Faze4 Fully 3D-Printed Six-Axis Arm: Cycloidal Gearbox, Assembly and Two-Layer Control Stack

📌 What this guide covers

Faze4 comes from the same team as PAROL6 (Source Robotics / Petar Crnjak), but it takes a very different route:instead of simplifying the mechanics to cut cost, it takes the core mechanism of an industrial arm —the cycloidal pin-wheel reducer— recreates it by 3D printing, and packs one into every joint. The result is roughly 1000 parts, about 15 kg, with all cabling routed internally, and looks and behaves close to a small industrial arm.

Why it is worth building:Low backlash + maintainability. A cycloidal drive meshes many teeth at once and resists shock loads by nature. Printed in plastic, when the disc wears or breaks you simply print another one — something you cannot do with an off-the-shelf actuator module.

This guide runs in robot data → cycloidal principle → printing → assembly → power distribution → control stack — six sections. Accept one premise up front:the time cost of this machine is in assembly (measured in weeks), not in printing.

📐 Section 1: Robot Data (Copy These Down Before Writing Code)

This table is the baseline for everything that follows: the controller needs it, the simulation needs it, and it must be updated whenever you change the mechanics. The team also provides Faze4_Work_Envelope.m scripts that compute the workspace envelope —run them before planning a trajectory, so the arm does not drive into itself.

🌀 Section 2: Why the Cycloidal Pin-Wheel Drive Has Low Backlash

How it works: an eccentric shaft drives the cycloidal disc in an eccentric motion; the outer teeth of the disc engage several pins of the pin ring simultaneously, and the tooth profile converts the eccentric motion into low-speed co-rotating output. It is cheap, shock resistant and gives a large reduction ratio; compared with a harmonic drive its drawbacks are slightly larger volume and tighter manufacturing tolerances.

• Sources of backlash: tooth-profile accuracy of the cycloidal disc, clearance between pins and rollers, and the fit of the output mechanism. So print quality of the cycloidal disc directly determines joint feel and noise.
• Maintainability: when it wears or breaks, printing a new disc and swapping the rollers restores it — the biggest advantage over ready-made actuator modules.
• Parametric: the official files include cyclo_disk_STEP_files.zip, letting you tune the tooth profile to your needs — always re-check the clearances afterwards.

🖨️ Section 3: Printing Strategy

1. Dial in your printer using the official 3D printing guide: it covers materials, printers and print settings. The total printed volume is large, so print one or two critical parts first to verify dimensional accuracy and interlayer strength, then batch the rest.
2. Print several cycloidal discs in one run: pick the set with the tightest dimensions — a practical way to reduce backlash. Even batches from the same printer vary slightly, so spare parts never go to waste.
3. Stock extra bearings and screws: the project explicitly warns that bearings and screws can be damaged or ruined during assembly. Buy screws locally (cheap and you control the quality).
4. About 1000 parts in total: printing is only the start. Batch prints per joint module and sync them with your assembly pace — far less chaotic than printing everything first and assembling afterwards.

🔧 Section 4: Assembly (Modules First, Then the Whole Arm)

🔌 Section 5: Distribution PCB and First Power-Up

• Use the V2 distribution PCB: upstream states clearly that V1 is obsolete and not recommended. Check the BOM and reference designators before ordering boards, and run continuity and short checks before applying power.
• First power-up in three steps: ① supply power without motors connected and verify every rail voltage; ② connect the motors but send no motion commands, checking driver LEDs and standby current; ③ jog one joint at a time through small angles, verifying direction and limits axis by axis.
• Calibrate the zero position joint by joint: this machine has many joints and large reduction ratios, so calibration error gets amplified. To verify: rotate J1 by 90°, measure the actual end displacement and compare it with the kinematics result — a large deviation means the reduction ratio or direction parameter is wrong.

📦 V2 Distribution PCB Kit (redrawn by LateAI · ready for fabrication)

The official repository ships the V2 Gerbers and a PDF schematic; every designator and net name is buried in the drawings, so any change means hunting through the PDF.This pack is a V2 netlist plus companion files, rebuilt by checking Schematic.PDF pin by pin: 142 components, 105 nets and 429 pin connections, all in machine-readable BOM and netlist form — easy to script against and to version-diff, and usable with the bundled extension to regenerate schematic + PCB in one click inside EasyEDA Pro.

What is in the pack

How to use it

1. You just want boards made: order parts from bom.csv and solder by designator; the outline is 115 × 195 mm, 2 layers, 1 oz copper, green solder mask / white silkscreen is enough.
2. You want to edit the design yourself:Use EasyEDA Pro throughout. Install the bundled .eext extension (steps below), then click menu Faze4 V2 → ① Generate Schematic,② Generate PCB to redraw automatically; without the extension, import the included netlist_v2.net via the Pro version's File → Import → Netlist (select type Protel2) to bring all 105 nets into the PCB.
3. Verify before committing to production: see the 3 items to review below — those three have no unique answer in the PDF, so sensible defaults were assumed.

Install the extension and redraw in one click (EasyEDA Pro)

3 items to review

• Power pins of the KF301-6P driver terminal: assumed 1=PUL 3=DIR 5=EN,2/6=GND,4=+5V.
• 21P arm harness: the 9 inputs use odd pins, even pins carry +24V.
• +24V source: connected to the 4P power terminal.

📥 Controller / Distribution Board Production Files (Full V1 / V2 / V3 Set)

The pack above is the netlist and one-click redraw tooling; this section lists every production file for all three board revisions, one download at a time (assembly drawing / drill drawing / pick-and-place / BOM / Gerber / STEP / firmware / netlist and sourcing tables). For the complete connection data, open the matching interactive page — click a net to see its pins, click a component to see its designator; the data there matches these files exactly.

V2 · Distribution Board (current upstream revision · recommended)

V1 · Distribution Board (obsolete · not recommended upstream · reference only)

V3 · Teensy 4.1 Controller Board (site design draft · not fabricated yet)

🕹️ Section 6: Control Stack (Teensy Firmware + Matlab / ROS Upper Layer)

This Teensy low level + Matlab / ROS upper layer split is the highlight of its architecture: the low layer owns real-time behaviour and servo control, the upper layer owns trajectory and planning, joined by a clean interface — simulate every new motion before running it on hardware, which saves a lot of disassembly time.

🎯 Bonus · Real 3D Trajectory Lab (explore it in your browser before building)

Building it takes roughly 1000 parts and weeks of assembly, but its kinematics can be explored right now in a browser: the Faze4 Real 3D Trajectory Lab uses real parts converted from the official STLs (base + six joint links, including the external shape of the in-joint cycloidal reducer) and solves 6-axis inverse kinematics live in the page — the end effector actually traces sine / circle / triangle on a vertical drawing board, with all six joint angles and the tracking error shown alongside; drag the target point and the whole arm repositions instantly.

🚧 Pitfalls to avoid

• Most of the time goes into assembly: progress as joint module → per-module acceptance → whole arm; do not assemble everything in one go.
• Print several cycloidal discs per run: keep the best-fitting set to noticeably cut backlash and noise.
• Use the V2 distribution PCB: upstream clearly marks V1 as not recommended; do not cut that corner to save a fabrication fee.
• Do not tension the J4 belt: adding a tensioner there only increases load.
• Route cables internally during assembly: it is what keeps the arm looking clean, and reworking it afterwards is near impossible.
• 15 kg / 1000 parts: plan your bench, supports and load capacity as you would for industrial equipment, not a desktop toy.
• Keep the URDF in sync first: if you change the mechanics without updating the URDF, simulation and hardware diverge and planning will fail.
• Inconsistent licence labels: defer to the LICENSE file in the repository and ask the author before commercial use.
• Want to weigh effort against payoff first: PAROL6 from the same team takes a similar approach, with a livelier community and more mature documentation — read up before deciding.

🔗 Official resources

• Official documentation siteNine chapters: design decisions / assembly / printing / electronics / code / first power-up / robot data
• Robot DataL1~L5 links, per-joint reduction ratios, timing belt specs
• Printing guideMaterials, printers and print settings
• Electronics · distribution PCB V2 and first power-upParts list, wiring and power-up procedure
• Code · firmware and upper-layer notesHow the three Teensy firmware variants differ
• GitHub · Faze4-Robotic-armSTL_V2 / cycloidal disc STEP / BOM / assembly manual / distribution PCB
• Same team · PAROL6A similar six-axis arm from the same author, with a more active community

❓ FAQ

✅ Next steps

Back to project 01 in the Build Your Own Robot section, which links to this guide, the GitHub design files and the distribution PCB documentation. Suggested order: read the design decisions → copy the robot data → print and verify key parts → assemble joint modules and accept one by one → power up the V2 distribution PCB → bring up Teensy + ROS / Matlab → iterate in simulation with the URDF.