Sourccey Specifications

Builder-first coverage of the mechanical, electrical, and software stack.

Mobility

4 mecanum wheels

Omnidirectional drive, 60 rpm max

Arms

5 DOF + gripper

Per arm, 6 revolute joints

Compute

Raspberry Pi 5

5.1V regulated, PCIe USB hat

Vision

120 degree FOV

U20CAM-720p, 30 fps

Power

12V 10Ah LiFePO4

Up to 5000 cycles

Footprint

414 mm diameter

1030 mm tall, 15.88 kg

Detailed Specifications

Each module below includes the current configuration, materials, and architecture. Items marked TBD are awaiting final validation.

Motion

Mechanical Architecture

Mobility, linear actuation, and arm geometry.

Wheels: 4 mecanum wheels, 12 kg-cm torque, 60 rpm max speed, 96 mm diameter. Open-loop PWM, no encoders, motor driver model TBD.
Linear Actuator: 12V 2A, 100N load.
Arms: 6 revolute joints per arm. 5 DOF plus 1 DOF gripper.

Joint Torque: Shoulder yaw: 30 kg-cm (2.9 N-m)
Shoulder pitch: 50 kg-cm (4.9 N-m), 1.5x effective via planetary gearbox
Elbow pitch: 50 kg-cm (4.9 N-m)
Wrist pitch: 30 kg-cm (2.9 N-m)
Wrist roll: 30 kg-cm (2.9 N-m)
Gripper: 30 kg-cm (2.9 N-m)
Joint Lengths: Base to shoulder yaw: 50 mm
Shoulder yaw: 70 mm
Shoulder pitch: 157.5 mm
Elbow pitch: 160 mm
Wrist pitch: 73 mm
Gripper: 110 mm
Motor Interfaces: Shoulder yaw: M3x8. Elbow pitch: M3x16. Screwed into motor horn.

Fabrication

Mechanical Construction

Print specs, materials, and structural design.

3D Printer: Bambu P1S, 0.4 mm nozzle.
Primary Structure: PLA. Cosmetic parts: 10% infill, crosshatch. Load bearing: 20-40% infill, gyroid. Layer height 0.08-0.28 mm.
Load Paths: Servo horn -> 3D printed link -> motor housing at end of link.

Fasteners: Metric steel: M2 (wire clips), M2.5 (camera), M3 (all else).

Service

Assembly & Serviceability

Designed for repairability and part swaps.

Tools: Electric screwdriver. No adhesives required for structural joints.
Optional: Loctite for mecanum wheel set screws.
3D Printed Parts: Public STLs allow replacements and open-source modifications.

Servos: Individually addressable and replaceable with FD software.

Actuation

Arm Actuation System

Smart servos, bus control, and end effector.

Actuator Type: Smart serial servo motors: Feetech STS3215 and STS3250.
Bus Adapter: Waveshare Bus Servo Adapter (A) V1.1.
Control Mode: Absolute position control with configurable velocity and torque limits.

Backdrivability: Limited due to high resistance under load.
End Effector: Single actuated gripper, PLA structure with 5% gyroid TPE pad.

Sensing

Sensing & Feedback

Core motion feedback loops.

Joint Position: Absolute encoder 0-4095 integrated in Feetech motors.
Velocity: Derived from position feedback.
Current / Load: Servo-reported current.

Electronics

Compute & Electronics

Processing, I/O, and networking.

Main Processor: Raspberry Pi 5, 5.1V regulated through Pi 5 power HAT.
Attachments: PCIe to USB HAT, cooling fan.
Motor Control Bus: Daisy-chained UART, 1,000,000 baud, per-servo unique ID (1-6 left, 7-12 right).

I/O: USB (4 cameras, 1 speaker, 2 motor drivers, 1 LCD), GPIO (wheel motors, actuator sensing, battery detection).
Networking: Local Wi-Fi or direct-to-Pi access point.

Perception

Vision System

Camera placement and capture specs.

Camera: Innomaker U20CAM-720p, 30 fps, 360x240 default, 640x480 max, USB 2.0.
Field of View: 120 degree FOV.
Placement: Fixed head mount angled 20 degrees down and end effector mount.

Synchronization: Software timestamping with joint state data.

Power

Power System

Energy storage, regulation, and protection.

Battery: 12V 10Ah LiFePO4, approx 5000 cycles.
Charging: 2A and 5A 12V DC chargers.
Regulation: Buck converter, central power rail with local regulation for compute and peripherals.

Protection: Per-rail current limiting (board dependent). Burnout protection TBD.
Power Consumption: Min and max consumption TBD.

Software

Software & AI

Training, data capture, and onboard intelligence.

Software Stack: Sourccey desktop and Lerobot-Vulcan tooling (software stack details in progress).
Starting AI: XVLA with 4 micromodels: folding T-shirts, shorts, jeans/pants, and long shirts.
External AI: Capabilities scale with the host computer. Rented compute is planned for users who need stronger training or inference.

Data Specs: Image data: 30 fps, RGB, configurable resolution (default 360x240)
Motor data: positional only (observation and commanded position)
Storage: .mp4 for images, .parquet for motor data, JSON metadata

Control

Interfacing & Control

On-robot UI, desktop tools, and teleoperation.

On-Robot Interface: 7-inch 1024x600 touchscreen with custom firmware and controls.
Desktop Software: Downloadable app with features similar to the touchscreen UI.
Teleoperators: Two arms with Feetech motors (15 kg-cm torque), calibrated via zeroing from set position.

Oculus Quest: Inverse kinematics maps controller position to end effector pose.

Calibration

Calibration & Configuration

Autocalibration, zeroing, and config storage.

Joint Limits: Autocalibration moves to mechanical limits detected via current feedback.
Zeroing: Absolute encoder-based, with homing and quick preset zeroing.
State Representation: Joint angles derived from motor encoders.

Configuration Storage: JSON per arm. Robot calibration files on Raspberry Pi, teleoperator calibration on host computer.

Physical

Physical Characteristics

Dimensions and mass.

Footprint: 414 mm diameter. Handles 25 mm. Arms at side +162.5 mm, arms out +622.5 mm.
Height: 1030 mm (40.5 in).
Mass: 15.88 kg.

Environment

Environmental

Operating ranges and physical requirements.

Temperature: 0-40 C operating ambient temperature.
Humidity: Non-condensing, <80% (based on Raspberry Pi 5 specs).
Physical Environment: Even floor required. Shallow carpets ok; hardwood or tile recommended.

Clearance: Max table height shoulder can clear: 775 mm (30.5 in).

Safety

Safety & Constraints

Limits, protections, and known guardrails.

Limits: Torque, velocity, and acceleration limits configurable per motor.
Material Strategy: Low-mass PLA parts reduce injury risk; limits enforced in software and servo firmware.
Fault Handling: Overcurrent protection and off switch. Fuse and pinch zone guidance TBD.

Known Limitations

Payload limits vary by pose and reach.
Wi-Fi dependency (DFS considerations).
Calibration relies on current-based limit detection and may drift with wear.
Threading strategy not standardized; repeated fastener cycles may degrade printed threads.
Performance metrics not fully characterized (repeatability, accuracy, latency).

In Progress

We are actively validating performance metrics and long-run durability. This section will expand with repeatability, motor cycle life, and end-to-end latency benchmarks once testing is complete.

Updates forthcoming

FAQ

Answers for builders, tinkerers, and curious humans.

Can't find what you need? Reach out and we'll help you design your build path.

Sourccey is Vulcan Robotics' personal home robot. It is fully open source, trainable, and built to help you learn robotics and AI while teaching it real-world tasks through demonstration.

Yes. Hardware, electrical, and software designs are open source. You can inspect, modify, and contribute through our public repositories on GitHub.

The main processor is a Raspberry Pi 5, powered by a regulated 5.1V rail with a PCIe-to-USB hat and cooling fan for peripherals.

You can use the onboard 7-inch touchscreen, the desktop software, or teleoperate using dedicated arms. Oculus Quest input is supported through inverse kinematics that maps controller position to the end effector.

Yes. Sourccey uses an up-to-date Lerobot-Vulcan fork for training and control. Learn more on the Lerobot-Vulcan repository.

Yes. Public STLs let you print replacements, and servos are individually addressable for quick swaps. The entire mechanical stack is designed with serviceability in mind.

Still have questions?

Join the community or browse the repos for deeper technical details.

Join Discord View GitHub