment, each ROBBI AI unit undergoes a simulation‑based training cycle using virtual joint models. Once installed, it records:

• Video of each weld (30 fps, 1080p)
• Process parameters (current, voltage, speed)
• Laser‑derived gap measurements

If a weld deviates from quality thresholds (e.g., excessive spatter, lack of fusion), the AI halts the arc and flags the event. Operators review the footage, annotate the cause, and feed the annotation back into the learning algorithm. Over time, the system builds a defect‑recognition library, reducing false‑positive stops by up to 35 % after the first 200 welds (internal ES AB data).

Comparison: ROBBI AI vs. Conventional Robotic Welding

*ROI estimates based on ES AB case studies of 3‑axis MIG cells (average 150 kW power) and 6‑axis TIG cells (average 80 kW power).

Real‑World Impact

Early adopters report a 30 % reduction in setup time for mixed‑size pipe runs and a 20 % increase in first‑pass quality for TIG welds on aluminum alloys. The AI’s ability to stop a weld before a defect propagates also cuts re‑work costs by an estimated $45 k per year for a typical 1 M‑weld shop.

Bottom Line

ES AB’s ROBBI AI brings “physical AI” to the shop floor, merging vision‑guided perception with real‑time process control. By generating its own weld path, continuously adjusting parameters, and learning from each arc, the system bridges the gap between flexible, low‑volume fabrication