Collaborative Robots
Hand Guiding, Teach Pendant, Offline: Which Is Fastest
Three ways to program a collaborative robot, measured by setup time, accuracy and where each one breaks down. Hand guiding wins under 30 waypoints, offline wins above 200.
For a task with fewer than about 30 waypoints, hand guiding is fastest, often finishing in 15 to 40 minutes. Between 30 and 200 waypoints the pendant wins because it holds precision without drift. Above roughly 200 waypoints, or when the part exists as CAD and the cell does not exist yet, offline programming wins outright, and it is the only method that does not consume production time.
What each method actually is
The three methods are not competitors for the same job. They occupy different points on a curve of setup effort against precision, and most working cells use two of them.
| Method | Setup for 20 points | Typical accuracy | Robot occupied | Best for |
|---|---|---|---|---|
| Hand guiding | 15 to 40 min | 0.5 to 2 mm | Yes | Short runs, frequent changeover, coarse paths |
| Teach pendant, jogging | 40 to 120 min | 0.05 to 0.3 mm | Yes | Precise insertion, fixed fixtures |
| Offline from CAD | 2 to 8 h, off-line | Depends on calibration | No | Long paths, welding, deburring, new cells |
| Hybrid: offline plus touch-up | 2 to 8 h plus 20 to 60 min | 0.05 to 0.3 mm | Briefly | Almost every serious production cell |
Hand guiding: fast, and less accurate than it feels
The operator grasps the arm and moves it, recording positions with a button. It is genuinely quick and it needs no programming skill, which is the whole appeal. Two limits deserve honesty.
First, precision. A hand-guided point lands where the human hand stopped, which for an unsupported arm is typically within 0.5 mm to 2 mm. That is fine for picking from a tray and hopeless for a 0.1 mm clearance insertion. Most controllers offer a numeric nudge after recording, which recovers the precision but costs the speed advantage.
Second, drift over many points. Guiding 60 points by hand takes longer than the pendant and produces less consistent spacing, because human fatigue is real and the operator has no reference frame between points.
The pendant: still the precision instrument
Jogging with the pendant addresses the arm in a chosen coordinate frame, which is what makes it precise and repeatable. Working in the tool frame for approach vectors and in the workpiece frame for placement turns a fiddly 3D problem into a set of one-dimensional moves.
The cost is time and skill. A 20-point insertion path with correct approach and retract vectors, blend radii and a sensible speed profile is comfortably an hour for someone who knows the controller, and considerably more for someone who does not. That hour is production time on a real machine.
Offline: the only method that scales
Paths are generated from CAD in a simulation environment, checked for reach, singularities and collisions, then posted to the controller. For welding a seam of 3 m, for deburring a contour, or for a paint path, no manual method competes: an offline system can place hundreds of points in minutes and re-generate them all when the part revision changes.
The break-even is governed by one thing, calibration. An offline path is only as good as the agreement between the virtual cell and the real one. Cell calibration typically takes half a day and reduces the residual error to a level where a short touch-up pass fixes the rest. Skipping calibration is the reason offline programming has a reputation for not working.
- Reach and singularity checks catch layout mistakes before steel is cut, which is where offline pays for itself even on small jobs.
- Version control of paths becomes possible, because the source is a file rather than a controller memory.
- Cycle time can be estimated before purchase, which changes the quotation conversation.
Choosing in practice
A rule of thumb that survives contact with real cells: count waypoints, then count how many times the task will change.
- Under 30 points, changes weekly. Hand guiding with numeric touch-up. Total setup under an hour.
- 30 to 200 points, stable for months. Pendant, structured with frames and subroutines so a fixture move does not mean reteaching everything.
- Over 200 points, or geometry driven by CAD. Offline, plus calibration, plus a touch-up pass on the critical points.
- Any new cell that does not physically exist. Offline, always, because the alternative is discovering a reach problem after installation.
Frequently asked questions
Is hand guiding accurate enough for assembly?
For coarse placement yes, for close-tolerance insertion no. Expect 0.5 mm to 2 mm from the guided motion itself. Most controllers allow a numeric correction after recording, which restores precision at the cost of the speed advantage.
Do I need special hardware for hand guiding?
Yes. Hand guiding is a collaborative operation type and requires a three-position enabling device plus safe speed monitoring. Arms with joint torque sensing usually support it natively; others need a dedicated guiding handle.
When does offline programming pay for itself?
Above roughly 200 waypoints, whenever the part geometry comes from CAD, and on any cell that does not exist yet. Below that threshold the calibration effort usually outweighs the savings unless the path will be regenerated often.
Why do offline paths not fit the real robot?
Because the virtual cell and the physical cell disagree. Fixture position, tool centre point and robot kinematic error all contribute. Cell calibration, typically half a day, reduces the mismatch to a level where a short touch-up pass closes the gap.
Can I mix methods in one program?
Yes, and most production cells do. Generate the bulk path offline, then touch up the few points where tolerance matters using the pendant, and keep hand guiding for quick changes to pick positions.
Sources
- ISO 10218-1:2025, Robotics, Safety requirements, Part 1International Organization for Standardization, enabling device and reduced speed requirements
- ISO/TS 15066:2016, hand guiding as a collaborative operation typeRequirements for guiding devices and safe speed
- ISO 9283, manipulating industrial robots, performance criteria and test methodsInternational Organization for Standardization, definitions of pose accuracy and repeatability