2026-10-22: Authoritative motor-capacity assessment below supersedes earlier motor-sizing wording here and in linked records. Other rail worksheet entries remain unvalidated, not zero. RELEASE HOLD: no battery input, final regulator/protection, powered commissioning or layout release approved. Do not connect a battery to provisional bench headers.

Head load review, section 7 distinguishes manufacturer examples from unverified worst cases. Head board full consumption, LCD peak and populated regulator thermal capacity remain unknown. Both displays are alternatives, not additive loads. Do not treat module radio current as complete devboard current. High-current regulator capability and Jetson loaded-voltage headroom remain conditional as explained in the review package.

Design calculations, not measured facts:

Complete and sum load tables first. Then rate fuses/protection for load curves and fault interrupt capability, FETs for voltage/SOA/hot resistance, connectors/wires for continuous and pulse temperature rise, capacitors for effective bias/ripple/inrush, and regulators for load transients and thermal margin. MLCC nominal values are not effective capacitance. Per-cell policy cannot be inferred from a converter UVLO. AP63205 is prohibited as servo regulator; candidate D24V150F5 remains unapproved for simultaneous stall.

Inventory confirmed by owner:15 XL330-M288-T at5.0V; exact hardware/firmware revisions still needed. Remaining inputs: battery label/capacity/C-rating/protection and balance connector;measured normal/peak/stall-duration/regen profiles;speaker model;head regulator marking;actual harness lengths/gauges;ambient/cooling;compute mode and runtime goal. Without these,final regulator,fuse,shunt,FET and harness ratings are blocked—not guessed. Motor feed/return must bypass PCB J4/J6/USB. U2D2-side VDD wire omitted does not prove DATA power-off isolation. See U2D2 Migration for passive wiring/contact-current and USB-on/motor-off test gates.

Analysis/records only; no schematic, PCB, fabrication or order changes. User confirms 15 ROBOTIS XL330-M288-T at nominal 5.0 V. Adopt 30 A / 150 W provisional supply-capability envelope for sizing comparison, not continuous motor consumption, a selected converter, a safe stall duration, or an approval of purchased hardware. 25 A / 125 W is a comparison option. Lower-capacity operation remains conditional on torque acceptance, enforceable limits and measured evidence.

Fresh read-only inspection: native BOM 72 instances/16 grouped types, no motor converter, power FET, fuse or implemented high-current protection. J4 is low-power bench boundary; pins 5/6 NC. J6 is B3B-PH-K-S diagnostic only: 1 GND/2 NC/3 DATA, TP5 passive DATA. PCB summary has 61 net nodes, no routeSegment/via nodes, four copper layers and inherited 160 x 100 mm rectangle. A fill object does not establish a qualified motor path. Prior migration records 44 populated nets. No fabricated distribution exists. External purchased U2D2 is communications only: GND+DATA, VDD conductor omitted/insulated; motor feed AND return must bypass PCB/J4/J6/USB/U2D2. No new ERC claim is made by this documentation task.

Official web/manual-source facts below are explicitly web-source-supported, not independently tolerance-qualified hardware measurements. Durable source links are retained for audit; PDF-derived facts use validated inline citations. Access date 2026-10-22.

Sensitivity to global table 1.5 A: 22.5 A/112.5 W; 25 A margin 2.5 A/11.1%, 30 A margin 7.5 A/33.3%. Supply capacity may be continuous-rated under a defined thermal environment, while motor demand can be transient; both need separate qualification. Nominal stall is neither a safe mission nor a guarantee bounding fault current. A 30 A-capable source may deliver far more fault energy/current before protection clears.

Register clarification: Current Limit(38) factory initial value is 1750; its allowed range is 0–1750, not an initial value of zero. Mode0 is current/torque control and mode5 is current-based position per the official XL330 source/control-table includes linked above. Any 800-unit proposal is deliberately below the factory setting and requires explicit write/readback before torque enable. For a genuinely thermally qualified 15 A supply, illustrative budget allocation: each motor commanded/control-limited to 800 units ≈0.800 A, 15×0.800=12.0 A; reserve 0.30 A for aggregate control/standby uncertainty (17mA×15=0.255A used conservatively, even though XL330 source-current measurement may include electronics); 1.20 A for uncharacterized overshoot/model/measurement error; 1.50 A unallocated headroom. Sum allocations=15.0 A. These reserves are planning assumptions, not measured error bounds; if actual overhead is included, do not double-count in a final measured budget. Quantization: n=floor(1000×I_target[A]); n=800 gives approximately 800 mA, one-count step approximately 1mA; unit approximation is not calibrated accuracy.

Use only verified current mode 0 or current-based position mode 5 as appropriate, EEPROM Current Limit≤800, signed Goal Current bounded by it, torque initially OFF; read back mode/current limit/goals/firmware/ID on all 15 before arming. EEPROM limit retention alone does not prove RAM goal/watchdog retention. Disable startup-torque, torque-on-goal-update and unreviewed RAM restore; repeat verification after power cycle, brownout, reboot, firmware update and factory reset. Reject writes/mode changes or missing readbacks while enabled. A single service owns U2D2; reject stale trajectories and prohibit automatic rearm.

Require independent aggregate sensing/limit and normally-off hardware cutoff, reviewed timing and energy coordination. On USB loss, service crash, invalid ID/configuration, watchdog timeout, branch overload or sensor fault: stop fresh commands, request torque-off where communication survives, latch hardware motor-power off independently of Linux; mechanical restraint and post-cutoff regeneration containment required. Watchdog setting is quantized in 20 ms; e.g. value5=100ms illustrates arithmetic only, NOT an approved response time. Choose safe time from mechanics/energy and test. Even though XL330 senses input current, no ideal instantaneous bus-current clamp, tolerance or transient ceiling is guaranteed. Low-capacity option is rejected until measured worst combined RMS/peaks fit actual derated source without UV resets, unsafe motion or excessive torque loss and the owner accepts restricted performance.

At load, I_bat≈[5×I_motor+P_motor_distribution_loss]/(VIN_loaded×eta_motor)+sum(P_other_switching_rails/(VIN_loaded×eta_j))+sum(I_LDO_load+IQ_LDO)+I_direct. Include the Jetson DC branch, U2D2 USB load counted ONCE within Jetson/peripherals, torso3V3, audio, face, sensors and safety supervisor; do not add USB motor power. VIN_loaded=pack voltage minus battery/cable/protection sag; solve source impedance iteratively at VIN_min. Other rail figures in the original inventory remain unknown; face 5 V/1 A remains only owner allowance. Thus TOTAL battery current/runtime/fuse size is UNVERIFIED, not just motor reflected current.

Illustrative motor-only conversion, excluding distribution losses and other rails; efficiency assumed, not a guarantee or selected pack cutoff:

9 V is illustrative only, NOT approved LiPo discharge endpoint. Actual cell/BMS/cutoff must determine minimum. At sleep use measured no-load/disabled IQ, not fixed high-load efficiency: Pololu typical100mA input enabled; at illustrative12V disabled60–120microamp. At high load an efficiency measured at converter terminals already includes internal IQ; do not add it again. For a conservative model whose efficiency excludes controller losses add IQ separately and state that convention. i7A exact sleep/no-load IQ not established here. 95.5% i7A example is at full rated load/12V/Tc25C, not guaranteed at30A in hot torso.

Prefer protected external star feed/return with separately protected individual motor spurs, or evidence-based small branches. Shared DATA may daisy-chain, but do not inadvertently share downstream power in a star-plus-chain loop. Do not switch motor ground. Qualify open-return/alternate return through USB and reference spur. No blanket approval of EH motor power chains: two nominal stalled motors2.94A leave only0.06A/2% catalog margin, and global1.5A figures leave none. Three nominal stalled motors4.41A exceed3A. One motor1.47A is below3A arithmetically, NOT proof of harness thermal/drop/fault suitability.

For a path to motor k, Vpin_min=Vreg_min−I_total×Rtrunk_hot−sum(I_segment×Rsegment_hot)−I_k×Rspur_hot−dynamic droop. All resistances include feed AND return, crimps, connectors, protection, shunts, joints and disconnects. Vpin_max=Vreg_max+load-release/regen overshoot; verify every motor pin stays within manufacturer3.7–6.0V, with measurement uncertainty. Adopt provisional design operating target 4.5–5.5 V at motor terminals to retain torque/rail margin; this tighter band is an engineering target, not a motor guaranteed performance specification.

At regulator5V±4% illustrative tolerance floor4.8V, only0.30V is left for total static drop to a4.5V target (before dynamic reserve); whole shared loop at30A would require R≤10milliohm, at22.05A≤13.61milliohm. Manufacturer minimum3.7V alone gives1.1V/30A=36.7milliohm, but that is not a good5V torque target and excludes dynamic margin. At5.2V high tolerance only0.8V remains to6.0V maximum, or0.3V to5.5V target. Split voltage budget among trunk/branch/contact/protection/transients rather than allocating all to wire.

Illustrative conductor CALCULATIONS, not ampacity certification: assume copper rho20=0.0175 ohm·mm²/m, temperature coefficient0.00393/C and conductor80C, giving R80=1.236R20; these are explicit analytical material assumptions, not verified cable data. Use measured/max specified hot cable resistance for release. For one-way length L, Rloop=2Lrho/A×1.236.

These omit contacts/protection; first trunk example already exhausts0.30V static allowance. At30A every additional1milliohm costs30mV/0.9W. For an allocated trunk drop0.15V, Rtrunk≤5milliohm and A≥4.33mm² at0.50m one-way/80C, before joints; at0.25m A≥2.16mm². The final wire must separately satisfy insulation, bundling, flex, ambient and fault-time thermal limits. Never force large trunk wire into EH crimps.

Using JST post-environment20milliohm/contact as planning bound (NOT a guaranteed hot resistance): one mated feed+return pair40milliohm gives1.47A drop58.8mV and heat86.4mW; at2.94A117.6mV/345.7mW. Two such connector interfaces in a cable loop double these values. Real hot/aged/fretted/crimp resistance needs evidence, so this is not final thermal certification. Illustrative0.50m motor spur plus two interfaces totals146.35milliohm: at1.47A0.215V/0.316W, at2.94A0.430V/1.265W, excluding trunk/fuse. A trunk0.196V plus single spur0.215V would give4.389V from4.8V floor: FAIL provisional4.5V static target even before dynamics. Shorter cables, lower resistance contacts/path or tighter regulator tolerance require actual analysis, not turning voltage up beyond safe high-end margin.

PASS means stated narrow analytical condition only; FAIL means unsuitable for stated use; UNVERIFIED means missing evidence, not implied acceptable.

Coordinate main fuse near pack, input reverse protection/inrush, motor high-side disconnect and branch interruption. Verify DC interrupt rating≥pack voltage and prospective fault current; normal current/inrush must not nuisance-trip hot derated protection; clearing I²t/time must remain below wire/contact/FET SOA damage and branch faults must clear before upstream healthy branches/Jetson collapse where required. Converter current-limit/hiccup may prevent a fuse from ever clearing: check both minimum available fault current and maximum capacitor/battery discharge. No arbitrary fuse/PTC rating is selected from capacity alone. The initial PTC screening allowance hold≥1.5×peak would imply45A for30A peak before hot derating; it is not a manufacturer rating, general coordination rule, or selection approval. Any PTC must independently satisfy manufacturer hot hold/trip curves, pulse duration, resistance/drop and fault-energy/selectivity checks; no PTC is selected here.

Transient planning equations only: C≥ΔI×Δt/ΔV after ESR stepΔI×ESR and wiring Ldi/dt; regen storage Ecap=0.5C(Vmax²−Vinitial²), motor energy from inertia/gravity and measured negative power integral. At5→6V storage is5.5J per farad; even1J would require0.182F ideally, without ripple/inrush/control-loop margin. Example demonstrates why arbitrary bulk caps are not a regen solution, NOT a capacitor recommendation. Qualify retained motor-side absorption after disconnect; pack may reject charge (full/BMS open/charging restrictions), and buck converters are not assumed bidirectional. No TVS, brake clamp, capacitor or parallel-converter solution selected without energy, repetitive power, clamp tolerance/dynamic voltage, stability and protection coordination checks.

Before energized tests approve source, protection, thermal and mechanical restraint plan. Use electronic load first; one restrained motor next; expand only after review. Record instrument calibration/bandwidth/probe placement, VIN/VOUT at source and worst motor pins, per-branch/total/battery I(t), RMS/peaks, peak duration, duty, negative power integral and all component temperatures. Numeric hardware-dependent times/temperature limits missing below must be filled before execution, not guessed.

Oscilloscope low-voltage limits apply while commanded powered operation is expected; an intentional shutdown ramp below3.7V is not itself a failure, but no unsafe torque/phantom powering/overshoot may occur. Response deadlines, safe stall duration, off-state leakage threshold and hot margins remain explicit release blockers until derived from actual hardware and mechanical hazard analysis.

Assigned records work: existing files reviewed COMPLETED; independent quantitative motor/options/drop/input/reflection/protection assessment COMPLETED with named unknowns; official sourcing COMPLETED for available facts, reverse-energy guarantees and hot-system ratings UNVERIFIED; lower-capacity approval BLOCKED by measured limits/torque and fault validation; final regulator/pack/harness/thermal/transient approval BLOCKED by inputs above; physical tests UNATTEMPTED (no hardware evidence); schematic/PCB/manufacturing changes UNATTEMPTED intentionally. No layout release. No claim of whole-project completion.