2 layers159.9 × 47.5 mm
LVP G0S4
Omar-Magdy0/ForgeDriveHW
ForgeDrive is an open-source research and development platform for modular, scalable, and industrial-inspired systems.
69 partsMIT
4 layers32 × 55 mm
Furuta pendulum
MHVandborg/furuta_pendulumA Furuta (rotary inverted) pendulum system controlled by a neural network running on a custom PCB.
- Size
- 32 × 55 mm
- Copper layers
- 4
- Thickness
- 1.6 mm
- Parts
- 67
- Nets
- 81
- Pads
- 247
- Tracks, vias
- 4, 0
- Routed
- Yes
- KiCad
- KiCad 10.0 (file 20260206)
- Created
- 24 May 2026
- Updated
- 15 Jul 2026
Overview
Furuta pendulum is an open-source printed circuit board by MHVandborg, from the MHVandborg/furuta_pendulum repository on GitHub. It is a 4-layer board of 32 × 55 mm with 67 parts on 81 nets, routed with 4 tracks and 0 vias.
It is built around the ATSAMD51J20A-AU, DRV8313PWP and AMS1117-3.3, with the USBLC6-2SC6, GZ2012D601TF and TS-1088-AR02016 among its other parts. Its parts include 18 capacitors, 17 resistors, 6 diodes, 5 connectors and 3 ICs. It belongs with the SAM D and SAM boards of this library.
Main components
9
ATSAMD51J20A-AU
TQFP-64(10x10)-40℃~+85℃ 1.71V~3.63V 120MHz 12bit 12bit 1MB 256KB 32 Bit 51 ARM Cortex-M4F Built-in FLASH TQFP-64(10x10) Microcontrollers (MCU/MPU/SOC) ROHSMicrochip TechnologyMicrocontrollers (MCU/MPU/SOC)
DRV8313PWP
HTSSOP-28-EP-4.5mmHTSSOP-28-EP-4.5mm Brushless DC (BLDC) Motor Driver ROHSTexas InstrumentsMotor Driver ICs
AMS1117-3.3
SOT-223-40℃~+125℃ 0.003%Vout 1 1.1V@(800mA) 15V 1A 3.3V 5mA 72dB@(120Hz) Fixed Over Current Protection、Short Circuit Protection、Thermal shutdown Positive SOT-223 Voltage Regulators - Linear, Low Drop Out (LDO) Regulators ROHSAdvanced Monolithic SystemsVoltage Regulators - Linear, Low Drop Out (LDO) Regulators
USBLC6-2SC6
SOT-23-60.35pF、0.8pF 100nA 150W@8/20us 15V 2 5V 6A@8/20us 6V ESD SOT-23-6 ESD and Surge Protection (TVS/ESD) ROHSUMWESD And Surge Protection (TVS/ESD)
GZ2012D601TF
0805-55℃~+125℃ 1 300mΩ 500mA 600Ω@100MHz ±25% 0805 Ferrite Beads ROHSSunlordFerrite Beads
TS-1088-AR02016
SMD,4x3mm-30℃~+80℃ 1.6N 100,000 cycles 12V 2mm 3.9mm 3mm 50mA Black Round Button SMD(SMT) Tab SPST Surface Mount,Vertical Without Bracket SMD,4x3mm Tactile Switches ROHSXUNPUTactile SwitchesGZ2012D101TF
08051 100Ω@100MHz 150mΩ 800mA ±25% 0805 Ferrite Beads ROHSSunlordFerrite Beads
SMBJ12A
SMB-65℃~+150℃ 12V 14.7V 19.9V 1uA 30.2A 390pF 600W@10/1000us TVS SMB ESD and Surge Protection (TVS/ESD) ROHS晶导微电子ESD And Surge Protection (TVS/ESD)SMBJ5.0A
DO-214AA(SMB)-65℃~+150℃ 5V 600W@10/1000us 65.3A@10/1000us 7V 800uA 9.2V TVS Unidirectional DO-214AA(SMB) ESD and Surge Protection (TVS/ESD) ROHSMDDESD And Surge Protection (TVS/ESD)
README
README.mdFuruta Pendulum
A Furuta (rotary inverted) pendulum system controlled by a neural network running on a custom PCB.
The system consists of a horizontal rotating arm driven by a BLDC motor, with a freely swinging pendulum attached at the end. The neural network controller learns to swing up and balance the pendulum using reinforcement learning trained in a physics simulation, then deployed directly onto the MCU.
End Goal — Double Furuta Pendulum
The ultimate goal is a double Furuta pendulum: motorised arm → first free pendulum → second free pendulum, all balanced simultaneously in the inverted position by a single neural network. The current hardware (PCB rev1) targets the single pendulum first as a validated stepping stone. Double Furuta requires a PCB rev2 with a third encoder.
Control Architecture
Single Furuta (PCB rev1 — current)
[AS5600 arm] [AS5600 pendulum 1]
| |
+-------+-------+
|
[Neural Network] MLP 2×64, trained with PPO
inputs: θ_arm, θ_p1, ω_arm, ω_p1
output: torque_command (−1…+1)
|
[SimpleFOC / DRV8313]
|
[GBM2804H-100T BLDC]Double Furuta (PCB rev2 — future)
[AS5600 arm] [AS5600 pendulum 1] [AS5600 pendulum 2]
| | |
+-----------+---+--------------------+
|
[Neural Network] MLP 2×64 (extended inputs), trained with SAC
inputs: θ_arm, θ_p1, θ_p2, ω_arm, ω_p1, ω_p2
output: torque_command (−1…+1)
|
[SimpleFOC / DRV8313]
|
[GBM2804H-100T BLDC]A single NN handles both swing-up and balancing via curriculum learning — it trains near vertical first, then faces progressively larger starting angles.
Repository Structure
furuta_pendulum/
├── firmware/ # Bare-metal C (ARM GCC, CMake) → firmware/README.md
├── hardware/
│ ├── pcb/ # KiCad schematic + layout → hardware/pcb/COMPONENTS.md
│ └── mechanical/ # Fusion 360 3D design files
├── neural_network/ # RL training + Gymnasium env + exported weights → neural_network/README.md
├── bootloader/ # UF2 bootloader binary for ATSAMD51J20A
└── docs/ # Datasheets and reference documentsToolchain
| Area | Tools |
|---|---|
| Firmware | ARM GCC, CMake, VS Code, Atmel-ICE + Tag-Connect TC2030 (SWD), UF2 bootloader |
| ML / Training | Python, TensorFlow / Keras, Gymnasium, PPO → SAC |
| Inference on MCU | TensorFlow Lite Micro (TFLM) |
| Mechanical | Fusion 360, PLA (FDM) |
| PCB | KiCad |
PCB Revision History
| Rev | Tag | Date | Notes |
|---|---|---|---|
| 1 | hw-rev1 | 2026-06-07 | First production run — JLCPCB. Single Furuta (2 encoders) |
| 2 | (planned) | TBD | Add AS5600 #3 + third I2C bus for double Furuta |
Further Reading
| Document | Contents |
|---|---|
| ROADMAP.md | Full phased development plan with checklists |
| hardware/pcb/COMPONENTS.md | Component choices, rationale, and PCB design decisions |
| firmware/README.md | Firmware architecture, build instructions, peripheral map |
| neural_network/README.md | RL training approach, simulation, reward design, export pipeline |
Contact
If you use or build on this project, I'd love to hear about it! Feel free to open an issue or reach out directly.
Parts
67| Qty | Value | Footprint | References |
|---|---|---|---|
| 9 | 100n | Capacitor_SMD:C_0402_1005Metric | C8, C15, C16, C18, C22, C23, C24, C27, C28 |
| 8 | 10u | Capacitor_SMD:C_0805_2012Metric | C9, C10, C12, C13, C19, C20, C21, C26 |
| 4 | 4.7k | Resistor_SMD:R_0402_1005Metric | R22, R23, R24, R25 |
| 4 | M2 | MountingHole:MountingHole_2.2mm_M2 | H1, H2, H3, H4 |
| 4 | TestPoint | TestPoint:TestPoint_Pad_D1.5mm | TP1, TP4, TP5, TP6 |
| 3 | 10k | Resistor_SMD:R_0402_1005Metric | R11, R19, R26 |
| 3 | Fiducial | Fiducial:Fiducial_1mm_Mask2mm | FID1, FID2, FID3 |
| 2 | 1k | Resistor_SMD:R_0402_1005Metric | R15, R17 |
| 2 | 2k | Resistor_SMD:R_0402_1005Metric | R14, R16 |
| 2 | 33 | Resistor_SMD:R_0402_1005Metric | R1, R2 |
| 2 | 330 | Resistor_SMD:R_0402_1005Metric | R13, R18 |
| 2 | 5.1k | Resistor_SMD:R_0402_1005Metric | R3, R4 |
| 2 | GREEN | LED_SMD:LED_0805_2012Metric | D3, D4 |
| 2 | GZ2012D601TF | Inductor_SMD:L_0805_2012Metric | FB2, FB5 |
| 2 | TS-1088-AR02016 | furuta_pendulum:SW_TS-1088-AR02016 | BTN1, RESET1 |
| 1 | 10u | Inductor_SMD:L_0603_1608Metric | L1 |
| 1 | 4.7u | Capacitor_SMD:C_0603_1608Metric | C17 |
| 1 | AMS1117-3.3 | Package_TO_SOT_SMD:SOT-223-3_TabPin2 | U1 |
| 1 | ATSAMD51J20A-A | Package_QFP:TQFP-64_10x10mm_P0.5mm | U7 |
| 1 | BLM18PG121SN1D | Inductor_SMD:L_0603_1608Metric | FB3 |
| 1 | Conn_ARM_SWD_TagConnect_TC2030 | Connector:Tag-Connect_TC2030-IDC-FP_2x03_P1.27mm_Vertical | J5 |
| 1 | DRV8313 | Connector_PinHeader_2.54mm:PinHeader_2x07_P2.54mm_Vertical | J3 |
| 1 | Encoders | Connector_PinHeader_2.54mm:PinHeader_1x06_P2.54mm_Vertical | J7 |
| 1 | GZ2012D101TF | Inductor_SMD:L_0805_2012Metric | FB4 |
| 1 | LCD_ST7735S | Connector_PinHeader_2.54mm:PinHeader_1x07_P2.54mm_Vertical | J2 |
| 1 | RED | LED_SMD:LED_0603_1608Metric | D5 |
| 1 | SMBJ12A | Diode_SMD:D_SMB | D2 |
| 1 | SMBJ5.0A | Diode_SMD:D_SMB | D1 |
| 1 | USBLC6-2SC6 | Package_TO_SOT_SMD:SOT-23-6 | U3 |
| 1 | USB_C_Receptacle_USB2.0_14P | Connector_USB:USB_C_Receptacle_HRO_TYPE-C-31-M-12 | J8 |
| 1 | YELLOW | LED_SMD:LED_0805_2012Metric | D6 |
Similar boards
- 4 layers228 × 115.4 mm
- 4 layers50 × 50 mm
STMF103C8T6 Dev Board
hackclub/OnBoard
💡 Join 1,000 teenagers and make your first PCB with a $100 grant!
52 parts915 - 4 layers252 × 92 mm
Cooksense
misko/prompt2device
The separate vNext PCB skill and orchestrator package is available alongside the reference workflow below; existing projects keep their accepted toolchain until an explicit migration.
195 parts3MIT - 6 layers90 × 89 mm
Modular processing board
igorwolfs/modular-processing-hw
MCU Breakout board with Hirose BM28B connector
177 parts1 - 6 layers133.1 × 133.1 mm
AERIS H743-PCB
7903555032/AERIS-H743-PCB
A 6-layer STM32H743-based embedded development and flight-control/data-logging PCB designed as a practical hardware-design and PCB-layout project.
114 parts1 - 4 layers68.7 × 53 mm
- 2 layers95 × 50 mm
Gw v41-plus
shachi-lab/greaseweazle-v41-plus
Greaseweazle V4.1 based compatible board design for 3.5/5.25-inch FDDs
87 parts4
Questions
How big is the Furuta pendulum PCB?
It measures 32 × 55 mm and has 4 copper layers, 1.6 mm thick.
What parts does the Furuta pendulum use?
67 parts, built around the ATSAMD51J20A-AU, DRV8313PWP, AMS1117-3.3 and USBLC6-2SC6. The full list of its parts is on this page, with links to each part's symbol and footprint where PCBKit's library has them.
Where can I download the Furuta pendulum KiCad files?
From the MHVandborg/furuta_pendulum repository on GitHub, at the commit this page was made from: the board is hardware/pcb/furuta_pendulum.kicad_pcb and its schematic hardware/pcb/furuta_pendulum.kicad_sch. PCBKit opens the board in the browser, with nothing to install.
Can I use the Furuta pendulum design in my own project?
Its repository states no licence, so its author keeps all rights: ask MHVandborg before reusing the design.