Application Guide: How ESAB’s ROBBI AI Learns During Welding and What Shop-Floor Teams Should Test

Welding automation is usually judged by how well it handles variation, not by how neatly it repeats a fixed path. The Oct. 22 ESAB launch of ROBBI AI presents that challenge as an application problem: a cobot that uses camera and laser sensing, plus artificial intelligence, to revise motion and welding parameters while a joint is being worked. Its stated fit is MIG and TIG jobs, metal fabrication with changing fit-up, short runs with many part variations, pipe fabrication, and work that has been difficult to automate.

The practical appeal is not that a robot moves faster, but that the decision space during welding becomes responsive to the physical workpiece. The application enters where planned paths and real joints diverge.

Where the tool enters the process

Rather than depending on predefined paths and fixed weld sequences, ROBBI AI creates its own weld path and decides what to do as welding proceeds. Before the process begins, operators supply welding procedure parameters, including consumables, material type and weld type. The robot then uses sensing to evaluate the workpiece, identify the expected joint and decide how to proceed within parameters set for the application.

“Physical AI brings intelligence directly into the welding process,” said Dan Colvin, vice president of Advanced Technology Solutions at ESAB. “Instead of programming every movement and anticipating every possible variation, ROBBI AI assesses the joint, determines how to proceed and can make adjustments while welding. It is also designed to learn from every weld, helping it recognize conditions that can lead to defects and respond by adjusting welding parameters or stopping the weld when needed.”

For an application team, this creates a useful boundary: the procedure defines the allowed envelope, while the system adjusts motion and parameters within it.

Adapting while welding

“As welding progresses, ROBBI AI can adapt its motion and welding parameters to changing conditions,” Colvin said.

“For example, if a joint gap opens or closes, the system can adjust its travel motion, oscillation or welding parameters as appropriate within the established procedure. Rather than relying solely on a predetermined weld sequence, it can make welding decisions in real time.”

This is the clearest application signal in the release: the system is positioned not merely as an executor of a saved program, but as a sensor-driven participant in a variable physical task.

Feedback loop after each weld

Before installation, ROBBI AI completes a simulation-based learning period. After it is placed in a customer’s facility, it keeps collecting weld data and associated video that operators can review to investigate defects and find improvement opportunities.

When a condition that could compromise a weld is detected, the system can halt welding. Operators then review the recorded information and give feedback that helps the AI recognize similar conditions in future work, ESAB said.

For users who prefer traditional operation, the platform also supports conventional cobot use; full-AI mode can be disabled.

That optional mode matters for adoption. It lets a shop evaluate AI behavior without making the entire production line depend on it from the first day.

Data location and procedure guardrails

Artificial-intelligence processing occurs locally on a computer inside the customer’s facility, avoiding reliance on continuous communication with a cloud-based system. ESAB said this keeps learning data onsite and supports real-time decision-making.

“We use our existing InduSuite software to help keep AI-driven welding decisions within predefined welding procedure specifications and qualification requirements,” Colvin said. “ESAB’s approach is designed to reduce the risk of AI-driven adjustments that could result in welding outside those specifications or with an unapproved procedure.”

In an industrial setting, this is where application usefulness is measured: not by autonomous action alone, but by whether autonomous action remains inside certified process limits.

Hardware and integration options

The ESAB cobot system can be configured with a Universal Robots UR10e arm or a UR20e arm. A seventh-axis configuration is available to extend reach and support a larger work envelope; a wheeled cart is also offered so it can be moved around a shop.

Compatible welding power sources include EWM Titan XQ, EWM Tetrix XQ and ESAB Aristo Edge 500R.

These choices determine whether the system can be placed around existing assets or requires additional reach, mobility, and power-source compatibility.

Configuration item	Option
Arm	Universal Robots UR10e or UR20e
Reach extension	seventh-axis configuration
Mobility	wheeled cart
Power source compatibility	EWM Titan XQ, EWM Tetrix XQ, ESAB Aristo Edge 500R

Why TIG and pipe fabrication are difficult automation cases

ESAB identifies TIG welding and pipe fabrication for data centers and other infrastructure projects as potential applications for ROBBI AI, especially where specialized welding skills are required and conventional automation can be difficult to implement.

TIG welding can be difficult to automate because it requires precise control of the torch, filler wire and molten weld pool. In research published in the International Journal of Pressure Vessels and Piping, investigators studying robotic orbital TIG welding of 304L stainless-steel pipes found that maintaining a stable weld pool during the root pass was particularly challenging as the torch moved around the pipe and the welding position changed.

A manual welder can continually adjust torch trajectory and filler-metal feed to control the pool; reproducing that control through automation requires carefully developed welding parameters. The study also found that the process remained sensitive to geometric variations in pipe joints.

Those are exactly the conditions the application narrative addresses: a moving joint, a changing position, and a weld pool that cannot be treated as a fixed recipe.

Checklist for teams evaluating the application

• Is the task one of the stated target scenarios: variable joint fit-up, low-volume high-mix production, pipe fabrication, TIG welding, or another MIG or TIG task that has been hard to automate?
• Can operators provide the required procedure inputs before welding begins, including consumables, material type and weld type?
• Can the shop review weld data and video to investigate defects and provide feedback after a detected condition?
• Are the local processing expectations clear: AI occurs onsite, reducing reliance on continuous cloud communication?
• Is a conventional cobot mode available if full-AI mode needs to be disabled?

Step-by-step workflow to understand

1. Begin with the established welding procedure, not with a blank autonomous task. Provide consumables, material type and weld type.
2. Allow the system to evaluate the workpiece and identify the expected joint.
3. Let ROBBI AI create its own weld path within the parameters set for the application.
4. During welding, observe how the system adapts travel motion, oscillation or welding parameters when a gap opens or closes.
5. If a potentially weld-compromising condition is detected, treat the halt as a review trigger and inspect the recorded information.
6. Return operator feedback so the AI can recognize similar conditions in future work.
7. Use simulation-based learning before installation, then continue reviewing onsite data after deployment.

Common misconceptions to avoid

• This is not a tool that removes human-defined welding parameters; operators still set the procedure boundary.
• This is not a fully unrestricted autonomous welder; InduSuite is described as helping keep decisions within predefined specifications and qualification requirements.
• This is not necessarily a cloud-dependent workload; the release describes local processing inside the customer facility.
• This is not an all-or-nothing switch; conventional cobot use remains supported and full-AI mode can be disabled.

Action suggestions

1. Match the pilot to a difficult automation case first, such as pipe fabrication or TIG work where specialized skills are required.
2. Document the procedure inputs that operators must provide before any autonomous decision-making.
3. Build a review process around weld data and video, because the learning loop depends on operator feedback after defects or halts.
4. Check hardware fit early: Universal Robots UR10e, UR20e, seventh-axis reach, wheeled cart mobility, and compatible power sources.
5. Test both modes during evaluation: conventional operation and AI-assisted operation, with the stop behavior treated as an inspection point rather than a failure state.

The closing point is the most useful application takeaway. “With ROBBI AI, we are extending the role of AI beyond digital analysis and into the physical welding process, where intelligent decisions can directly influence how a weld is made,” Colvin said.