Perception and Sensors
Encoder Resolution: Why 17 Bit Is the Floor
Encoder resolution at the joint decides how finely a robot can position its tool. Here is the arithmetic from bits to millimetres, and why the gearbox matters more than the encoder.
A 17-bit encoder divides one revolution into 131,072 counts. Behind a 100:1 gearbox that is 13.1 million counts per output revolution, or 0.0000275 degrees per count, which at a 1 m arm length is well under a micrometre. Encoder resolution is therefore almost never the limit on robot accuracy. Gearbox backlash, structural deflection and thermal expansion are.
From bits to millimetres
| Bits | Counts per rev | Degrees per count | Behind 100:1 | At 1 m radius |
|---|---|---|---|---|
| 13 | 8,192 | 0.0440 | 0.00044 deg | 7.7 um |
| 15 | 32,768 | 0.0110 | 0.00011 deg | 1.9 um |
| 17 | 131,072 | 0.00275 | 0.0000275 deg | 0.48 um |
| 19 | 524,288 | 0.00069 | 0.0000069 deg | 0.12 um |
| 21 | 2,097,152 | 0.00017 | 0.0000017 deg | 0.03 um |
| 23 | 8,388,608 | 0.000043 | 0.00000043 deg | 0.007 um |
Compare the right-hand column with a typical robot repeatability of 0.02 mm, which is 20 micrometres. Even a 13-bit encoder resolves nearly three times finer than that. The reason manufacturers fit 17 to 23 bit devices is not tool resolution; it is smooth velocity control at low speed, where a coarse encoder produces visible stepping and audible cogging.
What actually limits the joint
| Source | Typical magnitude | At 1 m |
|---|---|---|
| Encoder resolution, 17 bit behind 100:1 | 0.0000275 deg | 0.0005 mm |
| Encoder accuracy error | 0.005 to 0.05 deg | 0.09 to 0.9 mm |
| Gearbox backlash | 0.5 to 3 arcmin | 0.15 to 0.9 mm |
| Gearbox transmission error | 0.3 to 2 arcmin | 0.09 to 0.6 mm |
| Structural deflection under load | varies | 0.05 to 1.5 mm |
| Thermal expansion, 10 K | ≈ 12 um/m/10 K | 0.12 mm |
Read the first row against the rest. Encoder resolution contributes roughly a thousandth of what backlash contributes. This is why paying for a higher-resolution encoder to improve accuracy is misdirected spending, and why secondary encoders on the output shaft, which measure past the gearbox, deliver far more.
Why a second encoder is worth more than more bits
A motor-side encoder measures the motor shaft, before the gearbox. Everything the gearbox does wrong happens downstream of it and is invisible. A second encoder on the output shaft measures where the joint actually is, which changes what the controller can correct.
| Property | Motor side only | Dual encoder |
|---|---|---|
| Sees gearbox backlash | No | Yes |
| Sees transmission error | No | Yes |
| Sees structural deflection at the joint | No | Partly |
| Typical accuracy improvement | baseline | 2 to 5x |
| Added cost per joint | 0 | significant |
| Control complexity | simple | two-loop, tuning required |
| Typical use | General industrial arms | Precision arms, collaborative joints |
The cost is real, which is why dual encoders appear on precision and collaborative arms rather than on every machine. Where they are fitted they also enable joint torque estimation from the deflection of a known compliance, which is how several collaborative designs sense contact without a dedicated force sensor.
Absolute against incremental
- Absolute encoders report position directly at power-on, so the robot knows where it is without homing. Multi-turn versions track revolutions across power cycles, historically with a backup battery and increasingly through a geared or energy-harvesting mechanism.
- Incremental encoders count from an unknown starting point and require a homing move. Cheaper, and unacceptable on any arm where an unexpected motion at startup is a hazard.
- Battery-backed multi-turn devices introduce a maintenance item: a flat battery loses the position and requires remastering, which on a production cell is downtime plus a calibration procedure.
Nearly all modern industrial arms use absolute multi-turn encoders for exactly the safety and downtime reasons above.
Frequently asked questions
What encoder resolution does a robot joint need?
Industrial arms use 17 to 23 bit absolute encoders, but not for tool resolution. Even 13 bit behind a 100:1 gearbox resolves under 8 micrometres at 1 m. The higher resolution buys smooth low-speed velocity control.
Does a higher-resolution encoder make a robot more accurate?
Almost never. Gearbox backlash and transmission error contribute roughly a thousand times more than encoder resolution. A secondary encoder on the output shaft, measuring past the gearbox, helps far more.
What is the difference between resolution and accuracy?
Resolution is the smallest change the device can report; accuracy is how close the reading is to the true position. An encoder can have 23-bit resolution and an arcminute of accuracy error at the same time.
Why do robots use absolute encoders?
Because they report position at power-on without a homing move, which removes an unexpected startup motion and its associated hazard, and avoids production downtime after every power interruption.
What happens when the encoder battery dies?
The multi-turn position count is lost and the robot must be remastered, which is downtime plus a calibration procedure. Battery replacement intervals belong in the planned maintenance schedule rather than in the fault log.
Sources
- ISO 9283, manipulating industrial robots, performance criteria and test methodsRepeatability and accuracy definitions used for the comparison
- ISO 8373, robotics vocabularyInternational Organization for Standardization, joint and axis terminology
- Robotics at NISTNational Institute of Standards and Technology, robot performance measurement and test methods