Design Book › First power-on

First power-on

What a designer must do

  1. Identify the exact servo and supply before you connect anything. The 7.4 V and 12 V STS3215 carry different stall ratings [1], and the SO-101 leader and follower supplies differ [2].
  2. Rehearse the stop before torque is on. The guideline asks that you be ready to stop with CTRL+C or by unplugging power before running a policy [3].
  3. Move one joint at a time at low speed first [4], [5].
  4. Log temperature, position and timing from the first motion [6], [7].
  5. A person at the bench signs each gate. An agent does not [8], [9].

Scope: a bring-up checklist for the SO-101 with 12 V servos. The list runs from bench checks to first motion to first policy. Every step cites its source. Do the phases in order, and do not skip forward.

The STS3215 voltage-rating guideline records two versions: 16.5 kg.cm stall at 6 V for the 7.4 V version, and 30 kg.cm for the 12 V version [1].

Servo reference

This checklist assumes the 12 V STS3215-C018. Read your servo's label. The database also lists the STS3215-C047, and a 12 V row with no suffix.

The database holds the C018's maker ratings: voltage range, rated torque, stall current, no-load speed and current, operating temperature and mass. Their source does not let us show them, so read them on the maker sheet before steps 5, 12, 13 and 25. The table gives the values this site can show.

Servo reference
ItemValueKind
Stall torque30 kg.cm at 12 VClaim, project README
Baud rates19200 to 1000000 bps listed; default 1000000Spec, from a driver library, on the STS3215 page
Encoder4096 steps per turnSpec, from a driver library, on the STS3215 page
Temperature in one bench testIssues at 71 C, with no built-in thermal protection apparentBench test [6]

Phase 1: identify, on paper

Phase 1: identify, on paper
StepDoPass whenSource
1Read the label on every servoAll follower servos carry one 12 V suffix. The 7.4 V and 12 V versions carry different stall ratings[1]
2Compare with the build listAny departure is written down. The SO-ARM101 list in the database names 7.4 V C001 servos for a follower, not 12 V[1]
3Confirm gearingFollower is 6 servos at 1/345. A leader uses several gear ratios[10]
4Run the description checklistNo must failure open[11], [12]
5Write the torque budgetRequired joint torque at duty is at or below the rated torque of the C018, from its maker sheet. Gravity is included[13], [14]
6Write the test plan and hazard listPlan exists. Hazard list covers the gripper and each hazard class[15], [16], [17]

Phase 2: mechanical, power off

Phase 2: mechanical, power off
StepDoPass whenSource
7Inspect for damage. Secure cablesPre-power checklist signed[8]
8Check stops and clearance on the assembled armEach joint has a stop. No part hits another through the range[18], [19]
9Weigh each removable linkCompared with the model[20]
10Measure key link lengthsCompared with CAD[21]
11Measure backlashMeasured value goes in the error budget. A bench test of the STS3215 measured about 1 deg of encoder backlash against a datasheet value of at most 0.5 deg[22]

Phase 3: electrical, servos unplugged

Phase 3: electrical, servos unplugged
StepDoPass whenSource
12Read the supply with a meterReads 12 V for a follower. Inside the voltage range on the C018 maker sheet. Label fitted[2]
13Check the supply ratingSupply and driver limits are entered as constraints. The stall sum is 6 times the C018 stall current from its maker sheet. The database does not hold the kit supply's current rating, so read its label[23], [24]
14Check continuous loadContinuous load is near 80 percent of supply capacity or less. FAA rule of thumb, used by analogy[25]
15Fit protectionA fuse or breaker sits close to the source. It opens before anything downstream overheats. Wire is rated above it[26], [27], [28]
16Check wirePasses both the drop check and the heat check. Thin wire crossing joints has support[29], [30]
17Rate any switchSwitch nominal rating is at least 2 times motor current at 12 VDC. FAA table[31]
18Check bus wiringOne supply per bus segment, as the voltage-rating guideline checks. Leader and follower supplies are not swapped[1], [2]

Gate A: before servo power

Do not continue until all of these are true. A person signs [8].

Gate A: before servo power
StepDoPass whenSource
19Brief everyone at the benchHands stay clear of pinch zones[32]
20Rehearse the stop: CTRL+C, and pull the power plugTime recorded. The guideline names these two stops and no rated emergency stop. The rehearsal is this checklist's step[3]
21If an agent is connected, post a person at the power cutCut-off rehearsed and timed. Access level written down[9], [33]

Phase 4: powered, torque off

Bring the bus up first, one servo at a time with no load, as in TTL serial bus [34], [35]. Step 22 assumes that is done. Do not power six unconfigured servos at once.

Phase 4: powered, torque off
StepDoPass whenSource
22Power up and read each servoAll 6 answer. Position, voltage and temperature read sane on the real arm[36], [10]
23Calibrate this armYour own offsets are stored. Another setup's offsets do not count[37]
24Move each joint by hand, read the signPositive motion matches the model[5]
25Set the host temperature limitLimit recorded. Read the C018 operating limit on its maker sheet. The database holds no maker overheat threshold. One test reports issues at 71 C. A later page comment says firmware protection cuts torque near 70 C[6]
26Set speed limits on the controllerLimit lives on the robot side, not only in simulation config[38]
27Set position limitsTargets stay inside joint range and short of hard stops[12], [39]
28Set an under-voltage warningWarning point set on the bus. The aircraft points, 13 V on a 14 V system and 26 V on a 28 V system, are analogies only[40]
29If cameras are used, check them by nameEach feed is the expected one, after any replug[41]

Gate B: before first motion

This is the gate before first motion. A person signs [8].

The database has no guideline on what an STS3215 does at the moment torque turns on. Treat the first enable as an unknown: hands clear, a person on the power plug [32], [3].

Phase 5: first powered motion

Phase 5: first powered motion
StepDoPass whenSource
30Limit motion for the first runPeak joint speed logged[4]
31Bound commandsTargets pass through a rate limiter and a force limit. The guideline is written for simulated drives; this step applies the same idea on the host[42]
32Jog one joint positive, a small move. Repeat per jointDirection matches the model[5]
33Jog each joint through full range at low speedNo cable snags. Targets stop short of limits[43], [39]
34Run a step response per jointSteady-state offset logged[44]
35Timestamp commands on the real loopJitter computed. For a PI loop, rate varies at most plus or minus 20 percent over any ten samples[7], [45]
36Log command-to-response delay per jointDelay recorded for the model[46]
37Run the worst duty cycle and log temperature until it settlesStays under the host limit. One test saw no apparent built-in thermal protection at 71 C. A later page comment says firmware protection cuts torque near 70 C[6], [47]

Stop conditions

Stop on any of these. Fix the cause before you continue. The stop itself is CTRL+C or the power plug [3].

Stop conditions
SymptomActionSource
Hand inside a pinch zoneStop[32]
Violent shakingStop. Suspect unmodeled delay. Log it[46]
Joint moves opposite to the modelStop. Fix the model or sign[5]
Temperature at the host limitStop. One bench test of the C018 showed issues at 71 C with no built-in protection apparent, and calls for manual shutdown there. A later page comment says firmware protection cuts torque near 70 C[6]
Servo enters overload protectionStop. Stall is a momentary maximum, so size against rated torque, not stall[13]
Cable snagsStop. Reroute[43]
Bus voltage at the warning pointStop. Count drops and current limit in the torque budget[48]
Fixture moved or camera repluggedStop. Rerun the alignment and camera checks[49], [41]

Phase 6: replay, then first policy

Phase 6: replay, then first policy
StepDoPass whenSource
38Replay one motion on the robot and in simulationPaired logs saved[50]
39Compute the gap per jointTable for position, velocity and torque[51]
40Fit the model to the logsFit has real bounds[52], [53]
41Run sim-to-simTransfer succeeds. The guideline treats that as a possible first step toward deployment. See Sim-to-sim and sim-to-real[54]
42Diff training and deployment post-processingPolicy-output post-processing matches[55]
43Consider scaled-down actions for the first runSquint used 0.15 and 30 Hz. Any rate mismatch is recorded[56], [57]
44Keep the scene as trainedLayout, lighting and props match. For the typing policy, the keyboard stays fixed[58], [59]
45Run the policy with the stop readyA person is at the bench[3], [32]
46Report real successTrial count and the simulation figure stated[60]

Where the database is thin

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.

  1. The STS3215 7.4 V version is rated at 16.5 kg·cm stall torque at 6 V; the 12 V version is rated at 30 kg·cm. Level: guide. github.com 2026-11-08“The 12V version has a stall torque of 30kg.cm.”
  2. Label the SO-101 leader and follower supplies, because they differ at 5 V and 12 V and must not be swapped. 5 and 12 V Level: must. docs.nvidia.com 2026-11-08
  3. 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
  4. 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
  5. 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
  6. In a Robonine STS3215 test, issues appeared at 71°C and no built-in thermal protection was apparent, so manual shutdown is needed at that temperature. A later page comment says firmware protection cuts torque near 70°C. Level: should. robonine.com 2026-11-08
  7. Measure control loop timing on hardware, because it differs from simulation. Level: should. docs.nvidia.com 2026-11-08
  8. Inspect the robot for damage and secure cables before each power-up. Level: should. docs.nvidia.com 2026-11-08
  9. Keep a person within reach of a physical power cut during agent-driven hardware tests; Project Fetch's organizer had to grab the robot and power it off. Level: must. anthropic.com 2026-11-08
  10. The SO-101 leader uses several gear ratios, while the follower uses 1/345 gearing. Level: guide. huggingface.co 2026-11-08
  11. Before any tuning, check mesh scale and total mass to catch a millimetre-for-metre mistake. Level: must. mujoco.readthedocs.io 2026-11-08
  12. Keep position command ranges inside joint ranges, except for a gripper that must close past pad contact. Level: must. mujoco.readthedocs.io 2026-11-08
  13. A servo's stall torque is its maximum momentary output and differs from its continuous rating. Real-world output is generally closer to the performance-graph measurements. Level: guide. emanual.robotis.com 2026-11-08
  14. Include gravity in the joint torque budget from the equations of motion, tau = M ddq + c + g. Level: should. hades.mech.northwestern.edu 2026-11-08
  15. Test each policy on hardware through a systematic plan; good simulation results do not show that it works there. Level: must. docs.nvidia.com 2026-11-08
  16. Include grippers, end effectors and accessories in the risk assessment. From a vendor blog summary, not a standard. Level: guide. blog.robotiq.com 2026-11-08
  17. Include struck-by, caught-between, crushing, trapping, slip, trip, fall and electrical hazards in the hazard list. From a NIOSH overview page. Level: guide. cdc.gov 2026-11-08
  18. Fit non-jamming mechanical stops wherever overtravel harms the mechanism or system. Level: must. standards.nasa.gov 2026-11-08“[DDMR 30] Non-jamming mechanical stops shall be incorporated into all mechanisms where”
  19. Measure clearances on the highest level of assembly possible. Level: should. standards.nasa.gov 2026-11-08“[DDMR 8] Clearance measurements shall be performed on the highest level of assembly possible.”
  20. This guideline is held back: its source does not allow us to publish it yet.
  21. Compare CAD against the as-built robot as part of gap measurement. Level: should. docs.nvidia.com 2026-11-08
  22. 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
  23. Check supply limits before motor selection, since battery and amplifier limits can set the available peak current. Level: must. maxonmotor.com 2026-11-08
  24. Check power supply requirements, such as maximum motor terminal voltage and current limits, during the first motor selection. Level: must. maxonmotor.com 2026-11-08
  25. Hold continuous load near 80 percent of source capacity when no monitoring is used; this is a rule of thumb, not a load analysis. 80 percent Level: guide. raanz.org.nz 2026-11-08“can exceed 80 percent output load limits of the”
  26. Protect every power wire with a fuse or breaker located as close as possible to the source bus. Level: must. raanz.org.nz 2026-11-08“with circuit breakers or fuses located as close”
  27. Make the circuit protector open before anything downstream overheats or smokes. Level: must. raanz.org.nz 2026-11-08“breaker must always open before any compo-”
  28. Size wire to carry continuous current above the protector rating, including its time-current curve. Level: must. raanz.org.nz 2026-11-08“continuous current in excess of the circuit”
  29. Size each wire for both voltage drop and overheating; both must be met. Level: must. raanz.org.nz 2026-11-08“(2) The size should be sufficient to pre-”
  30. Give wire under size 20 extra clamps and support, and keep it out of excessive vibration, repeated bending, or frequent disconnection. Level: should. raanz.org.nz 2026-11-08“desirable to use wire sizes smaller than #20,”
  31. Derate a switch nominal rating by factor 2 for 12 VDC motor loads. 2 Level: should. raanz.org.nz 2026-11-08“12 VDC Motor 2”
  32. While motors are powered, no hand may be inside a zone where links can trap it. Level: must. docs.nvidia.com 2026-11-08
  33. Treat an agent's access level to a robot as a core part of the system's safety review, because small tool changes can make large capability changes. Level: must. anthropic.com 2026-11-08
  34. Connect only one servo and no load at first power-up. Level: must. github.com 2026-11-08
  35. Add servos one at a time and reverify communication after each one. Level: should. github.com 2026-11-08
  36. Have an agent verify robot connection facts against the hardware, because online information varies in accuracy. Level: should. anthropic.com 2026-11-08
  37. Calibrate each SO-101 before running a policy; another setup's released offsets do not calibrate your robot. Level: must. isaac-sim.github.io 2026-11-08“they do not calibrate another robot.”
  38. Do not treat Isaac Lab joint_velocity_limit as a safety clamp, because its enforcement depends on the backend. Level: must. isaac-sim.github.io 2026-11-08“is not a backend-independent safety clamp”
  39. During policy transfer, stop targets short of the hard limits of each joint. Level: should. isaac-sim.github.io 2026-11-08“keep targets away from hard joint stops”
  40. Aircraft warning lights can be triggered when battery bus voltage falls below 13 V on a 14 V system or 26 V on a 28 V system. 13 V Level: guide. raanz.org.nz 2026-11-08“bus voltage drops below 13 volts on a 14-volt quantity within the rated capacity of the re- system or 26 volts on a 28-volt system.”
  41. Verify camera assignments after any replug, because indices can change. Level: must. docs.nvidia.com 2026-11-08
  42. Bound large drive forces with a force limit or a rate limiter on drive targets. Level: should. docs.omniverse.nvidia.com 2026-11-08
  43. Route cables so that they cannot snag during motion. Level: should. docs.nvidia.com 2026-11-08
  44. Plot a real step response to quantify stiction, then add joint friction in simulation; the UR10e showed about 0.25 degrees of offset. 0.25 degrees Level: should. isaac-sim.github.io 2026-11-08“observed contact offsets of approximately 0.25 degrees from the commanded setpoint.”
  45. Keep sampling rate variation to at most plus or minus 20 percent over any ten samples for a PI loop. 20 percent Level: should. wescottdesign.com 2026-11-08
  46. Suspect unmodeled delay when a policy shakes violently on hardware. Level: should. arxiv.org 2026-11-08
  47. Run a duty thermal test until temperature reaches a steady trend or five thermal time constants have passed. 0.1 C/min Level: should. arxiv.org 2026-11-08“0.1◦ C/min), or (ii) five thermal time constants have elapsed (estimate τth from initial temperature”
  48. Count supply voltage, voltage drops and current limit when calculating available actuator torque. Level: should. standards.nasa.gov 2026-11-08“Supply voltage, voltage drops, current limit”
  49. Moving any fixture or device invalidates the layout check, so verify positions again before the next run. Level: should. docs.nvidia.com 2026-11-08
  50. To measure the actuation gap, replay the same motions on hardware and in simulation and compare the two logs. Level: should. docs.nvidia.com 2026-11-08
  51. Report the gap per joint in position, velocity and torque, not as one robot-wide number. Level: should. docs.nvidia.com 2026-11-08
  52. Fit armature, damping, friction, gains, inertia and delay to hardware logs with optimization, rather than hand-tuning, when matching a specific robot in MuJoCo. Level: should. mujoco.readthedocs.io 2026-11-08
  53. Expect mujoco.sysid to use box-constrained least squares, so give each parameter real bounds. Level: must. mujoco.readthedocs.io 2026-11-08
  54. Treat successful sim-to-sim transfer as a possible first step toward sim-to-real deployment. Level: guide. isaac-sim.github.io 2026-11-08“Successful transfer can be a first step toward sim-to-real deployment.”
  55. Apply the same policy-output post-processing during real-world inference as in Isaac Lab. Level: must. isaac-sim.github.io 2026-11-08“is also applied during real-world inference”
  56. On a first SO-101 transfer, consider scaling actions down and raising control rate at deployment; Squint used 0.15 and 30 Hz and reports smoother paths. 0.15 action scale Level: guide. arxiv.org 2026-11-08
  57. Record any deliberate change between training and hardware control rate; Squint accepts such a mismatch and states it. Level: should. arxiv.org 2026-11-08
  58. Keep the SO-101 typing keyboard fixed: its policy uses a fixed A–Z key map without camera input and cannot adjust to movement. Level: must. isaac-sim.github.io 2026-11-08“it cannot adjust to a moved keyboard.”
  59. Keep the real workspace layout, lighting and props consistent across calibration, teleoperation and evaluation. Level: should. docs.nvidia.com 2026-11-08
  60. Report real success with its trial count, as Squint does with 91.3 percent over 80 trials on 8 tasks, and show the simulation figure beside it. Level: should. arxiv.org 2026-11-08
  61. Multi-servo startup transients are a frequent power fault. Level: should. github.com 2026-11-08
  62. For jitter or dropped packets, check supply droop, common ground, cable length, connectors, branch topology, baud rate and half-duplex direction timing. Level: should. github.com 2026-11-08
  63. Avoid long star branches and loose connectors on the bus. Level: should. github.com 2026-11-08