How to use this book
Do these steps in order, and do not skip the checks.
- Write the job as numbers: payload, reach, speed, duty cycle, supply, budget.
- Pick a robot type (chapter 2). Find a catalogued robot near your numbers and use it as a sanity check.
- Fix the joints and travel (chapter 3).
- Size each joint for continuous torque, peak torque, speed and heat (chapter 4).
- Choose the actuator family and the transmission (chapter 5, chapter 6).
- Choose bearings, structure, fasteners and process (chapter 7, chapter 8).
- Choose sensors, power and buses (chapter 9, chapter 10).
- Compare CAD, the model and the built robot. Fix every mismatch [1].
- Do the first powered motion only after the safety checks pass [2].
What the book holds
The book answers three questions, and each has its place.
| Question | Where the answer is |
|---|---|
| What must this design satisfy? | The guidelines in each chapter. Each one carries a numbered note: the guideline in the catalog's own words, a link to its source and the date it was read. The note also gives the source's exact words when the source allows copying. |
| What can I buy or copy, and what does it cost? | The catalog: Parts and Robots |
| How do I prove a guideline holds? | The Check column of the guideline tables: calc, sim, model-lint, measure or review. model-lint and calc need no hardware. measure does. This edition states the checks. It does not ship a tool that runs them on a CAD file. |
This book connects them. It states the method, cites the guideline, and links to the part. It holds no number of its own. Every number links to a part, a robot or a category page, or carries a note with its source and read date.
Before you trust a guideline
- Read its scope. The guidelines come from many sources with different scope. The touch-temperature limit comes from crewed spacecraft [3], and the strain wave rated-torque guideline comes from one gear maker [4].
- One model, many values. A part here may hold maker specs, seller claims and bench data that disagree. Two variants of one model are two parts: STS3215-C001 and STS3215-C018.
- Measured models are rare. Few parts have facts of kind
measuredin this edition. See the depth cameras D415 and D455, whose depth error comes from a published test. The strict form is in NASA-STD-5017B: measured motor data, with flight-representative drive electronics for the torque-speed curve [5]. - Worst cases. Combine only worst cases that can occur together. Hot driving torque and cold resisting torque do not stack [6], [7].
- Three views of one robot. Compare CAD, the robot model and the real arm. Typical mismatches: mesh scale, total mass, joint direction, gear ratio [1], [8], [9], [10].
- First powered motion. A bad first motion can hurt a person. Be ready to stop before you run anything [11], [2], [12], [13].
Levels
Every guideline has one of three levels.
| Level | Meaning | Example |
|---|---|---|
must | Breaking it causes failure or harm | In continuous operation, keep motor torque below the thermal limit, not peak or stall [14] |
should | Good practice | Keep a regulating reserve [15] |
guide | Rule of thumb | The square-root duty law [16] |
The level is set within the scope of the source. The life-test guideline [17] is a must, but its scope is a NASA mechanism standard. For a bench arm, read it as the method a stricter program uses, then decide. Do not lower a must that protects a person.
Value types
Every fact on a part page carries a kind. Trust them in this order for sizing: measured, then spec, then claim.
| Kind | Source | How to use it |
|---|---|---|
spec | Maker datasheet or spec table | Design input, subject to conditions and tolerance [18] |
claim | Marketing page or seller listing | Screening only. Confirm before you commit. |
measured | Independent or lab test | Best evidence, but only at the stated test conditions |
One part shows why this matters. For STS3215-C001 the database holds stall torque from three sources: the maker wiki, a seller table and the project README. They give different voltages and do not agree. The database keeps all three. This site shows only the README value, 16.5 kg.cm at 6 V: the seller does not allow copying, and the wiki's terms are not confirmed. Choose the one you can defend. Backlash shows the same gap: the backlash guideline records a bench test of the STS3215 at about 1 deg against a datasheet value of at most 0.5 deg [19].
How to read a range
Category pages and later chapters give ranges as minimum, median and maximum.
- A spec range covers fact rows, not parts. One part with three voltage rows counts three times.
- A spec range mixes spec, claim and measured rows. Open the part pages before you rely on an extreme value.
- A two-value fact counts at both ends. Raspberry Pi Pico lists a supply of 2 to 5 V, and both numbers enter the range of its category.
- The catalog converts values to SI for ranges. Part pages keep the source unit.
- A catalog price range covers every price row, including volume breaks. Chapters 9 and 10 therefore quote quantity 1 rows from the same price table, and say so.
- Where the database has no number, the book says so ("not listed", "not in database"). It does not fill the gap.
- Where the database has a number but its only source does not allow copying, or its terms are not confirmed yet, a table says not shown, and the text names the part without the number. A part or robot with no value this site may show is listed by name only.
Topics and chapters
This table shows where each topic of the guidelines is covered.
| Topic | Chapter |
|---|---|
| Kinematics, workspace, singularities | 3 |
| Actuator sizing, thermal limits | 4 |
| Gears, belts, screws | 6 |
| Bearings, structure | 7 |
| Fasteners, manufacturing | 8 |
| Electrical, battery, bus | 10 |
| Control loops | 11 |
| Robot description | 12 |
| Format conversion | 13 |
| Simulation setup | 14 |
| RL training | 15 |
| Sim-to-sim and sim-to-real | 16 |
| Safety, agent access | 17 |
| Isaac Sim, Isaac Lab, Newton, importers | 25 |
| RL configs that ship | 26 |
| Gap metrics, measured SO-101 data | 27 |
| Serial-bus servos | 28 |
| Sensor timing, calibration, mounting and limits | 29 |
Sources
Each guideline in this chapter with its source, the date it was read and the exact words of the source. Level: must means breaking it causes failure or harm, should is good practice, guide is a rule of thumb.
- Before any tuning, check mesh scale and total mass to catch a millimetre-for-metre mistake. Level: must. mujoco.readthedocs.io 2026-11-08
- Inspect the robot for damage and secure cables before each power-up. Level: should. docs.nvidia.com 2026-11-08
- Keep continuous touch surfaces at or below 44 C. 44 C Level: must. ntrs.nasa.gov 2026-11-08“The maximum allowable continuous touch temperature for surfaces in crewed spacecraft is 44 degrees Centigrade”
- Use strain wave gear rated torque as allowable continuous load at 2000 rpm input. 2000 rpm Level: must. harmonicdrive.net 2026-11-08
- For NASA-STD-5017B EC brushless motors, measure Kt, Km, the torque-speed curve, phase resistances, phase inductances and end-to-end commutated torque, or calculate them from measured values. Use flight-representative drive electronics for the torque-speed curve. Level: should. standards.nasa.gov 2026-11-08“[DDMR 46] Each electronically commutated (EC) brushless motor shall have the”
- Apply torque margin under worst-case conditions through the whole mechanism life, including supply voltage, current limit and controller settings. Level: must. standards.nasa.gov 2026-11-08“[DDMR 9] All torque margin requirements shall be applied under worst-case”
- Combine only worst cases that can occur together, for example do not stack hot driving torque with cold resisting torque. Level: should. standards.nasa.gov 2026-11-08“Only those worst-case combinations that can occur simultaneously need to be considered.”
- Verify each exported joint direction; use the onshape-to-robot inv suffix when a joint must turn opposite to the CAD mate. Level: should. onshape-to-robot.readthedocs.io 2026-11-08
- Write the axis on every moving URDF joint, because an omitted axis silently becomes the x axis. Level: should. wiki.ros.org 2026-11-08“<axis> (optional: defaults to (1,0,0))”
- The SO-101 leader uses several gear ratios, while the follower uses 1/345 gearing. Level: guide. huggingface.co 2026-11-08
- While motors are powered, no hand may be inside a zone where links can trap it. Level: must. docs.nvidia.com 2026-11-08
- On an SO-101 bench setup, be ready to stop with CTRL+C or by unplugging power before running a policy. Level: must. docs.nvidia.com 2026-11-08
- Limit motion on a first hardware run of a sim-trained SO-101 policy; Squint scales actions down for safety. Level: guide. arxiv.org 2026-11-08
- In continuous operation keep motor torque below the thermally set continuous limit, not the peak or stall value. Level: must. hades.mech.northwestern.edu 2026-11-08
- Include a regulating reserve of about 20 percent for bad tolerance cases of motor, load, amplifier and supply. 20 percent Level: should. maxonmotor.com 2026-11-08
- Scale torque rating with duty by the square root law: continuous torque is the 50 percent duty torque divided by 1.414, and the 25 percent duty torque is 1.414 times the 50 percent duty torque. 1.414 Level: guide. pergatory.mit.edu 2026-11-08
- Life test mechanisms whose failure could cost a human life to total operational and ground cycles times at least 4.0. 4.0 Level: must. standards.nasa.gov 2026-11-08“maintenance cycles) by a minimum life test factor of 4.0.”
- Treat motor catalog data as average values and allow for field strength and wire resistance tolerances. Level: should. maxonmotor.com 2026-11-08
- A bench test of the STS3215 measured about 1° of encoder backlash against a ≤0.5° datasheet value. Level: guide. github.com 2026-11-08