1. Programming the Robot Before the Robot Exists
A robot program is a sequence of targets, each a position and orientation of the tool, linked by motions, guarded by signals and wrapped in logic, and the higher goal of robot programming is to build that sequence without taking the robot out of production. Offline programming does exactly that: the robot, its cell, the fixtures, the workpiece and the tools are modeled in a simulation environment, the path is taught and verified against the model, collision free and fully timed, and the finished program is downloaded to the real controller. The robot is programmed while it keeps welding the current batch.
Path planning is the layer below the program: deciding the series of tool poses that follows the workpiece geometry, clears the tool and the cell, and respects the robot’s limits, all while the end effector delivers the process, the weld, the paint, the glue or the arc at the correct speed and orientation. This article develops the two together, the geometric path planning and the offline programming workflow, from the cell model to the validated program, and closes with the practical engineering rules that separate a program that runs from one that crashes on the first dry cycle.
The offline program is the meeting place of three truths: the truth of the CAD model, the truth of the calibrated cell and the truth of the process parameters, and the program is only as good as the agreement between the first two.
2. The Cell Model and Its Calibration
Offline programming is a simulation whose worth depends entirely on how faithfully the simulation models the real cell. The robot model, its kinematic chain and its reach envelope, is published by the manufacturer, and the simulation builds the cell from the robot model plus the fixtures, the conveyor, the safety fences and the workpiece geometry imported from the CAD system. The world of the simulation, its coordinate frames and the placement of every object, must match the real world, and that match is made by calibration, the measurement of the real cell that aligns the model to the floor.
The calibration binds the robot to the world. The robot base frame, the workpiece frame and the tool frame are the three coordinate systems that must agree, and the calibration procedure, usually measuring known reference points on the fixture with the real robot, derives the transform between the robot’s world and the simulation’s world. A weld programmed in the simulation that trusts an uncalibrated coordinate transform lands the torch centimeters away from the joint in reality, not because the path was wrong but because the two worlds disagreed. The calibration targets, the accuracy of the measurement and the stability of the fixture define the achievable placement accuracy of the imported program.
The model must also carry the dynamic truth: the reach, the singularities and the axis limits of the robot, the weight of the tool, the cable dressings that a collision checker must respect, and the real geometry of the workpiece as it arrives, with its tolerances and its weld gaps. A simulation that demands a perfect model is defeated by the real part before the first cycle, so the well built offline program is validated against the real as-built fixture, the coordinate transform is re checked and the critical path is taught with the measured deviation, not the nominal geometry.
3. Path Planning Fundamentals
The path is planned in a space larger than the robot’s own: the task space, the space of tool positions on the workpiece, and the joint space, the space of the robot’s axis angles. The planner defines the sequence of task targets along the workpiece, the weld seam divided into a series of torch poses, and then solves for each task pose the joint configuration that places the tool there, and the sequence of joint configurations that moves the robot between them without striking itself, the workpiece or the cell.
Path planning is ruled by three constraints. The first is reachability, that every task target lies inside the robot’s working envelope at a valid joint configuration, a constraint that is checked by the inverse kinematics solver and violated the moment a target is planned too close to the robot or beyond its stroke. The second is collision freedom, that the whole robot body, the tool, the cables and the workpiece, sweep through space without touching anything, a check performed against the full geometry of the cell at every point of the path. The third is singularity avoidance, because the robot’s axes cannot move through certain configurations, the wrist straight, the center of rotation aligned, where the solution degenerates and the robot would request an impossible speed.
Process motions add a fourth constraint: the tool must travel at a controlled speed and orientation along the path, the welding torch leaned at the correct angle and dragged at the correct travel speed, the paint gun kept perpendicular, the glue nozzle held at a constant standoff. The planner is therefore a negotiation between the geometric path, the collision space and the process window, and the program that survives dry cycling is the one that keeps the tool inside all three at every instant.
4. Building the Program Offline
The offline programming workflow starts in the simulation, where the programmer reads the workpiece geometry and defines the work objects: the target poses, the weld seams, the reference frames. The path is then taught on the virtual robot, either by clicking through the geometry, picking the torch pose at the start and at the end of each seam and letting the software interpolate the in between, or by exporting the path directly from the CAD geometry, the seam divided automatically into pose after pose at the correct spacing. The programmer reviews the generated path in the simulation, watching the robot sweep through the cell, checking the reach, the collisions and the process angles before a single line of controller code is written.
The logic wraps the geometry: the signals that drive the positioner, the weld gun, the clamps and the conveyor, the branch and wait statements that synchronize the robot with the cell, the search and offset logic that adapts the path to the real part. The cycle time is simulated with the real acceleration and speed parameters, giving the takt estimate that the cell design needs, and the program is versioned and stored with the cell model, so the next product variant starts from the previous validated program instead of an empty screen.
The translation to the controller is automatic: the offline tool exports the program in the controller’s native language, the motions, speeds, zones and signals, ready to be loaded, and the download is the first time the physical robot learns what it will do next Tuesday. Because every motion was already rehearsed in the simulation, the download is followed not by weeks of teach, but by a dry cycle and a process check.
The dry cycle is the bridge between the simulated world and the real world: the robot runs the program with the tool disabled, and the commissioning engineer watches the sweep, verifies the clearances with the real cell, and corrects the deviations that the model never captured.
5. Collision Checking and Optimization
The value of the offline program is measured by how little of it the commissioning engineer has to correct, and the two tools that keep that number small are collision checking and path optimization. The collision checker runs the full robot model, the tool, the cables and the workpiece through the entire path at the real joint angles, detecting the interference before it costs a fixture, a torch or a shift. The modern checker also tests the path envelope with a safety margin, so the robot clears the cell not by a paint-thick film but by a documented clearance, and it flags the close passages for the engineer’s judgment.
The optimization layer works on the same path: it looks for the shorter, smoother, faster route between the same targets, the path that reduces the cycle time, the passes that swing wide unnecessarily, the approach that makes an extra half turn, and it searches the joint space for the configuration changes that shorten the overall travel. The optimization is constrained by everything the planner respected, the reach, the collisions, the process angles and the singularities, so the optimized path is genuinely better, not merely different, and the improvement is visible in the simulated cycle time before it is ever measured on the floor.
Positioner coordination is a special optimization: the robot and the positioner move together, and the combined motion keeps the weld seam flat, always in the ideal orientation, while the robot works in the easy part of its envelope. The coordinated motion is planned as one, the two arms and the two axes synchronized, and the offline tool resolves the joint solution that keeps the torch in the process window while the positioner turns the part for the next pass. The result is a weld that is faster, better angled and less reworked than any static-table program, and it is achieved entirely in the simulation.
6. The Commissioning Conversation
The offline program does not eliminate commissioning; it changes its nature. The engineer no longer teaches every point by jogging the pendant, but verifies the downloaded program against the real cell: the calibration is confirmed, the critical clearances are walked, the tool center point is verified, and the process is tuned on the first real cycles. The ideal commissioning is boring, a sequence of confirmations that the simulation predicted correctly, and the rare collision or overshoot is traced back to the model, the calibration or the geometry, not to the programming language.
The search and offset features connect the program to the real parts: a seam search motion, the torch probing for the real joint before the weld, an adaptive offset that shifts the path to the measured position, a force or vision sensor that feeds the correction, all written in the offline program and verified in the simulation. The program becomes robust to the tolerance that the simulation labeled nominal, and the cell runs longer between interventions.
The knowledge captured in the program is the real asset: the validated path, the calibrated frames and the process parameters are stored with the cell model, so the next part variant, the next product family, starts from a program that already knows the cell. The offline programming pays its wage not once, in the saved commissioning hours of the first program, but every time a new program is derived from the library of validated work.
7. The Offline Programming Procedure
- Build and calibrate the cell model against the real fixture
- Import the workpiece geometry and define the work objects
- Plan the path in the task space and resolve the joint space
- Add the process logic, signals and safety interlocks
- Run the collision check and optimize the path and cycle time
- Export the controller program and download it to the robot
- Dry cycle, verify calibration, and tune the process on first parts
| Element | Contribution |
|---|---|
| Calibrated model | the placement accuracy of every imported target |
| Task space path | the tool poses on the workpiece geometry |
| Joint resolution | reachable, smooth, singularity free configurations |
| Collision check | clearance of the whole sweep through the cell |
| Process logic | signals, branches and adaptive searches |
| Optimization | shorter cycle time within the same constraints |
Commissioning rule: if the real cell disagrees with the program, suspect the calibration and the geometry before the logic. The offline program inherits the accuracy of the model it was taught in.
Robot path planning and offline programming move the machine learning out of production and into the simulation, where it is faster, safer and cheaper to be wrong. The discipline of the calibrated model, the collision checked path and the validated program lets a robot cell go from empty floor to running parts in a fraction of the time a pendant taught program would demand, and the knowledge survives in the cell model, ready to program the next product before it is even built.
