Importance of Sensing and Machine Vision in Robotics

Updated: 4/2026

Difference between Offline and Online sensing
Prior to digging into sensing and machine vision as applied to robotic welding, it is worthwhile to understand the offline and online nature of these technologies.
Offline sensing is performed prior to or after welding, not during the process. While delivering valuable information about where the weldment features and joints are located, it will increase its takt (production) time.
Online sensing occurs while the weld is happening. The sensing is done ahead of the weld, using the collected data to correct the path and parameters before the robot weld torch gets to that part of the weld. These real-time corrections keep the torch within the joint maintaining the stick-out, work angle and push angle required for welding consistency.

Touch-sensing [offline]

A welding circuit can be used as a continuity tester, with the circuit closing (electricity flowing) when the welding wire or torch nozzle contacts the weldment. Welding robot controllers have high-speed inputs that can detect that “contact” signal and record the robot’s position, much like a coordinate measuring machine (CMM).
Comparing this sensed position against the original position where the welds were programmed, determines the offset to the NEW position of the weldment, and results in the shifted positions required to match. These shifts (corrections) can be performed in 1, 2 or 3 directions, although an equal number of searches is required.
Example: 2 directional corrections require searching in 2 directions
Being an offline sense, every search takes time and the robot is not welding during that time. Search speed is limited by the response time of the robot’s high-speed input. Too high a search speed results in inaccurate, delayed position detection and/or a crashed/bent welding torch.
Weldments with non-conductive or inconsistently-conductive surfaces due to scale, rust, dirt, oil, oxidation (to name a few) or other surface contamination would disqualify this approach.

Laser touch-sensing [offline]
A faster approach to touch sensing is to use a non-contact, laser-triangulation sensor to measure the distance between the torch and the weldment. Connected to the same high-speed input as conventional touch-sensing, this approach results in 4-times faster detection while eliminating the need to contact the weldment with the torch.
This does require additional cost for the sensor, enclosure and mounting and the need to calibrate prior to use, but it does work on non-conductive/poorly conductive weldments. The sensing distance is 6”-12”, keeping the sensor away from the welding arc. Highly-reflective surfaces can prove troublesome with the potential for missed or inaccurate sensing corrupting the results.

Through-arc seam tracking (also referred to as arc-sensing or TAST) [online]

TAST works by monitoring the voltage or current of the welding arc to automatically adjust the weld torch’s path in real-time, compensating for variations in the weld joint. This is achieved by having the robot perform a weaving motion while welding, allowing the system to interpret changes in the arc’s electrical feedback caused by variations in torch-to-workpiece (stick-out) distance. The system then corrects the torch’s vertical position and cross-seam position to keep it centered on the joint.
Weaving requires a fillet weld with heights at least ¾ of the material thickness to generate good results. American Welding Society (AWS) D1.1 welding code provides data tables defining this minimum size. This approach is best suited for long or curved seams with results on sharp corners or abrupt changes being less effective. Weld material thinner than 2-4mm is not well suited to the heat input required for this approach.

Laser seam tracking [online]

This sensor uses a laser to project a line onto a weld joint, which is then viewed by a camera. This system uses this real-time 3D weld joint information to guide a welding robot, making dynamic adjustments to its path to compensate for variations, misalignment, or distortion. The benefits include improved weld quality, increased efficiency, and reduced need for pre-programming and manual adjustments.
Because this sensor collects weld joint information ahead of the weld torch, it needs to be mounted near the torch tip. This limits joint access, requiring some planning for when and how it is used. This proximity to the welding arc also requires protective housing, water or gas cooling and a consumable lens.
In addition to weld joint tracking, joint start and stop points can be detected, surface normal orientation can be maintained, and joint gap variation can be adaptively filled.
Like laser touch sensing, highly-reflective surfaces can prove troublesome with the potential for missed or inaccurate sensing corrupting the results.

Laser weld inspection [online or offline]

A variation of the approach used by laser seam tracking, weld inspection detects 3D geometric weld defects (undercut, burn-through, porosity and spatter). Internal weld defects are not detectable using this technology which can only see the surface features of the weld.
Offline inspection can be performed 4X faster than welding speed when not combined with online seam tracking. Inspection results can be both qualitative and quantitative, including Go/No-Go detection.