Backlash in robotics is the lost motion in a drivetrain: the small angular or linear clearance between mating parts, such as gear teeth, that allows the output to move a little without any movement of the input. Also called gear backlash, it is one of the most important error sources in robot actuators, because every joint that uses a gearbox, belt or lead screw inherits some of it.
Backlash matters most when a joint reverses direction or when external load changes sign. During that reversal the motor turns through the gap before the teeth make contact again, so the joint position briefly stops following the command. In precision arms, legged robots and haptic devices, this hidden dead zone degrades positioning accuracy, stiffness and control stability.
How backlash arises in a drivetrain
Backlash appears because mating parts need deliberate clearance to run. Gear teeth require room for lubricant film, thermal expansion and manufacturing tolerance; without it, the mesh would bind, heat up and wear rapidly. The clearance is measured along the pitch circle as circumferential backlash or, at the output, as an angular backlash in arcminutes.
Each stage of a transmission contributes its own gap, and the gaps add up. Because backlash measured at the output shaft is not reduced by the gear ratio, a multistage gearbox accumulates lost motion from every mesh, plus clearances in bearings, keyways, couplings and splines. Wear later enlarges these gaps over the service life of the joint.
The following table lists the usual mechanical sources of lost motion in a robot joint and what drives their size.
| Source | Mechanism | What increases it |
|---|---|---|
| Gear mesh clearance | Gap between tooth flanks | Loose center distance, thin teeth, wear |
| Shaft-hub connection | Play in keys, splines or set screws | Shock loads, fretting |
| Bearing play | Radial and axial clearance | Missing preload, worn races |
| Belt and pulley | Tooth-to-groove clearance and stretch | Low tension, worn belt |
| Lead screw nut | Axial clearance between thread flanks | Plain nut without preload |
Backlash values across gearbox types
Gearbox selection largely determines how much backlash a joint will have. Manufacturers quote it in arcminutes (arcmin), where one degree equals 60 arcmin. Values below 1 arcmin count as near-zero backlash, while general-purpose spur gearboxes often exceed 30 arcmin.
The table below gives typical catalog ranges; actual values depend on the quality class, size and preload of each specific unit.
| Transmission | Typical backlash | Common use in robots |
|---|---|---|
| Harmonic drive | Near zero, below 1 arcmin | Collaborative and industrial arm joints |
| Cycloidal drive | About 0.5 to 1 arcmin | Heavy industrial arm axes |
| Precision planetary gearbox | About 1 to 5 arcmin | Wrist axes, mobile manipulators |
| Economy planetary gearbox | About 10 to 30 arcmin | Hobby and low-cost robots |
| Spur gear train | About 20 to 60 arcmin or more | Toy servos, simple mechanisms |
| Quasi-direct drive actuator | Small, low ratio of 6:1 to 10:1 | Legged robots, haptic devices |
A harmonic drive achieves its low value because the flexspline engages many teeth at once under elastic preload. A planetary gearbox shares load across several planets but still needs clearance in each mesh. A quasi-direct drive actuator keeps backlash small mainly by using few gear stages and a low ratio, which also improves backdrivability.
Effects of backlash on robot control
Backlash turns a linear drivetrain into a nonlinear dead-zone element. When the motor reverses, the output stays still until the gap closes, and then the teeth meet with an impact. For a controller, this produces delay, loss of stiffness and sudden torque transients precisely when precise motion is needed.
The consequences scale with arm length. A backlash of 5 arcmin at a shoulder joint with a 1 m reach yields about 1.5 mm of uncertainty at the tool, which is unacceptable for assembly but harmless for coarse material handling. The table below summarizes typical symptoms.
| Symptom | Underlying cause | Where it is noticed |
|---|---|---|
| Limit cycling | Position loop hunts inside the dead zone | Holding a fixed pose |
| Contour error at reversals | Output lags while gap closes | Circular or zigzag paths |
| Tooth impact noise | Flanks collide after free travel | Fast direction changes |
| Reduced joint stiffness | Load moves output within clearance | Machining, contact tasks |
| Poor repeatability | Final position depends on approach direction | Pick-and-place calibration |
A key distinction is motor-side versus load-side sensing. If the only encoder sits on the motor, the controller cannot see the backlash at all and reports a position the output never reached. A second, output-side encoder closes the loop around the gap, but it can also make the loop less stable, because the mechanical path inside the loop now contains the nonlinearity.
Methods to reduce or compensate backlash
Designers fight backlash with mechanical elimination, preload and software compensation, usually in combination. The right mix depends on accuracy targets, cost and how much friction the joint can tolerate.
- Zero-backlash transmissions such as harmonic and cycloidal drives remove the problem at the source, at higher cost and with some compliance.
- Split or spring-loaded gears keep two gear halves pressed against both flanks, eliminating clearance at low to moderate loads.
- Preloaded ball screws and bearings remove axial play in linear actuators.
- Dual-motor antibacklash drives apply opposing torque from two motors so that one always holds contact.
- Backlash compensation in software adds a stored offset to the command whenever motion reverses.
Every remedy carries a cost, as the decision table below shows. Preload increases friction and heat, and fixed software offsets drift as gears wear.
| Situation | Recommended approach | Main trade-off |
|---|---|---|
| Precision arm joint | Harmonic or cycloidal reducer | Cost and torsional compliance |
| Linear axis with lead screw | Preloaded ball screw nut | Higher friction, lower efficiency |
| Large rotary axis | Dual-motor torque bias | Extra motor and control effort |
| Low-cost robot | Software reversal offset | Accuracy fades with wear |
| Legged or haptic robot | Low-ratio quasi-direct drive | Larger, heavier motor |
Measuring backlash and common mistakes
Backlash is measured by locking the input and rocking the output with a small, defined torque, typically about 2 percent of rated torque, while a dial indicator or output encoder records the travel. Catalog values usually refer to this method, so comparing figures from different vendors requires checking the test torque and the definition used.
A frequent mistake is to confuse backlash with torsional stiffness. A harmonic drive has almost no backlash but measurable elastic windup under load, whereas a stiff spur train may have large clearance. Another error is calibrating a robot from one approach direction only, which hides the gap until the robot reverses in production. Engineers should also re-measure periodically, because backlash grows with wear and lubrication loss.
A non-technical view
Imagine turning a loose steering wheel. For the first few degrees, nothing happens to the car, and only then do the wheels respond. Backlash is that small free play inside a robot’s gears. When a motor changes direction, it first spins through the gap before the joint actually moves.
For a robot that only stacks boxes, this hardly matters. For a robot that threads screws or holds a surgical tool, a tiny gap becomes a visible error at the fingertip. That is why precise robots use special gearboxes or clever software to keep this free play as close to zero as possible.
A concrete example
A small electronics manufacturer builds a desktop SCARA robot to insert connectors into circuit boards. The first prototype uses an economy planetary gearbox with about 15 arcmin of backlash on the first joint and a motor-mounted encoder only. During testing, connectors land correctly when the arm approaches from the left but miss by about 0.6 mm when approaching from the right.
The engineers measure the gap with a dial indicator, add a reversal offset in the controller as a short-term fix, and then switch the first joint to a harmonic drive with under 1 arcmin. After the change, the insertion error drops to about 0.05 mm from either direction, the software offset is no longer needed, and first-pass assembly yield rises noticeably.