A robot can have precisely machined components and still miss its target. Small deviations in joint axes, bearing fits, link geometry and tool mounting combine with structural deflection, temperature changes and control behavior at the tool center point. The practical question is not simply how tight each tolerance should be, but which deviations matter at the working position and under the actual load.

Precision engineering for robotics connects individual component specifications to system-level performance. A useful evaluation starts with the task: where the tool must move, what it must do there, and how much error the process can tolerate. Those conditions determine whether tighter machining, better bearings, calibration or a stiffer structure will deliver the most useful improvement.

Accuracy and repeatability answer different questions

Accuracy describes how close the robot comes to a desired position or pose. Repeatability describes how closely it returns to the same position or pose under stated conditions. A robot can repeatedly stop at nearly the same place and still be offset from the intended target.

That distinction changes how a specification should be assessed. A fixed welding fixture may allow an operator to teach points on the installed robot. Repeatability can then be more important than absolute accuracy, provided the fixture and process remain stable. A robot programmed offline to reach coordinates generated from a digital model depends more directly on absolute accuracy and on the alignment between the model and the physical cell.

Position is only part of the result. A tool may reach the correct point while approaching at the wrong angle. For insertion, drilling or optical inspection, orientation error can affect contact or measurement even when positional error appears acceptable. Ask whether a reported figure covers position, orientation or both, and how it was measured.

Also distinguish the robot flange from the tool center point: the working point on the end effector. An angular error at the flange produces a larger positional displacement at the end of a longer tool. For small angles, the lateral displacement is approximately the tool length multiplied by the angular error in radians. A flange measurement alone therefore may not represent performance at the process point.

How component tolerances become tool-position errors

A dimensional tolerance sets the permitted variation in a feature, such as a bearing seat diameter or the distance between mounting faces. Geometric tolerances describe relationships such as position, orientation, form or runout. In a robot assembly, both matter: a shaft can have an acceptable diameter while its axis is misaligned with the mating structure.

Each joint and link establishes the position and orientation of the next. A small angular deviation near the base can affect the entire downstream reach; an error closer to the tool acts over a shorter remaining distance. The effect also changes with robot posture, because joint orientations and lever arms change as the arm moves.

Consider a simplified arm with a joint axis tilted slightly from its intended direction. At a short reach, the resulting tool displacement may be small. With the arm extended, the same tilt acts over a longer distance and can produce a larger displacement. This is a geometric illustration, not a prediction for every joint: the direction of the tilt and the configuration of the arm determine which part of the tool pose is affected.

Errors do not always add in the same direction. Some can partly cancel at one posture and reinforce each other at another. A tolerance stack-up that adds every limit as if it created the maximum error in the same direction can be useful for a conservative boundary, but it may not represent typical assembled behavior. Statistical methods can estimate variation across units, but only when the assumed distributions, correlations and assembly process are justified.

For a technical comparison, ask for a system error budget that connects significant component variations to tool-point error across the intended workspace. A single stack-up at a nominal posture can miss the location where the application is most sensitive.

Where tighter tolerances tend to matter most

Not every feature deserves the same manufacturing effort. The most consequential features are those that establish motion axes, locate the tool or influence a long downstream lever arm. Their importance depends on the task and the assembly design.

• Joint-axis alignment: Errors in the position or orientation of a joint axis affect the kinematic model. Features that locate bearings, housings and gearboxes deserve attention because their deviations can move the effective axis.
• Bearing interfaces: Seat geometry, shaft fits and housing alignment affect how a bearing is supported. Bearing clearance or preload can influence stiffness, friction and motion; excessive preload can also increase heat and shorten service life.
• Transmission interfaces: Gear and reducer mounting, coupling alignment and shaft support can contribute to lost motion or uneven behavior. A tight housing tolerance does not by itself establish the performance of the complete transmission.
• Tool and fixture mounting: Locating surfaces, fastener clamping and tool changer interfaces affect the relationship between the robot flange and the process point. A tool can be accurately made but inaccurately located after each change.
• Structural members: Link dimensions affect the kinematic geometry, while material, section shape and joints affect stiffness. Tight dimensions alone cannot prevent a slender arm from deflecting under load.

For bearing selections in particular, a precision grade is only one part of the decision. Mounting fits, lubrication, load direction, operating temperature and contamination control determine whether the assembly can retain the intended behavior in service. A highly precise bearing installed in a distorted housing will not reliably deliver its potential at the tool.

Build an error budget around the operating task

An error budget translates a process requirement into contributions that can be designed, measured and controlled. Begin by defining an acceptable position and orientation error at the tool center point, relative to a stated workpiece or reference frame. Identify the critical workspace, payload, applied process forces, approach directions, speed and temperature conditions.

Then separate sources by how they behave. Manufacturing and assembly variations often create systematic geometric offsets for a particular robot. Transmission clearance can make the final position depend on the direction of approach. Elastic deflection varies with payload, posture and process force. Temperature can change dimensions and alignments over time. Measurement uncertainty also affects how confidently the result can be assessed.

These categories should not automatically receive equal allowances. If an extended arm must press against a workpiece, stiffness and force direction may dominate the available error budget. If the task is unloaded inspection after warm-up, geometric calibration, camera or probe mounting, and measurement quality may be more influential. Sensitivity analysis helps identify which input variation produces the largest change in the relevant output.

When a tolerance appears critical, trace it through the assembly. A drawing may tightly control a bearing bore, for example, while allowing a less controlled mating face to determine the housing orientation. Datum choices, assembly sequence and fastening can therefore matter as much as the tolerance value on an individual feature. The budget should describe the assembled relationship that affects the tool, not merely collect dimensions from separate drawings.

Static geometry is only part of the operating error

A robot measured without a payload can behave differently when carrying a tool, holding a component or applying force. Links, joints, bearings, transmissions and mounting structures all have finite stiffness. Under load, their combined deflection changes the tool pose. A heavy base does not compensate for a flexible tool bracket, and an accurate gearbox does not eliminate bending in the arm.

Dynamic motion adds other effects. Acceleration can excite vibration, while servo following error can cause the tool to lag behind the commanded path. For a pick-and-place task, the error after settling may be the relevant measure. For dispensing or machining, the path during motion can matter more than the final stopped position. A positioning specification does not automatically describe path accuracy.

Temperature matters because materials expand and because heating is rarely uniform. Motors, gearboxes and brakes can warm some parts of the structure while the surrounding environment cools others. A robot calibrated when cold may show a different offset after sustained operation. The appropriate response might be thermal stabilization, compensation, a revised duty cycle or a wider process allowance; tighter room-temperature machining alone does not address every thermal effect.

In demanding environments, the load case should include more than payload mass. A subsea manipulator, for example, may encounter pressure, temperature variation and forces from contact with equipment. A robot working around large energy infrastructure may need long reach or heavy tooling, making deflection and mounting stiffness especially relevant. These are different engineering problems even if both applications ask for precise positioning.

What calibration can—and cannot—correct

Calibration estimates the relationship between commanded motion and measured robot geometry, then uses that information to improve positioning. It can reduce repeatable, systematic errors caused by differences between the nominal model and the assembled machine. Its value is often greatest when a robot must reach coordinates defined outside its own teaching process.

Calibration does not remove physical flexibility or clearance. A correction derived under one payload, temperature or posture may not fully represent behavior under another. It also depends on the quality of the measurement system and on the locations sampled: a model fitted in one area of the workspace may provide less improvement elsewhere.

Include the tool and the workpiece frame in the calibration plan. An accurately calibrated robot can still miss a feature if the tool center point is defined incorrectly or the fixture is displaced. Where tools are changed, the repeatability of the tool interface becomes part of the practical result. Where fixtures are moved, the method used to re-establish their position matters.

Calibration and manufacturing tolerances should be chosen together. Looser but stable geometry may be acceptable if it can be measured and compensated within the application. Tighter physical control may be necessary when errors change with load, when calibration access is limited, or when the process cannot tolerate variation between calibrations.

How to compare specifications and acceptance tests

Published figures are useful only when their test conditions match the intended decision. Two robots can report similar repeatability while being tested at different loads, speeds, positions or distances from the flange. Ask suppliers to identify the measurement point, approach procedure, environmental conditions and whether the result refers to a single direction, a set of poses or the usable workspace.

Recognized industrial robot performance test standards can provide a common framework for defining and measuring characteristics. A standard reference alone does not make two figures interchangeable: the selected tests and conditions still need to match. Request the test method and results relevant to the task rather than treating a standard reference as a substitute for them.

An acceptance plan should cover the conditions most likely to expose the failure mode. Depending on the application, that can include:

• Measurements at representative poses, including extended or heavily loaded configurations.
• Approaches to the same target from different directions, to reveal direction-dependent behavior.
• Position and orientation checks at the actual tool center point, with the intended tool installed.
• Measurements after the robot reaches its normal operating temperature and duty cycle.
• Path measurements during motion when the process depends on trajectory rather than stopped position.
• Fixture and base stability checks, so movement outside the robot is not mistaken for robot error.

The measurement system must be suitable for the required uncertainty and must have a clear reference relationship to the robot and workpiece. Agree on how results will be evaluated before testing, including whether the requirement applies to individual readings, a defined statistical measure or another stated criterion.

When a tighter tolerance is the wrong purchase

Tighter tolerances can increase machining, inspection, assembly and rejection costs. They are justified when the controlled feature has a meaningful effect on the task and that effect cannot be addressed more effectively elsewhere. They are less useful when the dominant error comes from tool flex, thermal drift, fixture movement or an unsuitable control strategy.

Before specifying a tighter part, identify the error it is intended to reduce and how that reduction will be verified at the tool. Compare it with alternatives such as a shorter tool, a stiffer bracket, a revised approach direction, improved thermal control or calibration. The best choice may be a combination: sufficient physical precision to keep the robot stable and repeatable, followed by calibration to correct the remaining systematic offset.

For procurement or design review, the most useful next step is to write the requirement in the language of the process: the allowed tool pose or path error, at the relevant positions, under the expected loads and operating conditions. Component tolerances can then be assigned where their influence is demonstrable, and the finished system can be tested against the outcome the robot must deliver.