---
title: "Measured actuator benchmarks"
description: "Measured robot actuator performance next to datasheet claims: torque-speed curves, thermal derating, backlash and friction from vendor test reports, papers and blogs, each with its method and uncertainty."
canonical: https://actuatorlist.com/benchmarks
last-updated: 2026-11-03T08:48:14.000Z
---
# Measured actuator benchmarks

Datasheets are claims; these are measurements. 21 runs on 16 catalog actuators (21 published, 0 lab, 0 community), as of 2026-11-03. Each run states its rig, how the numbers were taken (raw rig data, a table, a plot digitised by us, or figures quoted in the text) and an uncertainty per metric. A digitised plot is never raw data, and vendor test reports are the maker's own measurements, not independent ones. Interactive charts: https://actuatorlist.com/benchmarks. JSON: https://actuatorlist.com/api/v1/benchmarks. MCP tool: get_benchmarks.

## Datasheet vs measured

Highest measured peak torque per actuator against the catalog's datasheet peak, biggest shortfall first. A curve that stops below the stall rating counts as a shortfall: the run never saw more.

| Actuator | Datasheet peak | Measured peak | Measured vs datasheet | Run | Origin |
| --- | --- | --- | --- | --- | --- |
| [XC430-W150-T](https://actuatorlist.com/actuators/dynamixel-xc430-w150-t) | 1.6 N·m | 0.74 N·m | -54% | [robotis-performance-graph](https://actuatorlist.com/benchmarks#run=dynamixel-xc430-w150-t/robotis-performance-graph) | Published (vendor) |
| [XL430-W250-T](https://actuatorlist.com/actuators/dynamixel-xl430-w250-t) | 1.5 N·m | 0.84 N·m | -44% | [robotis-performance-graph](https://actuatorlist.com/benchmarks#run=dynamixel-xl430-w250-t/robotis-performance-graph) | Published (vendor) |
| [XM430-W350-T](https://actuatorlist.com/actuators/dynamixel-xm430-w350-t) | 4.1 N·m | 2.98 N·m | -27% | [robotis-performance-graph](https://actuatorlist.com/benchmarks#run=dynamixel-xm430-w350-t/robotis-performance-graph) | Published (vendor) |
| [XM540-W270-T](https://actuatorlist.com/actuators/dynamixel-xm540-w270-t) | 10.6 N·m | 8.6 N·m | -19% | [robotis-performance-graph](https://actuatorlist.com/benchmarks#run=dynamixel-xm540-w270-t/robotis-performance-graph) | Published (vendor) |
| [XH540-W270-T](https://actuatorlist.com/actuators/dynamixel-xh540-w270-t) | 9.9 N·m | 8.1 N·m | -18% | [robotis-performance-graph](https://actuatorlist.com/benchmarks#run=dynamixel-xh540-w270-t/robotis-performance-graph) | Published (vendor) |
| [STS3250 (12V, 50 kg.cm)](https://actuatorlist.com/actuators/feetech-sts3250) | 4.9 N·m | 4.71 N·m | -4% | [robonine-servo-test](https://actuatorlist.com/benchmarks#run=feetech-sts3250/robonine-servo-test) | Published (blog) |
| [qdd100 beta 3](https://actuatorlist.com/actuators/mjbots-qdd100) | 16 N·m | 16 N·m | matches | [mjbots-beta3-spec](https://actuatorlist.com/benchmarks#run=mjbots-qdd100/mjbots-beta3-spec) | Published (vendor) |
| [RobStride 00](https://actuatorlist.com/actuators/robstride-00) | 14 N·m | 14 N·m | matches | [robstride-manual-2026-07](https://actuatorlist.com/benchmarks#run=robstride-00/robstride-manual-2026-07) | Published (vendor) |
| [RobStride 02](https://actuatorlist.com/actuators/robstride-02) | 17 N·m | 17 N·m | matches | [robstride-manual-2026-07](https://actuatorlist.com/benchmarks#run=robstride-02/robstride-manual-2026-07) | Published (vendor) |
| [RobStride 03](https://actuatorlist.com/actuators/robstride-03) | 60 N·m | 60 N·m | matches | [robstride-manual-2026-10](https://actuatorlist.com/benchmarks#run=robstride-03/robstride-manual-2026-10) | Published (vendor) |
| [RobStride 04](https://actuatorlist.com/actuators/robstride-04) | 120 N·m | 120 N·m | matches | [robstride-manual-2026-07](https://actuatorlist.com/benchmarks#run=robstride-04/robstride-manual-2026-07) | Published (vendor) |
| [CyberGear Micromotor](https://actuatorlist.com/actuators/xiaomi-cybergear) | 12 N·m | 12 N·m | matches | [xiaomi-manual-2023](https://actuatorlist.com/benchmarks#run=xiaomi-cybergear/xiaomi-manual-2023) | Published (vendor) |
| [STS3215 (12V, 30 kg.cm)](https://actuatorlist.com/actuators/feetech-sts3215) | 2.94 N·m | 3.43 N·m | +17% | [robonine-servo-test](https://actuatorlist.com/benchmarks#run=feetech-sts3215/robonine-servo-test) | Published (blog) |

## Measured leaderboard

Continuous torque density: the highest torque a run held for 10 minutes or without reaching the thermal limit, per kg of catalog weight. Heat sinks and ambient differ between runs; read the run before ranking two close entries.

| Rank | Actuator | Continuous torque | Weight | Measured N·m/kg | Run |
| --- | --- | --- | --- | --- | --- |
| 1 | [RobStride 04](https://actuatorlist.com/actuators/robstride-04) | 40 N·m | 1420 g | 28.2 | [robstride-manual-2026-07](https://actuatorlist.com/benchmarks#run=robstride-04/robstride-manual-2026-07) |
| 2 | [RobStride 02](https://actuatorlist.com/actuators/robstride-02) | 9 N·m | 380 g | 23.7 | [robstride-manual-2026-07](https://actuatorlist.com/benchmarks#run=robstride-02/robstride-manual-2026-07) |
| 3 | [RobStride 03](https://actuatorlist.com/actuators/robstride-03) | 20 N·m | 880 g | 22.7 | [robstride-manual-2026-10](https://actuatorlist.com/benchmarks#run=robstride-03/robstride-manual-2026-10) |
| 4 | [CyberGear Micromotor](https://actuatorlist.com/actuators/xiaomi-cybergear) | 6 N·m | 317 g | 18.9 | [xiaomi-manual-2023](https://actuatorlist.com/benchmarks#run=xiaomi-cybergear/xiaomi-manual-2023) |
| 5 | [RobStride 00](https://actuatorlist.com/actuators/robstride-00) | 5 N·m | 310 g | 16.1 | [robstride-manual-2026-07](https://actuatorlist.com/benchmarks#run=robstride-00/robstride-manual-2026-07) |
| 6 | [qdd100 beta 3](https://actuatorlist.com/actuators/mjbots-qdd100) | 3.3 N·m | 507 g | 6.51 | [mjbots-beta3-spec](https://actuatorlist.com/benchmarks#run=mjbots-qdd100/mjbots-beta3-spec) |

## Torque-speed curves

| Actuator | Bus | Method | Points (rpm: N·m) | Uncertainty | Peak efficiency | Run |
| --- | --- | --- | --- | --- | --- | --- |
| XC430-W150-T | 12 V | digitised | 7 rpm: 0.74; 22 rpm: 0.7; 40.5 rpm: 0.6; 52 rpm: 0.5; 61.5 rpm: 0.4; 72.5 rpm: 0.3; 82 rpm: 0.2; 91 rpm: 0.09 | ±0.02 N·m | 46% | [robotis-performance-graph](https://actuatorlist.com/benchmarks#run=dynamixel-xc430-w150-t/robotis-performance-graph) (Published (vendor)) |
| XH540-W270-T | 12 V | digitised | 4.5 rpm: 8.1; 7.7 rpm: 7.2; 11.5 rpm: 6.3; 16.5 rpm: 5.4; 21.5 rpm: 4.5; 24.8 rpm: 3.6; 28.3 rpm: 2.7; 31.4 rpm: 1.8; 36.3 rpm: 0.36 | ±0.1 N·m | 48% | [robotis-performance-graph](https://actuatorlist.com/benchmarks#run=dynamixel-xh540-w270-t/robotis-performance-graph) (Published (vendor)) |
| XL430-W250-T | 11.1 V | digitised | 3 rpm: 0.84; 10 rpm: 0.72; 16.5 rpm: 0.6; 23.5 rpm: 0.48; 30 rpm: 0.36; 37 rpm: 0.24; 47 rpm: 0.06 | ±0.02 N·m | 23% | [robotis-performance-graph](https://actuatorlist.com/benchmarks#run=dynamixel-xl430-w250-t/robotis-performance-graph) (Published (vendor)) |
| XM430-W350-T | 12 V | digitised | 5.5 rpm: 2.98; 9 rpm: 2.8; 16 rpm: 2.45; 22.5 rpm: 2.1; 28.5 rpm: 1.75; 33.5 rpm: 1.4; 37.5 rpm: 1.05; 42 rpm: 0.7; 45.5 rpm: 0.35; 48 rpm: 0.05 | ±0.05 N·m | 45% | [robotis-performance-graph](https://actuatorlist.com/benchmarks#run=dynamixel-xm430-w350-t/robotis-performance-graph) (Published (vendor)) |
| XM540-W270-T | 12 V | digitised | 1.5 rpm: 8.6; 4.5 rpm: 8; 9.5 rpm: 7; 13.5 rpm: 6; 17 rpm: 5; 20.5 rpm: 4; 23.5 rpm: 3; 26 rpm: 2; 28 rpm: 1; 29.5 rpm: 0.4 | ±0.1 N·m | 46% | [robotis-performance-graph](https://actuatorlist.com/benchmarks#run=dynamixel-xm540-w270-t/robotis-performance-graph) (Published (vendor)) |
| RobStride 00 | 48 V | digitised | 100 rpm: 14; 150 rpm: 12; 200 rpm: 9.5; 225 rpm: 8; 250 rpm: 6.6; 275 rpm: 4.5; 300 rpm: 2; 315 rpm: 0.5 | ±0.3 N·m | n/a | [robstride-manual-2026-07](https://actuatorlist.com/benchmarks#run=robstride-00/robstride-manual-2026-07) (Published (vendor)) |
| RobStride 01 | 48 V | digitised | 220 rpm: 17; 250 rpm: 15.4; 275 rpm: 14; 300 rpm: 12; 325 rpm: 10; 350 rpm: 7.8; 375 rpm: 4.6; 400 rpm: 1.4; 407 rpm: 0.5 | ±0.3 N·m | n/a | [robstride-manual-2026-07](https://actuatorlist.com/benchmarks#run=robstride-01/robstride-manual-2026-07) (Published (vendor), not ranked) |
| RobStride 02 | 48 V | digitised | 220 rpm: 17; 250 rpm: 15.4; 275 rpm: 14; 300 rpm: 12; 325 rpm: 10; 350 rpm: 7.8; 375 rpm: 4.8; 400 rpm: 1.5; 408 rpm: 0.5 | ±0.3 N·m | n/a | [robstride-manual-2026-07](https://actuatorlist.com/benchmarks#run=robstride-02/robstride-manual-2026-07) (Published (vendor)) |
| RobStride 03 | 48 V | digitised | 120 rpm: 60; 135 rpm: 54.5; 150 rpm: 47.5; 165 rpm: 38.5; 180 rpm: 28.5; 186 rpm: 20; 187 rpm: 10 | ±1 N·m | n/a | [robstride-manual-2026-10](https://actuatorlist.com/benchmarks#run=robstride-03/robstride-manual-2026-10) (Published (vendor)) |
| RobStride 04 | 48 V | digitised | 95 rpm: 120; 110 rpm: 109; 130 rpm: 96; 150 rpm: 79; 155 rpm: 70; 170 rpm: 53; 180 rpm: 47; 186 rpm: 38; 190 rpm: 10 | ±2 N·m | n/a | [robstride-manual-2026-07](https://actuatorlist.com/benchmarks#run=robstride-04/robstride-manual-2026-07) (Published (vendor)) |
| CyberGear Micromotor | 24 V | digitised | 40 rpm: 12; 110 rpm: 12; 130 rpm: 11.3; 150 rpm: 10.4; 170 rpm: 9.2; 190 rpm: 7.8; 210 rpm: 6.2; 230 rpm: 4.7; 250 rpm: 3; 270 rpm: 1.4; 280 rpm: 0.65 | ±0.2 N·m | n/a | [xiaomi-manual-2023](https://actuatorlist.com/benchmarks#run=xiaomi-cybergear/xiaomi-manual-2023) (Published (vendor)) |

## Thermal derating

| Actuator | Ambient | Torque: time to limit | Uncertainty | Continuous | Run |
| --- | --- | --- | --- | --- | --- |
| qdd100 beta 3 | 20 °C | 16 N·m: 1 s; 10 N·m: 60 s; 6 N·m: 400 s; 3.3 N·m: continuous | ±0.5 N·m | 3.3 N·m | [mjbots-beta3-spec](https://actuatorlist.com/benchmarks#run=mjbots-qdd100/mjbots-beta3-spec) (Published (vendor)) |
| qdd100 beta 3 | 20 °C | 12.5 N·m: 5 s; 10 N·m: 30 s; 8 N·m: 120 s; 6 N·m: 300 s; 4 N·m: 800 s; 2.5 N·m: continuous | ±0.5 N·m | 4 N·m | [mjbots-blog-2020-08](https://actuatorlist.com/benchmarks#run=mjbots-qdd100/mjbots-blog-2020-08) (Published (vendor), other revision) |
| RobStride 00 | 25 °C | 14 N·m: 3 s; 13 N·m: 7 s; 12 N·m: 10 s; 10 N·m: 20 s; 8 N·m: 53 s; 7 N·m: 94 s; 6 N·m: 285 s; 5 N·m: continuous | ±0.5 N·m | 5 N·m | [robstride-manual-2026-07](https://actuatorlist.com/benchmarks#run=robstride-00/robstride-manual-2026-07) (Published (vendor)) |
| RobStride 01 | 25 °C | 17 N·m: 10 s; 16 N·m: 14 s; 15 N·m: 18 s; 14 N·m: 29 s; 13 N·m: 42 s; 12 N·m: 71 s; 11 N·m: 165 s; 10 N·m: 250 s; 9 N·m: 838 s; 7 N·m: continuous | ±0.5 N·m | 9 N·m | [robstride-manual-2026-07](https://actuatorlist.com/benchmarks#run=robstride-01/robstride-manual-2026-07) (Published (vendor), not ranked) |
| RobStride 02 | 25 °C | 17 N·m: 10 s; 16 N·m: 14 s; 15 N·m: 18 s; 14 N·m: 29 s; 13 N·m: 42 s; 12 N·m: 71 s; 11 N·m: 165 s; 10 N·m: 250 s; 9 N·m: 838 s; 7 N·m: continuous | ±0.5 N·m | 9 N·m | [robstride-manual-2026-07](https://actuatorlist.com/benchmarks#run=robstride-02/robstride-manual-2026-07) (Published (vendor)) |
| RobStride 03 | 25 °C | 60 N·m: 13 s; 50 N·m: 34 s; 40 N·m: 67 s; 30 N·m: 393 s; 20 N·m: continuous | ±0.5 N·m | 20 N·m | [robstride-manual-2026-10](https://actuatorlist.com/benchmarks#run=robstride-03/robstride-manual-2026-10) (Published (vendor)) |
| RobStride 04 | 25 °C | 120 N·m: 3 s; 115 N·m: 9 s; 110 N·m: 10 s; 105 N·m: 13 s; 100 N·m: 15 s; 90 N·m: 24 s; 80 N·m: 38 s; 70 N·m: 61 s; 60 N·m: 108 s; 50 N·m: 324 s; 40 N·m: continuous | ±0.5 N·m | 40 N·m | [robstride-manual-2026-07](https://actuatorlist.com/benchmarks#run=robstride-04/robstride-manual-2026-07) (Published (vendor)) |
| CyberGear Micromotor | 25 °C | 12 N·m: 28 s; 11 N·m: 45 s; 10 N·m: 60 s; 9 N·m: 90 s; 8 N·m: 160 s; 7 N·m: 320 s; 6 N·m: 700 s; 5 N·m: 1800 s; 4.5 N·m: 2500 s; 4 N·m: continuous | ±0.25 N·m | 6 N·m | [xiaomi-manual-2023](https://actuatorlist.com/benchmarks#run=xiaomi-cybergear/xiaomi-manual-2023) (Published (vendor)) |

## All runs

| Run | Actuator | Date | Origin | Method | Groups | Source |
| --- | --- | --- | --- | --- | --- | --- |
| `cubemars-ak80-9/staccatoe-2024` | [AK80-9 (V3.0 KV100)](https://actuatorlist.com/actuators/cubemars-ak80-9) | 2024-04 | Published (paper) | reported | Transmission | [arXiv 2404.05039: StaccaToe, a single-leg robot that mimics the human leg and toe](https://arxiv.org/abs/2404.05039) |
| `cubemars-ak80-9/orthosis-backdrive-2021` | [AK80-9 (V3.0 KV100)](https://actuatorlist.com/actuators/cubemars-ak80-9) | 2021-10 | Published (paper), other revision | reported | Transmission | [arXiv 2110.01562: Enhancing voluntary motion with modular, backdrivable, powered hip and knee orthoses](https://arxiv.org/abs/2110.01562) |
| `dynamixel-xc430-w150-t/robotis-performance-graph` | [XC430-W150-T](https://actuatorlist.com/actuators/dynamixel-xc430-w150-t) | 2026-11 | Published (vendor) | digitised | Torque-speed | [ROBOTIS e-Manual: XC430-W150 performance graph](https://emanual.robotis.com/docs/en/dxl/x/xc430-w150/) |
| `dynamixel-xh540-w270-t/robotis-performance-graph` | [XH540-W270-T](https://actuatorlist.com/actuators/dynamixel-xh540-w270-t) | 2026-11 | Published (vendor) | digitised | Torque-speed | [ROBOTIS e-Manual: XH540-W270 performance graph](https://emanual.robotis.com/docs/en/dxl/x/xh540-w270/) |
| `dynamixel-xl430-w250-t/robotis-performance-graph` | [XL430-W250-T](https://actuatorlist.com/actuators/dynamixel-xl430-w250-t) | 2026-11 | Published (vendor) | digitised | Torque-speed | [ROBOTIS e-Manual: XL430-W250 performance graph](https://emanual.robotis.com/docs/en/dxl/x/xl430-w250/) |
| `dynamixel-xm430-w350-t/robotis-performance-graph` | [XM430-W350-T](https://actuatorlist.com/actuators/dynamixel-xm430-w350-t) | 2026-11 | Published (vendor) | digitised | Torque-speed | [ROBOTIS e-Manual: XM430-W350 performance graph](https://emanual.robotis.com/docs/en/dxl/x/xm430-w350/) |
| `dynamixel-xm540-w270-t/robotis-performance-graph` | [XM540-W270-T](https://actuatorlist.com/actuators/dynamixel-xm540-w270-t) | 2026-11 | Published (vendor) | digitised | Torque-speed | [ROBOTIS e-Manual: XM540-W270 performance graph](https://emanual.robotis.com/docs/en/dxl/x/xm540-w270/) |
| `feetech-sts3215/robonine-servo-comparison` | [STS3215 (12V, 30 kg.cm)](https://actuatorlist.com/actuators/feetech-sts3215) | 2026-11 | Published (blog) | reported | Torque-speed, Transmission | [Robonine: Practical comparison of smart serial servo motors for robotics](https://robonine.com/smart-serial-servo-motor-comparison-robotics/) |
| `feetech-sts3215/robonine-servo-test` | [STS3215 (12V, 30 kg.cm)](https://actuatorlist.com/actuators/feetech-sts3215) | 2026-11 | Published (blog) | reported | Torque-speed, Control | [Robonine: Testing of Feetech STS3215 servomotor: backlash, repeatability and torque](https://robonine.com/testing-of-feetech-sts3215-servomotor-backlash-repeatability-and-torque/) |
| `feetech-sts3215/hardwarex-backlash-2026` | [STS3215 (12V, 30 kg.cm)](https://actuatorlist.com/actuators/feetech-sts3215) | 2026 | Published (paper) | reported | Transmission | [Kotov, HardwareX 26 (2026) e00767: An open-source test stand for backlash measurement in low-cost UART servo motors](https://pmc.ncbi.nlm.nih.gov/articles/PMC13087586/) |
| `feetech-sts3250/robonine-servo-test` | [STS3250 (12V, 50 kg.cm)](https://actuatorlist.com/actuators/feetech-sts3250) | 2026-11 | Published (blog) | reported | Torque-speed | [Robonine: Feetech STS3250 smart actuator, evaluation of accuracy, torque and backlash](https://robonine.com/feetech-sts3250-smart-actuator-evaluation-of-accuracy-torque-and-backlash/) |
| `mjbots-qdd100/mjbots-beta3-spec` | [qdd100 beta 3](https://actuatorlist.com/actuators/mjbots-qdd100) | 2026-11 | Published (vendor) | tabulated | Torque-speed, Thermal | [mjbots: qdd100 beta 3 servo product page](https://mjbots.com/products/qdd100-beta-3) |
| `mjbots-qdd100/mjbots-blog-2020-08` | [qdd100 beta 3](https://actuatorlist.com/actuators/mjbots-qdd100) | 2020-08-07 | Published (vendor), other revision | reported | Thermal | [mjbots blog: Up-rating the qdd100 beta thermal bounds](https://blog.mjbots.com/2020/08/07/up-rating-the-qdd100-beta-thermal-bounds/) |
| `mjbots-qdd100/mjbots-blog-2020-03` | [qdd100 beta 3](https://actuatorlist.com/actuators/mjbots-qdd100) | 2020-03-25 | Published (vendor), other revision | reported | Torque-speed | [mjbots blog: Ground truth torque testing for qdd100](https://blog.mjbots.com/2020/03/25/ground-truth-torque-testing-for-qdd100/) |
| `robstride-00/robstride-manual-2026-07` | [RobStride 00](https://actuatorlist.com/actuators/robstride-00) | 2026-07 | Published (vendor) | digitised | Torque-speed, Thermal | [RobStride RS00 user manual (260713)](https://github.com/RobStride/Product_Information/blob/main/Product%20Literature/CAN2.0/RS00/RS00User%20Manual260713.pdf) |
| `robstride-01/robstride-manual-2026-07` | [RobStride 01](https://actuatorlist.com/actuators/robstride-01) | 2026-07 | Published (vendor), not ranked | digitised | Torque-speed, Thermal | [RobStride RS01 user manual (260713)](https://github.com/RobStride/Product_Information/blob/main/Product%20Literature/CAN2.0/RS01/RS01User%20Manual260713.pdf) |
| `robstride-02/robstride-manual-2026-07` | [RobStride 02](https://actuatorlist.com/actuators/robstride-02) | 2026-07 | Published (vendor) | digitised | Torque-speed, Thermal | [RobStride RS02 user manual (260713)](https://github.com/RobStride/Product_Information/blob/main/Product%20Literature/CAN2.0/RS02/RS02User%20Manual260713.pdf) |
| `robstride-03/robstride-manual-2026-10` | [RobStride 03](https://actuatorlist.com/actuators/robstride-03) | 2026-10 | Published (vendor) | digitised | Torque-speed, Thermal | [RobStride RS03 user manual (261013)](https://github.com/RobStride/Product_Information/blob/main/Product%20Literature/CAN2.0/RS03/RS03User%20Manual261013.pdf) |
| `robstride-04/robstride-manual-2026-07` | [RobStride 04](https://actuatorlist.com/actuators/robstride-04) | 2026-07 | Published (vendor) | digitised | Torque-speed, Thermal | [RobStride RS04 user manual (260713)](https://github.com/RobStride/Product_Information/blob/main/Product%20Literature/CAN2.0/RS04/RS04User%20Manual260713.pdf) |
| `steadywin-gim6010-8/hackaday-spider-bot-2025` | [GIM6010-8](https://actuatorlist.com/actuators/steadywin-gim6010-8) | 2025-03 | Published (blog), not ranked | reported | Torque-speed | [Hackaday.io: Steadywin GIM6010-8 vs Xiaomi CyberGear micromotor](https://hackaday.io/project/202366-the-unnamed-spider-bot/log/239239-steadywin-gim6010-8-vs-xiaomi-cybergear-micromotor) |
| `xiaomi-cybergear/xiaomi-manual-2023` | [CyberGear Micromotor](https://actuatorlist.com/actuators/xiaomi-cybergear) | 2023 | Published (vendor) | digitised | Torque-speed, Thermal | [Xiaomi CyberGear micromotor instruction manual (English translation)](https://github.com/belovictor/cybergear-docs/blob/main/instructionmanual/instructionmanual.md) |

## Run notes

### cubemars-ak80-9/staccatoe-2024

Measured: Backlash 9 arcmin (±1).

Method: Backlash measured at the knee actuator output on the assembled leg; torque constants on the authors' dynamometer. The paper states no uncertainty for the backlash: ±1 arcmin is our estimate (rounding the quoted 0.15° alone accounts for ±0.3 arcmin).

Revision: not stated

Note: 0.15° backlash at the knee output. The same paper found the measured torque constants of its AK actuators below the specified values.

### cubemars-ak80-9/orthosis-backdrive-2021

Measured: Backdrive friction 0.25 N·m (±0.25).

Method: Static backdrive: an analog torque wrench turns the unpowered output until it moves. Dynamic: the output driven by hand at 1 and 2 Hz over 70°.

Revision: not stated; predates the V3.0 KV100 in the catalog

Note: Reported only as under 0.5 N·m static, so it is entered as 0.25 ± 0.25 N·m to cover 0 to 0.5 N·m. Dynamic backdrive stayed under 2 N·m at 1 to 2 Hz, rising with frequency from inertia.

### dynamixel-xc430-w150-t/robotis-performance-graph

Measured: T-N curve with 8 points at 12 V (±0.02 N·m).

Method: Speed and efficiency read off the e-Manual's N-T graph (about 0.02 N·m reading error). The graph does not print its voltage; the 12 V rating is assumed. The reading error is stated for torque only; our reading errors for speed and efficiency are not quantified.

Note: The graph stops at 0.74 N·m, under half of the 1.6 N·m stall figure. Efficiency near stall reads unreliably: the 0.05 at 7 rpm implies less electrical input than the 22 rpm point at lower torque, which is unlikely. We could not re-check the graph, so treat the lowest-speed efficiency as approximate.

### dynamixel-xh540-w270-t/robotis-performance-graph

Measured: T-N curve with 9 points at 12 V (±0.1 N·m).

Method: Speed and efficiency read off the e-Manual's N-T graph (about 0.1 N·m reading error). The graph does not print its voltage; the 12 V rating is assumed. The reading error is stated for torque only; our reading errors for speed and efficiency are not quantified.

Note: Same frame as the XM540-W270 with a different motor; the graph reaches 8.1 N·m against a 9.9 N·m stall figure.

### dynamixel-xl430-w250-t/robotis-performance-graph

Measured: T-N curve with 7 points at 11.1 V (±0.02 N·m).

Method: Speed and efficiency read off the e-Manual's N-T graph (about 0.02 N·m reading error). The graph does not print its voltage; the 11.1 V rating is assumed. The reading error is stated for torque only; our reading errors for speed and efficiency are not quantified.

Note: The graph ends at 0.84 N·m, a little over half the 1.5 N·m stall figure. Efficiency near stall reads unreliably: the 0.05 at 3 rpm implies less electrical input than the 10 rpm point at lower torque, which is unlikely. We could not re-check the graph, so treat the lowest-speed efficiency as approximate.

### dynamixel-xm430-w350-t/robotis-performance-graph

Measured: T-N curve with 10 points at 12 V (±0.05 N·m).

Method: Speed and efficiency read off the e-Manual's N-T graph, measured by ROBOTIS under a gradually increasing load (about 0.05 N·m reading error). The graph does not print its voltage; the 12 V rating is assumed. The reading error is stated for torque only; our reading errors for speed and efficiency are not quantified.

Note: ROBOTIS says the graph, not the stall rating, is what to expect in use. The graph tops out near 3 N·m against a 4.1 N·m stall rating.

### dynamixel-xm540-w270-t/robotis-performance-graph

Measured: T-N curve with 10 points at 12 V (±0.1 N·m).

Method: Speed and efficiency read off the e-Manual's N-T graph (about 0.1 N·m reading error). The graph does not print its voltage; the 12 V rating is assumed. The reading error is stated for torque only; our reading errors for speed and efficiency are not quantified.

Note: Peak efficiency is about 46% near 2 N·m and falls under 10% at the top of the curve. Efficiency near stall reads unreliably: the 0.06 at 1.5 rpm implies less electrical input than the 4.5 rpm point at lower torque, which is unlikely. We could not re-check the graph, so treat the lowest-speed efficiencies as approximate.

### feetech-sts3215/robonine-servo-comparison

Measured: Peak torque 3.24 N·m (±0.2); Backlash 46.1 arcmin (±5.3).

Method: Peak torque from a 3.3 kgf reading on a 10 cm lever at full load. Backlash is the post's mean of 8.75 encoder counts, converted by us at about 0.088° per count; the post does not say how many readings the mean uses. The post states no uncertainties: ±5.3 arcmin is one encoder count (reading resolution, not the spread of the readings) and ±0.2 N·m is our estimate for the scale reading.

Revision: 12 V, 1:345 (C018)

Note: Largest backlash of the five servos compared, above the 0.5° the datasheet gives.

### feetech-sts3215/robonine-servo-test

Measured: Peak torque 3.43 N·m (±0.2); Repeatability 10.2 arcmin (±2).

Method: Static load of about 3.5 kg on a 10 cm lever read on a scale, and repeated cycles to a fixed position read on the lever. The post carries no date, so the run is dated when it was read. The post states no uncertainties: ±0.2 N·m and ±2 arcmin are our estimates, not figures from the post.

Revision: 12 V, 1:345 (C018)

Note: Stall slightly above the 30 kg·cm rating, but long holds tripped overload protection, which cut output to about a fifth.

### feetech-sts3215/hardwarex-backlash-2026

Measured: Backlash 37.2 arcmin (±5.3).

Method: Two STS3215 servos at default settings pull a 100 mm lever both ways; backlash is read on the servo's own output encoder, unloaded and under load. backlashArcmin holds the unloaded figure; the loaded one is in the note, and neither load is the protocol's ±5% of rated torque. ±5.3 arcmin is one count of the 4096-count encoder (0.088°), our reading resolution, not an uncertainty stated by the paper.

Note: Unloaded backlash about 0.62°; loaded about 1.30° (78 arcmin). Coupling two servos brought it to 0.47° to 0.67°.

### feetech-sts3250/robonine-servo-test

Measured: Peak torque 4.71 N·m (±0.5).

Method: Torque read on a lever: the peak held for a split second, and the highest load held before overload protection cut in. The post gives no lever length, scale accuracy or repeat count: ±0.5 N·m is our estimate, about 10% of the peak.

Note: 48 kg·cm peak against a 50 kg·cm rating; 25 kg·cm (2.45 N·m) was the most it held before protection cut in.

### mjbots-qdd100/mjbots-beta3-spec

Measured: Peak torque 16 N·m (±0.5); thermal derating with 4 points (±0.5 N·m).

Method: Torque against time before thermal limiting, measured by mjbots at zero speed, 24 V, 20 °C ambient, no airflow or heat sink. The page gives the first point as under 1 s; it is entered as 1 s. The page states no uncertainty: ±0.5 N·m on the peak and on the thermal torques is our estimate, not a figure from mjbots.

Note: Vendor measurement on the beta 3 revision. The page carries no date, so the run is dated when it was read.

### mjbots-qdd100/mjbots-blog-2020-08

Measured: thermal derating with 6 points (±0.5 N·m).

Method: Torque held until thermal throttling in a 20 °C lab with no forced airflow. The post gives the 12.5 N·m point as under 5 s; it is entered as 5 s. The post states no uncertainty: ±0.5 N·m is our estimate, not a figure from mjbots.

Revision: original qdd100 beta, not beta 3

Note: Earlier hardware than the catalog's beta 3, kept because it is the developer's own test with the method written up.

### mjbots-qdd100/mjbots-blog-2020-03

Measured: Peak torque 13.8 N·m (±1.25).

Method: A 1 m pole on the output pressed on a digital scale. Peak torque landed between 12.5 and 15 N·m depending on the unit; the run enters the mid-point, 13.75 N·m, with half that range as its uncertainty.

Revision: original qdd100 beta, not beta 3

Note: The test also found the torque constant below the value estimated from Kv and falling off at high current.

### robstride-00/robstride-manual-2026-07

Measured: T-N curve with 8 points at 48 V (±0.3 N·m); thermal derating with 8 points (±0.5 N·m).

Method: T-N points read off the manual's 48 V chart (about 0.3 N·m reading error). Overload times are the manual's table at 100 rpm on a 90 x 85 mm aluminium plate, 145 °C winding limit. Thermal torques are the table's whole-N·m figures and the manual states no measurement uncertainty, so the ±0.5 N·m is that rounding only. windingRiseC is the rise from 25 °C ambient to the winding limit, entered only for points that reached it.

Note: Vendor test data. The manual adds a stall table: with the rotor locked one phase carries the heat, so times are shorter than these.

### robstride-01/robstride-manual-2026-07

Measured: T-N curve with 9 points at 48 V (±0.3 N·m); thermal derating with 10 points (±0.5 N·m).

Method: T-N points read off the manual's chart (about 0.3 N·m reading error). Overload times are the manual's table at 100 rpm on a 260 x 280 mm aluminium plate, 135 °C winding limit. Thermal torques are the table's whole-N·m figures and the manual states no measurement uncertainty, so the ±0.5 N·m is that rounding only. windingRiseC is the rise from 25 °C ambient to the winding limit, entered only for points that reached it.

Not ranked: The curve and overload table match the RS02 manual entry for entry, so they may not have been measured on an RS01. Shown, but not ranked until RobStride confirms.

Note: The RS01 manual's overload table and T-N chart match the RS02 manual entry for entry, while the two manuals give different winding limits (135 and 145 °C). Treat this run as not measured on an RS01 until RobStride says otherwise.

### robstride-02/robstride-manual-2026-07

Measured: T-N curve with 9 points at 48 V (±0.3 N·m); thermal derating with 10 points (±0.5 N·m).

Method: T-N points read off the manual's 48 V chart (about 0.3 N·m reading error); the manual also has a 36 V chart. Overload times are the manual's table at 100 rpm on a 260 x 280 mm aluminium plate, 145 °C winding limit. Thermal torques are the table's whole-N·m figures and the manual states no measurement uncertainty, so the ±0.5 N·m is that rounding only. windingRiseC is the rise from 25 °C ambient to the winding limit, entered only for points that reached it.

Note: Vendor test data on a large heat sink. The manual rates 6 N·m continuous, below the 7 N·m its own overload table holds without limit.

### robstride-03/robstride-manual-2026-10

Measured: T-N curve with 7 points at 48 V (±1 N·m); thermal derating with 5 points (±0.5 N·m).

Method: T-N points read off the manual's chart (about 1 N·m reading error). Overload times are the manual's table at 100 rpm on a 215 x 220 mm aluminium plate, 145 °C winding limit. Thermal torques are the table's whole-N·m figures and the manual states no measurement uncertainty, so the ±0.5 N·m is that rounding only. windingRiseC is the rise from 25 °C ambient to the winding limit, entered only for points that reached it.

Note: Vendor test data. The chart stops near 187 rpm at 10 N·m, short of the 200 rpm no-load speed the same manual states. The last two points drop 10 N·m within 1 rpm; we could not re-check that against the chart, so treat the end of the curve as approximate.

### robstride-04/robstride-manual-2026-07

Measured: T-N curve with 9 points at 48 V (±2 N·m); thermal derating with 11 points (±0.5 N·m).

Method: T-N points read off the manual's chart (about 2 N·m reading error). Overload times are the manual's table at 50 rpm on a 345 x 345 mm aluminium plate (GB/T 30549-2014), 145 °C winding limit. Thermal torques are the table's whole-N·m figures and the manual states no measurement uncertainty, so the ±0.5 N·m is that rounding only. windingRiseC is the rise from 25 °C ambient to the winding limit, entered only for points that reached it.

Note: Vendor test data on the largest heat sink in the RobStride manuals. Its stall table rates 28.5 N·m continuous with the rotor locked.

### steadywin-gim6010-8/hackaday-spider-bot-2025

Measured: Peak torque 3 N·m (±1).

Method: A second-hand stall figure: the log quotes another builder's measurement and gives no rig. The log's own test compares current at a fixed load (2.29 A here, 1.99 A for a CyberGear). The log states no uncertainty: ±1 N·m is our estimate, wide because the figure is second-hand and the rig unknown.

Not ranked: The log suspects the ODrive Micro's current limit caused the 3 N·m stall, so it likely reflects the driver setup rather than the actuator. Shown, but not set against the datasheet.

Flagged: 3 N·m is 0.27x the 11 N·m rating. The log suspects the ODrive Micro's current limit, so this says more about the driver setup than the actuator.

Note: Weak evidence, kept because it is the only published stall test we found. A run with a stated current limit would replace it.

### xiaomi-cybergear/xiaomi-manual-2023

Measured: T-N curve with 11 points at 24 V (±0.2 N·m); thermal derating with 10 points (±0.25 N·m).

Method: T-N points read off the manual's chart (about 0.2 N·m reading error): the start and end of the flat 12 N·m section (40 and 110 rpm), every 20 rpm from 130 to 270 rpm, and the end of the curve at 280 rpm. Overload times are the manual's own table, taken at 24 rpm with a 120 °C winding limit. Its torques are given to 0.5 N·m and the manual states no measurement uncertainty, so the ±0.25 N·m is that rounding only. windingRiseC is the rise from 25 °C ambient to the limit, entered only for points that reached it.

Note: Vendor test data, not an independent run. The curve is flat at 12 N·m below about 110 rpm, which reads as the driver's torque cap rather than a motor limit.

## Submit a run

Anyone can add a run by pull request: one folder `data/benchmarks/<actuator-id>/<run-id>/` with `run.json`, raw CSVs, a rig description and rig photos, measured to the [test protocol](https://github.com/Miyamura80/getactuators/blob/main/docs/benchmarks/protocol.md). CI checks units, protocol and plausibility, a maintainer reviews it, and it is badged community. [How to submit](https://github.com/Miyamura80/getactuators/blob/main/CONTRIBUTING.md#benchmark-submissions).