A robot arm repeats the same motion thousands of times each day, and the accuracy of that motion is set by the weakest link in its drivetrain. A bearing housing off by 0.03 mm, a gear with too much backlash, or a coupling that winds up under load all compound into a position error the controller cannot correct. Robotics therefore ranks among the most demanding customers for precision machining: parts stay small, geometries grow complex, and the tolerance stack-up leaves little room for compromise. This article explains what makes robot components difficult to machine, which parts carry the motion, and how the tolerances on the drawing translate into arm performance.
Why tolerance eats the budget on an arm
Industrial and collaborative arms repeat a pose to a repeatability spec of ±0.02 to ±0.1 mm, and that number is the sum of every error in the chain. Gear backlash, bearing runout, coupling wind-up, and housing squareness each consume a slice of that budget. A gearbox housing that holds its bore centers to ±0.01 mm leaves room for the rest of the stack; a housing at ±0.05 mm forces the gear set to run tighter and wear faster.
Linear actuators and gantries add a separate requirement: the rail mounting faces must stay parallel within 0.02 to 0.05 mm over the full travel length, or the carriage binds. A part that looks like a generic machined bracket on the bench becomes a precision component the moment it carries a rotary joint. Our robotics machining work shows the kinds of components shops typically build for this industry.
The components that carry the motion
A short list of parts recurs across nearly every robot design. Gear housings and gearbox bodies wrap the harmonic or planetary drive and must keep their bore axes coaxial to within a few microns. Joint brackets and wrist castings get machined in a single setup so bearing bores and mounting faces stay true to one another. Shafts carry the output torque and run ground to Ra 0.4 to 0.8, with bearing journals held to ±0.005 mm. Couplings connect motor to reducer, stiff enough not to wind up yet forgiving enough to absorb misalignment. End-of-arm tooling plates, cable brackets, and sensor mounts complete the list.
Each of these is a machined part first and a robot part second, which is why a shop already doing tight-tolerance housings and shafts usually handles robotics work well. The gear machining guide, the shaft machining guide, and the notes on couplings cover these components in detail.
Materials chosen for stiffness, not just strength
Robot parts live in a weight-versus-stiffness trade. 6061 aluminum remains common for housings and brackets: it machines fast, anodizes, and weighs about a third of steel. 7075 aluminum goes where the load runs higher, with a yield strength around 500 MPa against 6061’s 240 to 280 MPa. Steel, often 4140 or hardened 1018, carries the shafts and gear blanks that see continuous torque. Engineering plastics such as acetal and PEEK appear as wear pads and cable guides where friction and light weight matter.
The hidden requirement is stiffness rather than raw strength. A thin aluminum arm that flexes 0.1 mm under load destroys repeatability even if it never breaks, so walls get beefed up or ribbed instead of downgraded. Shops that understand robotics design around deflection, not just yield strength.
Weight enters every material choice because arm inertia directly drives cycle time and settling after each move. Heavier links demand larger motors and add energy consumption through every operating hour, which is why aluminum dominates stationary housings while steel stays concentrated on the torque-carrying members that simply cannot shed mass. A buyer working from a concept model should ask which members actually rotate or translate, then reserve the higher-density alloy only for those load paths.
Drawing the part so it machines correctly
Finish the bores and bearing seats in one setup whenever the geometry allows, because moving the part introduces datum error. Call out coaxiality and perpendicularity on those bores, not only the diameters. Specify the bearing journal surface with a ground finish and a tight circularity, typically 0.005 to 0.01 mm. Keep bolt patterns on mating faces to a positional tolerance that matches the bearing, usually ±0.02 to 0.05 mm, and design for deburr: robot cable systems catch sharp edges, and a burr that survives plating becomes a maintenance problem.
Tell the shop which bores carry bearings and which are only clearance holes; inspection effort splits sharply between the two, and so does the price. Shops running statistical process control on critical bores can hold Cpk above 1.33 across a batch, a level that matters for parts entering a production arm rather than a one-off demo. Most robot builders start with a handful of machined joints before any casting or molding: a prototype arm part typically runs five to ten business days in 6061, at USD 80 to 300 per piece depending on size and setup count. Machine the first iterations from solid, validate the gear mesh and bearing fit, and only then consider a cast housing for volume. Because tolerance stacks are easy to misjudge on paper, build the first joint, measure actual backlash and runout, and feed those numbers back into the model before committing to production tooling.
