LeRobot Arms (SO-100/SO-101)

Home > Hardware > Arms > LeRobot Arms

Overview

LeRobot SO-100/SO-101 is an open-source robot arm jointly developed by HuggingFace and TheRobotStudio. With parts costing approximately $130 and a 3D printing-based design, it provides an accessible robotics research platform for everyone.

Item	Details
Development	HuggingFace + TheRobotStudio
License	Apache 2.0
Type	Low-cost 3D printed robot arm
DoF	6 (Base, Shoulder, Elbow, Wrist Flex, Wrist Roll, Gripper)
Parts Cost	~$130 (single arm) / ~$230 (Leader-Follower pair)
Release Date	SO-100: 2024, SO-101: May 2025

Key Significance

Democratization of Robotics Research

The most important contribution of LeRobot SO-100/SO-101 is dramatically lowering the barrier to entry for robotics research.

Ultra-low-cost Hardware: While existing research robot arms cost thousands to tens of thousands of dollars, this can be completed for $130
Fully Open Source: CAD files, BOM, firmware, and learning framework all publicly available
3D Printing Based: Manufacturable with standard FDM printers without industrial equipment
Community Ecosystem: Dataset/model sharing through HuggingFace Hub

This enables university research labs, startups, and individual researchers to participate in VLA (Vision-Language-Action) research. Especially with SmolVLA integration, end-to-end robotics learning is possible even on consumer-grade GPUs.

Detailed Comparison: SO-100 vs SO-101

Item	SO-100	SO-101
Release	2024	May 2025
Status	Deprecated	Currently recommended version
Wiring	Wire breakage issues at Joint 3	Improved wiring prevents breakage
Assembly	Gear removal required	No gear removal, simplified assembly
Leader Arm Motors	Same gear ratio used	Optimized gear ratios per joint
Range of Motion	Limited due to wiring	Limitations resolved with wiring improvements
RL Support	Limited	Leader can follow Follower (for RL intervention)
Code Compatibility	-	Compatible with SO-100 code

Key Improvements in SO-101

Wiring Stability: Complete resolution of cable breakage issues at Joint 3
Ease of Assembly: Can assemble directly without gear removal
Leader Arm Optimization: Differentiated gear ratios per joint for self-weight support and manipulation ease
Reinforcement Learning Support: Leader arm can follow Follower arm in real-time for human intervention

Hardware Specifications

Source and Verification Date: Specifications below are based on Feetech Official Specifications and LeRobot SO-101 Documentation (verified May 2025). Check official documentation for latest information.

STS3215 Servo Motor Specifications

Item	7.4V Version	12V Version
Stall Torque	16.5 kg.cm (6V) / 19.5 kg.cm (7.4V)	30 kg.cm
Rated Torque	5 kg.cm	-
Encoder	12-bit magnetic encoder (0-4096)	12-bit magnetic encoder
Rotation Range	360-degree continuous rotation, multi-turn support (±7 turns)	Same
Communication	TTL Serial Bus (bidirectional)	TTL Serial Bus
Gearbox	Metal gears	Metal gears
Features	Daisy chain connection, overcurrent protection, temperature/voltage/current feedback	Same

Leader Arm Motor Configuration (SO-101)

Joint	Motor ID	Gear Ratio	Purpose
Base (Shoulder Pan)	1	1/191	Base rotation
Shoulder Lift	2	1/345	Shoulder lift
Elbow Flex	3	1/191	Elbow bend
Wrist Flex	4	1/147	Wrist bend
Wrist Roll	5	1/147	Wrist rotation
Gripper	6	1/147	Gripper open/close

Follower Arm Motor Configuration

All 6 motors use STS3215 with 1/345 gear ratio (uniform high-torque configuration)

Bill of Materials (BOM)

Pricing Information: Prices below are based on SO-ARM100 GitHub BOM and Seeed Studio (verified May 2025). May vary by exchange rate and supply conditions.

Leader + Follower Pair (Dual Arm Configuration)

Part	Quantity	US Price	EU Price	Notes
STS3215 Servo (7.4V, various gear ratios)	12	$166.68	€146.4	6 Leader + 6 Follower
Motor Control Board	2	$21.20	€22.8	1 per arm
USB-C Cable (2 pack)	1	$7.00	€7.00
Power Supply (5V)	2	$20.00	€31.4	Standard Edition
Table Clamp (4pc)	1	$9.00	€9.7
Screwdriver Set	1	$6.00	€9.00	Star driver needed
Total	-	~$230	~€226	3D printing cost separate

Single Follower Arm

Part	Quantity	US Price
STS3215 Servo (7.4V, 1/345)	6	$83.34
Motor Control Board	1	$10.60
USB-C Cable	1	$3.50
Power Supply	1	$10.00
Miscellaneous (clamps, driver)	1	$15.00
Total	-	~$122

Where to Buy

US: PartaBot (includes assembled products)
Global: Seeed Studio ($220 motor kit, 3D printed parts +$35)
Other: WowRobo, RobotShop

Assembly and Build

3D Printing Requirements

Item	Recommended Specs
Material	PLA+ (basic), PETG or Nylon (high-temp environments)
Nozzle	0.4mm (0.2mm layer) or 0.6mm (0.4mm layer)
Infill	15% or more (35% or more for load-bearing areas)
Support	Full, except for slopes over 45 degrees
Bed Size	Minimum 205×250mm (Prusa) or 220×220mm (Ender)

Recommended Printers: Prusa MINI+, Creality Ender 3, Bambu Lab series

Outsourcing Cost Reference (EU):

Follower arm only: ~€50
Leader + Follower pair: ~€105

Assembly Process

Motor Pre-setup: Set unique ID and baudrate for each motor
Leader Arm Assembly: Sequential assembly of joints 1-6, motor horn attachment
Follower Arm Assembly: Same process, motor installation order important
Wiring Connection: Connect motors in daisy chain method
Calibration: Set center point and range of motion for each joint
Camera Mounting (optional): Camera mount on Follower arm

Power Supply Notes

Edition	Leader Arm	Follower Arm
Standard	5V	5V
Pro	5V	12V

Warning: Mixing up Leader and Follower voltages risks servo motor damage

Calibration

Calibration is essential for transfer learning of neural networks. For policies learned on one robot to work on another, they must return the same values at the same physical positions.

# Follower arm calibration
lerobot-calibrate \
    --robot.type=so101_follower \
    --robot.port=/dev/ttyACM0 \
    --robot.id=my_follower

# Leader arm calibration
lerobot-calibrate \
    --teleop.type=so101_leader \
    --teleop.port=/dev/ttyACM1 \
    --teleop.id=my_leader

LeRobot Ecosystem

LeRobot Framework

LeRobot is a PyTorch-based robotics learning framework developed by HuggingFace.

Installation Method 1: PyPI (Recommended)

# Install from PyPI (includes Feetech SDK)
pip install "lerobot[feetech]"

# CLI commands (default settings)
lerobot teleoperate --robot.type=so101
lerobot record --robot.type=so101 --repo-id=your-username/your-dataset
lerobot train --policy.type=act --dataset.repo_id=your-username/your-dataset

# Explicit port/camera specification (when multiple devices or defaults don't work)
lerobot teleoperate --robot.type=so101 --robot.port=/dev/ttyACM0
lerobot teleoperate --robot.type=so101 \
    --robot.port=/dev/ttyACM0 \
    --robot.cameras="{'cam': {'type': 'opencv', 'index': 0}}"

Note: --robot.port can be verified with lerobot-find-port command. Specify index with --robot.cameras if there are multiple cameras.

Installation Method 2: Source Build (for development/customization)

# Clone and install from source
git clone https://github.com/huggingface/lerobot.git
cd lerobot
pip install -e ".[feetech]"

# Run scripts directly
python lerobot/scripts/control_robot.py teleoperate \
    --robot.type=so101 \
    --robot.cameras="{'cam': {'type': 'opencv', 'index': 0}}"

python lerobot/scripts/control_robot.py record \
    --robot.type=so101 \
    --repo-id=your-username/your-dataset

python lerobot/scripts/train.py \
    --policy.type=act \
    --dataset.repo_id=your-username/your-dataset

HuggingFace Hub Integration

Dataset Sharing: 487+ community datasets public with lerobot tag
Model Sharing: Upload/download trained policy models
10M+ Frames: Diverse environment data used for SmolVLA pretraining

Supported Policy Architectures

ACT (Action Chunking with Transformers): Basic Imitation Learning
Diffusion Policy: Diffusion model-based policy
SmolVLA: Vision-Language-Action model
TDMPC: Model-based RL

VLA Research Applications

SmolVLA Integration

SmolVLA is a lightweight Vision-Language-Action model with 450M parameters, providing optimized integration with SO-100/SO-101.

Item	Details
Parameters	450M
Training Data	10M frames (487 community datasets)
Hardware Requirements	Consumer-grade GPU, MacBook support
Inference Speed	30% improvement with async inference, 2x throughput

Performance on SO-100/SO-101

SmolVLA’s community dataset pretraining improved success rate from 51.7% to 78.3% on SO-100 physical robot benchmarks.

Source: SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics (arXiv:2506.01844, June 2025). The figures report 51.7% success rate with task-specific training only, 78.3% with community dataset pretraining added.

# SmolVLA fine-tuning example
from lerobot.common.policies.smolvla import SmolVLAPolicy

policy = SmolVLAPolicy.from_pretrained("lerobot/smolvla_base")
# Fine-tune with custom dataset

Research Use Cases

Imitation Learning: Demo data collection via Leader-Follower teleoperation
Reinforcement Learning: Human-in-the-loop RL with Leader following Follower functionality
Sim2Real: Transfer from Isaac Sim/Lab simulation learning to real robot
Multimodal Learning: End-to-end learning of camera + language commands + actions

Dual Arm Mobile Robot Configuration

Component	Quantity	Purpose
SO-101 Arms	2	Bimanual manipulation
LeKiwi Base	1	Mobile platform
Anker 300Wh Battery	1	Power
Wrist RGB Cameras	2	Visual feedback
Head Depth Camera (2-DoF neck)	1	Environment perception
Total Cost	-	~$660

Troubleshooting

Common Issues

Problem	Solution
Motor not recognized	Check power/USB cables, set Waveshare board to ‘B’ channel jumper
Calibration failed	Move each joint slowly through full range of motion
Insufficient torque	Upgrade Follower to 12V power supply (Pro edition)

Linux USB Permission Setup (Recommended Method)

USB serial device access permission issues are best resolved by adding to the dialout group:

# Add user to dialout group (recommended, permanent, safest)
sudo usermod -aG dialout $USER
# Log out and log back in for changes to take effect

Using udev Rules (Alternative)

For more fine-grained control over specific devices, you can use udev rules:

# 1. Check USB adapter Vendor/Product ID
lsusb

# Or check with actual device path (replace ttyUSB0, ttyACM0 with your device)
udevadm info -a -n /dev/ttyUSB0 | grep -E "idVendor|idProduct"
# Or
udevadm info -a -n /dev/ttyACM0 | grep -E "idVendor|idProduct"

# 2. Create udev rule (dialout group ownership, 0660 permissions)
# CH340 adapter example (1a86:7523) - replace with your adapter ID
echo 'SUBSYSTEM=="tty", ATTRS{idVendor}=="1a86", ATTRS{idProduct}=="7523", GROUP="dialout", MODE="0660"' | \
  sudo tee /etc/udev/rules.d/99-lerobot-serial.rules

# 3. Apply rules
sudo udevadm control --reload-rules
sudo udevadm trigger

Note:

MODE="0666" or chmod 666 makes devices writable by all users and is not recommended for security reasons.

udev rules with GROUP="dialout" require user to be in dialout group for access. Run the usermod -aG dialout command above first.

Common USB-Serial Adapter IDs:

CH340/CH341: 1a86:7523
CP210x: 10c4:ea60
FTDI: 0403:6001

Finding USB Ports

lerobot-find-port
# Follow prompts to disconnect/connect cables to identify ports

LeRobot Arms (SO-100/SO-101)

Overview

Key Significance

Democratization of Robotics Research

Detailed Comparison: SO-100 vs SO-101

Key Improvements in SO-101

Hardware Specifications

STS3215 Servo Motor Specifications

Leader Arm Motor Configuration (SO-101)

Follower Arm Motor Configuration

Bill of Materials (BOM)

Leader + Follower Pair (Dual Arm Configuration)

Single Follower Arm

Where to Buy

Assembly and Build

3D Printing Requirements

Assembly Process

Power Supply Notes

Calibration

LeRobot Ecosystem

LeRobot Framework

Installation Method 1: PyPI (Recommended)

Installation Method 2: Source Build (for development/customization)

HuggingFace Hub Integration

Supported Policy Architectures

VLA Research Applications

SmolVLA Integration

Performance on SO-100/SO-101

Research Use Cases

Dual Arm Mobile Robot Configuration

Troubleshooting

Common Issues

Linux USB Permission Setup (Recommended Method)

Using udev Rules (Alternative)

Finding USB Ports

References

Official Documentation

SmolVLA

Where to Buy

Community

See Also