Worked example: an open humanoid, ToddlerBot 2XC
Result. ToddlerBot 2XC has the best parts list of any open humanoid in this database: 93 rows, 47 priced, 39 linked to catalog parts.
- Priced rows sum to $5,934.98. Servos and their horn sets are 68.6 percent of that. The computer is most of another 16.6 percent.
- 22 servo housings give the 30 actuated joints in the upstream model. Every class count in the parts list matches the model.
- The upstream model totals 3.4536 kg. The paper states a mass for the robot, but this site cannot show it. Compare the two before you trust the model's masses.
- Static knee load reaches 2.90 N·m at 120° flex on one leg, and 1.35 N·m on two legs. The knee servo's stall torque comes from the maker's manual, which this site cannot show. Compare these loads with the sheet of the servo you buy.
- The database has no rated torque for these servos. The rated-torque check cannot be closed from this database [1].
- The upstream model sets ±10 N·m on every joint, one value for five servo classes.
The arithmetic is in tools/book_calcs/toddlerbot.py; its output is tools/book_calcs/out/toddlerbot.md.
Why this robot
| Robot | BOM rows | Priced | Sum of priced rows | Why not chosen |
|---|---|---|---|---|
| ToddlerBot 2XC | 93 | 47 | 5,934.98 USD | chosen |
| ToddlerBot 2XM | 7 | 7 | 5,635.58 USD | motors only |
| Poppy Humanoid | 67 | 0 | none | no prices |
| Roboto Origin | 62 | 40 | 51,822.53 CNY | no robot mass in the database |
| Bolt | 31 | 0 | none | no prices |
| Sesame | 25 | 0 | none | no prices |
| Open Duck Mini v2 | 17 | 17 | 1,516.36 EUR | smaller list |
| Berkeley Humanoid Lite | 0 | 0 | none | no BOM rows; only an upper-bound price, not shown |
Parts list and cost
Source: the ToddlerBot sheet (toddlerbot) and its PCB files. All priced rows are USD.
| Group | Rows | Priced | Linked to a part | Sum qty x price (USD) | Share |
|---|---|---|---|---|---|
| Motors | 7 | 7 | 7 | 4,069.96 | 68.6% |
| Computers | 3 | 3 | 2 | 982.85 | 16.6% |
| Electronics | 15 | 15 | 1 | 250.76 | 4.2% |
| Parallel gripper | 1 | 1 | 1 | 206.78 | 3.5% |
| Hardware | 15 | 15 | 0 | 195.75 | 3.3% |
| Sensors | 4 | 4 | 2 | 138.91 | 2.3% |
| Materials | 2 | 2 | 0 | 89.97 | 1.5% |
| Power and TTL hub PCB | 18 | 0 | 15 | not priced | |
| 8-channel PCB | 28 | 0 | 11 | not priced | |
| Total | 93 | 47 | 39 | 5,934.98 |
Without the gripper the sum is $5,728.20.
What the sum leaves out:
- All 46 PCB rows are unpriced. The database notes that the docs do not say which ToddlerBot variant uses which board. Do not add both.
- Printed parts appear only as filament: PLA-CF 1 kg x 2 at $34.99, PLA 1 kg at $19.99. There is no print time and no failure allowance.
- The computer row ($955) is the largest single line. The sheet notes it "can replace with Jetson Orin Nano if budget is tight". The database lists a Jetson Orin Nano Super Developer Kit, but the sheet does not say which Orin Nano it means.
Actuators
The upstream model toddlerbot_2xc/toddlerbot_2xc.xml gives each joint a servo class. The table ties those classes to the parts list.
| Model class | Part | BOM housings | Axes each | BOM axes | Model joints | USD each | USD total |
|---|---|---|---|---|---|---|---|
| XC330 | XC330-T288 | 6 | 1 | 6 | 6 | 103.39 | 620.34 |
| 2XC430 | 2XC430-W250 | 2 | 2 | 4 | 4 | 298.89 | 597.78 |
| XM430-W210 | XM430-W210 | 4 | 1 | 4 | 4 | 310.39 | 1,241.56 |
| XC430 | XC430-T240BB | 4 | 1 | 4 | 4 | 137.89 | 551.56 |
| 2XL430 | 2XL430-W250 | 6 | 2 | 12 | 12 | 149.39 | 896.34 |
| Total | 22 | 30 | 30 | 3,907.58 |
The database holds stall torque, mass, voltage range and current for four of these classes. They come from the ROBOTIS e-Manual, and this site cannot show them. It holds no value for the XC430-T240BB beyond price and name.
The Motors group also holds horn sets: 10 HN11 at $8.05 and 4 HN12 at $20.47, $162.38 in all. So motors cost $4,069.96 / 30 = $135.67 per actuated joint. The optional gripper adds 2 more XC330.
Where each class sits, from the model:
| Class | Joints |
|---|---|
| XM430-W210 | both knees, both ankle pitch |
| 2XC430 | both hip pitch, both hip roll |
| XC430 | both ankle roll, both shoulder pitch |
| XC330 | neck yaw, neck pitch, two waist, both hip yaw |
| 2XL430 | both arms from shoulder roll to wrist roll |
The XM430-W210 sits at the knees and ankle pitch. The check below computes the knee load it must carry.
Do not borrow a value for the XC430-T240BB. The catalog has the XC430-W240 and the XC430-W150. Neither is this part. The part-revision guideline shows how far one line can move between revisions [2]: 48 Nm peak on the AK10-9 V2.0, 53 Nm on the V3.0 KV60. The gap stays a gap.
Bus voltage. The servo classes have different supply windows, and one shared bus has to stay inside the narrowest of them. The database holds those windows only from the ROBOTIS e-Manual, which this site cannot show, so read each class's minimum and maximum from its sheet. The STS3215 voltage-rating guideline's check asks for one supply per bus segment [3]. The BOM's 12 V, 4.5 A step-down sets the bus at 12 V. The BOM gives "2000 mAh Lipo Battery" with no voltage, so the battery side is not checked.
Current. Sum the stall current from each servo's sheet over all 30 joints. The database holds stall currents for four classes, from the same manual, which this site cannot show. Check whether a dual-axis housing's figure is per axis; the record does not say. The BOM has two 4.5 A converters, 9.0 A together, and does not say what each feeds. All joints do not stall at once, but the database has no duty profile to set the real peak [4], [5].
The 2XM variant. ToddlerBot 2XM moves the hips to XM430-W350 and the arms to 2XC430. Motors cost $5,635.58 against $4,069.96, $1,565.62 or 38.5 percent more. The upstream model grows from 3.4536 kg to 3.7689 kg.
Static knee check
All masses and lengths in this section come from the upstream simulation model, not from a weighed robot. The paper states a mass and a height for the whole robot (ToddlerBot, arxiv-toddlerbot), but this site cannot show them. Run the total-mass check [6] yourself: divide the model's 3.4536 kg by the paper's mass.
Inputs, from forward kinematics at the zero pose:
| Quantity | Value |
|---|---|
| Hip pitch axis to knee axis, L1 | 0.1015 m |
| Knee axis to ankle pitch axis, L2 | 0.1100 m |
| Total mass, M | 3.4536 kg |
| Mass below one knee, m_s | 0.2407 kg |
| Its centre of mass along the shank, s | 0.0985 m from the knee |
Assumptions:
- The body's centre of mass sits straight above the ankle axis of the loaded leg, so the ground force passes through it.
- The hip sits straight above the ankle. This fixes the shank tilt for each knee flex.
- Slow motion only: no dynamics, no contact impulse, no friction.
With these, H = sqrt(L1^2 + L2^2 + 2 L1 L2 cos(flex)), cos(tilt) = (L2^2 + H^2 - L1^2) / (2 L2 H), and:
- Two legs:
T = g sin(tilt) (M/2 x L2 - m_s x s) - One leg:
T = g sin(tilt) (M x L2 - m_s x s)
| Knee flex | Shank tilt | Arm L2 sin(tilt) | Two legs (N·m) | One leg (N·m) |
|---|---|---|---|---|
| 30° | 14.4° | 0.0273 m | 0.405 | 0.868 |
| 60° | 28.7° | 0.0528 m | 0.782 | 1.676 |
| 90° | 42.7° | 0.0746 m | 1.106 | 2.369 |
| 120° | 56.0° | 0.0912 m | 1.352 | 2.896 |
The model limits the knee to 120° (2.094 rad). These are the gravity terms of the knee torque budget [7]. The knee servo is the XM430-W210. The script also prints stall-to-load ratios, but its stall input comes from the ROBOTIS e-Manual, which this site cannot show. Take the stall and rated torque at your bus voltage from the servo's sheet and divide each by the loads above. A ratio under 1.00 fails.
How to read it:
- Stall is not the limit to design on. A servo's stall torque is its maximum momentary output and differs from its continuous rating [1]. That guideline's check sizes on rated torque. The database has no rated value for this servo, so a stall comparison alone is optimistic.
- Deep single-leg flex is the worst case. At 120° the one-leg load is 2.896 N·m, more than twice the two-leg load, before any dynamics.
- This check says nothing about walking. Walking adds dynamics, impact and duty. The database has no gait torque record for this robot.
- Hip pitch and ankle pitch are not checked. With the assumptions above their gravity moment is zero by construction, which would not hold in a real gait.
Upstream simulation against the catalog
| Item | Upstream model | Catalog | Reading |
|---|---|---|---|
| Force limit, torque model | motor ctrlrange="-10 10" on all 30 actuators | not shown: the stall values come from the ROBOTIS e-Manual | It fails the per-joint effort guideline by its own check [8]: one effort value on mixed actuators. Compare 10 N·m with each class's stall on its sheet. |
| Force limit, position model | position forcerange="-10 10" in toddlerbot_2xc_pos.xml | same | Same reading. The effort-limit guideline asks for the actuator specification [9]. |
| Armature | XC330 0.0048; XC430 0.0044; XM430-W210 0.0022; 2XL430 0.0083; 2XC430 0.0044 | gear ratio for four classes, not shown; no rotor inertia | The file sets a value per class [10]. The armature formula cannot be checked without rotor inertia [11]. |
| Damping | XC330 0.001; XC430 0.0072; XM430-W210 0.0064; 2XL430 0.0013; 2XC430 0.0028 | none | No catalog value to compare. |
| Frictionloss | XC330 0.006; XC430 0.014; XM430-W210 0.038; 2XL430 0.069; 2XC430 0.06 | none | No catalog value to compare. |
| Total mass | 3.4536 kg | paper claim, not shown | Compare the two yourself. |
The model closes a loop at the neck pitch with four connect constraints, and couples drive and driven joints with joint equalities. Expect more than one forward kinematics solution for a closed chain [12]. The leg check above does not depend on them.
If you train a policy on this model, set the force limit per class from each servo's stall torque on its sheet first, then lower it to the rated values once you have them.
What the database cannot tell you here
- Rated or continuous torque for any X-series servo in this BOM.
- Stall torque, mass, voltage range and current for these servos. The database holds them from the ROBOTIS e-Manual, but this site cannot show them.
- Any value for the XC430-T240BB beyond price and name.
- Battery voltage, and which rail each converter feeds.
- Which PCB belongs to which variant, or what a board costs.
- Measured joint torque, gait duty, or a weighed mass.
Guidelines this example exercises
| Guideline | Result here |
|---|---|
| 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. [1] | Open: its check needs rated values, and the database has none. |
| CubeMars lists the AK10-9 V2.0 at 48 Nm peak and the AK10-9 V3.0 KV60 at 53 Nm peak. [2] | Applied: no value borrowed for the XC430-T240BB. |
| 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. [3] | Check applied: one 12 V bus from the BOM's step-down. The servo voltage windows are not shown here. |
| Check supply limits before motor selection, since battery and amplifier limits can set the available peak current [4]; A multi-cell battery must provide the required capacity and current within the application's voltage limits. [5] | Open: no duty profile and no battery voltage. |
| Include gravity in the joint torque budget from the equations of motion, tau = M ddq + c + g. [7] | Gravity at the knee. |
| Expect closed chains to have multiple forward kinematics solutions. [12] | Applies to the neck pitch loop in the model. |
| Set each simulated joint's maximum force from the actuator specification when one exists [9]; Derive each joint's effort limit from its own motor and gearbox; a single large value across the robot is not acceptable. [8] | Fails: one ±10 N·m value on mixed servos. |
| Set armature for a geared actuator from rotor inertia and ratio, as J times G squared, where G is motor speed over output speed [11]; Add realistic armature on wrists, fingers and ankles, because reflected rotor inertia often exceeds their link inertia. [10] | Armature set per class; value not checkable. |
| Before any tuning, check mesh scale and total mass to catch a millimetre-for-metre mistake. [6] | Open here: the paper's mass is not shown on this site. |
The category pages are Servos and Robot computers. For the arm worked in full, see chapter 19.
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.
- 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
- CubeMars lists the AK10-9 V2.0 at 48 Nm peak and the AK10-9 V3.0 KV60 at 53 Nm peak. Level: guide. cubemars.com 2026-11-08
- 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.”
- Check supply limits before motor selection, since battery and amplifier limits can set the available peak current. Level: must. maxonmotor.com 2026-11-08
- A multi-cell battery must provide the required capacity and current within the application's voltage limits. Level: must. ntrs.nasa.gov 2026-11-08“adequate thermal management. The battery must provide both the capacity and current required”
- Before any tuning, check mesh scale and total mass to catch a millimetre-for-metre mistake. Level: must. mujoco.readthedocs.io 2026-11-08
- 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
- Derive each joint's effort limit from its own motor and gearbox; a single large value across the robot is not acceptable. Level: should. isaac-sim.github.io 2026-11-08“choosing both from the motor and gearbox rather than using one oversized value for the whole robot.”
- Set each simulated joint's maximum force from the actuator specification when one exists. Level: should. docs.omniverse.nvidia.com 2026-11-08
- Add realistic armature on wrists, fingers and ankles, because reflected rotor inertia often exceeds their link inertia. Level: should. mujoco.readthedocs.io 2026-11-08
- Set armature for a geared actuator from rotor inertia and ratio, as J times G squared, where G is motor speed over output speed. Level: should. docs.omniverse.nvidia.com 2026-11-08
- Expect closed chains to have multiple forward kinematics solutions. Level: should. hades.mech.northwestern.edu 2026-11-08