Force-Controlled Robotic Assembly: Start With Location
Direct answer: Force control lets a robot sense and regulate contact force, which widens the tolerance window for insertion and tightening. It does not remove it. In force-controlled assembly the first thing to verify is where the part sits each cycle and how repeatable that is, which is a fixture question. The second is the approach pose at every point, because approach angle sets how the tool is loaded.
Who this is for: This guide is written for manufacturing engineers evaluating force-controlled robotic assembly, where compliance is often expected to solve a positioning problem it cannot solve.
Scope: It covers what force control does and does not compensate for, how the locating scheme sets the usable search window, and why approach pose changes tool loading. It does not cover fastener design, torque specification, or the assembly sequence of the product itself.
What force control does, and what it does not
Force control gives the robot the ability to sense contact and regulate the force it applies, which makes compliant operations such as insertion, mating and controlled tightening feasible. Its strength is handling small positional deviation and managing force during contact.
It is not a substitute for location. Compliance operates within a search window, and outside that window the strategy simply fails more smoothly than a rigid one would. The window is finite, and its size is a design parameter rather than an unlimited allowance.
The first question in any evaluation is therefore not about the arm. It is where the part sits at the start of each cycle and how much that position varies, and the answer to that lives in the fixture.
In practice the projects that struggle are the ones where compliance was expected to absorb a fixture problem. According to ISO 10218-2:2025, it is the robot application that is assessed, and force-controlled assembly is an application property rather than an arm feature.
The locating scheme sets the search window
Datum faces and locating pins determine the position of the part relative to the robot base, and therefore whether a taught point remains valid on the next part. That relationship is what makes a programmed assembly sequence repeatable at all.
The scheme needs a clear separation between primary datums, which establish position, and auxiliary supports, which only prevent deflection and vibration. Where the two roles are mixed, repeatability tends to be intermittently good in a way that resists diagnosis.
On castings and weldments the raw dimensional variation of the part itself is a further input. Whether datums are taken on machined faces or on as-cast surfaces decides how much of that variation is transmitted into the assembly position.
A force-controlled assembly cell inherits whatever repeatability the fixture delivers, so the measured repeatability of the existing fixture is the single most useful number to bring into the evaluation.
Approach pose changes how the tool is loaded
The angle from which the robot approaches an assembly point determines the direction in which the tool is loaded. The ideal case puts the load along the tool axis, which gives the cleanest force feedback and the lowest side loading on the mechanism.
Real layouts rarely allow that at every point. Obstructions, deep locations and assembly points oriented in different directions force compromises, and each compromise introduces a side component that degrades both the force signal and tool life.
For that reason the approach pose at every point should be checked during design rather than discovered during teaching. The check requires the part model and the tool model together, since either one alone will miss the interference cases.
Decision table: what the part and fixture justify
The table maps what the part and its fixture can justify on their own against the evidence a supplier still has to produce.
| Evidence from the part and fixture | Approach it justifies | Evidence you still owe |
|---|---|---|
| Machined datum faces, measured repeatability well inside the search window | Taught positions with force control managing contact only | Measured repeatability data, and approach pose verified at every assembly point |
| As-cast or weldment datums with significant part-to-part variation | Either a revised datum strategy or added sensing, decided before robot selection | Actual variation range, since it decides whether the search window is adequate |
| Assembly points oriented in several directions | Pose study with part and tool models before layout is fixed | Side-load assessment at the compromised poses and its effect on tool life |
| Torque results required to be traceable | Tool and control architecture chosen for data capture, not added afterwards | The record format required, and where in the line it will be stored |
Torque traceability and what it demands upstream
Where tightening results have to be traceable, the requirement reaches back into tool selection and control architecture. Capturing torque and angle per fastener is straightforward when designed in and awkward when added to a cell already built.
ISO 5393:2017 describes performance test methods for rotary tools for threaded fasteners and provides a common basis for discussing tool capability. Agreeing what will be recorded, and to what, belongs in the specification rather than in commissioning.
The same applies to what happens on a failed result. A rule that is written down and executed is what turns traceability into a control; without it, the data is collected and nothing acts on it.
According to ISO 5393:2017, rotary tools for threaded fasteners are characterised by defined performance test methods, which gives a common basis for comparing tool capability rather than relying on supplier description. EVST specifies the record format at the same time as the tool.
A collaborative arm still needs a risk assessment
The presence of a collaborative arm does not by itself establish that an application is safe without assessment. The boundaries depend on the task, the tool, the part and the workspace, not on the class of robot.
ISO 10218-2:2025 addresses the robot application and ISO 12100:2010 provides the general risk assessment framework. ISO/TS 15066:2016 gives guidance specific to collaborative operation and is the appropriate reference where human access to the workspace is intended.
A powered tool such as a nutrunner changes the hazard picture regardless of the arm carrying it. That is a task property, and it belongs in the assessment on its own terms.
According to ISO/TS 15066:2016, collaborative operation is defined by the task and the workspace rather than by the class of robot, so a powered tightening tool changes the assessment regardless of which arm carries it. EVST assesses the task, not the badge on the arm.
What the footage supports
The footage shows a collaborative arm carrying a nutrunner spindle and working through assembly positions on a part held in a locating fixture, with the fixture datum faces and the approach poses visible.
It does not show torque values, assembly yield, cycle time or force control accuracy. Those depend on your fasteners, your part and your acceptance criteria, and have to be established under your own conditions.
What to put in the capital request
State the datum scheme and the measured repeatability of the existing fixture, the distribution of assembly points with the required approach direction for each, the fastener specification and any torque traceability requirement, the cycle requirement and the loading method.
Then state what the supplier still has to prove: that the search window is adequate for your measured variation, that every approach pose is reachable with acceptable side loading, and that torque data capture meets the record format you need.
Written this way the request describes a force-controlled assembly application rather than a robot purchase, which is what makes competing proposals comparable.
Frequently asked questions
Can force control compensate for a poorly located part?
Only within its search window. Outside that window the robot performs the wrong motion compliantly rather than correcting it. Force control widens the tolerance requirement, it does not remove it.
What decides repeatability in this kind of cell?
The locating scheme. Datum faces and pins establish where the part sits relative to the robot base, and auxiliary supports should only prevent deflection. Mixing the two roles produces repeatability that is intermittently good and hard to diagnose.
Why does approach angle matter?
Because it sets the direction in which the tool is loaded. A load along the tool axis gives the cleanest force signal and the lowest side loading; every compromise introduces a side component that affects both the signal and tool life.
Is a collaborative robot automatically safe for this?
No. The safety boundaries depend on the task, the tool, the part and the workspace. A powered tightening tool changes the hazard picture regardless of which arm carries it, and the application still requires assessment.
Project inputs for an application review
Send the following and the application can be reviewed against the real part instead of a robot specification:
- the part datum scheme, and measured repeatability of the fixture you use now
- the distribution of assembly points with the required approach direction for each
- the fastener specification and any torque and angle traceability requirement
- the cycle requirement, and how parts are loaded and unloaded
- the workspace layout, including any intended human access
If you are evaluating force-controlled robotic assembly, send the datum scheme and the measured repeatability of the current fixture, the assembly point distribution with required approach directions, the fastener and traceability requirements, and the cycle and loading method. The boundaries of the application can then be judged in that order. Related reading: robot product range, wider robotic factory integration, sensing and vision-assisted process context.