Cool SO-ARM101 Projects to Try Now: From Grasping to Imitation Learning
Looking for hands-on ways to explore SO-ARM101, LeRobot, and embodied AI? These six open-source projects take you from basic robotic arm teleoperation and dataset collection to imitation learning, wireless control, dual-arm manipulation, and reinforcement learning with Isaac Lab.
Whether you’re new to LeRobot or ready to build a mobile household robot, this guide ranks the projects by difficulty so you can find the right place to start.
1. Project 1: Teleoperation, Record and Visualize Dataset
Difficulty: ★★☆☆☆
Wiki Tutorial: https://wiki.seeedstudio.com/lerobot_so100m_new/#visualize-dataset
Start with the basics of robotic arm teleoperation and data collection.
In this project, an SO-ARM101 leader arm controls a follower arm while a USB camera records the task. You’ll learn how to connect the arm to a PC, perform simple teleoperation tasks, record demonstrations with LeRobot, and visualize the resulting dataset.
It’s a practical introduction to the complete data pipeline behind imitation learning—from robot control to dataset recording and visualization.
What You’ll Learn
- SO-ARM101 leader-follower teleoperation
- LeRobot dataset recording
- Camera-based visual feedback
- Dataset visualization
- The basics of imitation learning data collection
Required Equipment:
| Part | Quantity |
| ×1 | |
| × 1 | |
| PC | ×1 |
2. Train Your Own Imitation Learning Model: ACT, SmolVLA, Pi0.5, and GR00T
Difficulty: ★★★☆☆
Wiki Tutorial: https://wiki.seeedstudio.com/lerobot_so100m_new/#train-and-evaluate
Ready to move beyond teleoperation? Use your SO-ARM101 dataset to train an imitation learning policy with LeRobot.
This project introduces six leading models, including ACT (Action Chunking with Transformers), SmolVLA, Pi0.5, and GR00T N1.5. You’ll collect demonstrations, prepare the dataset, configure a model, train the policy, and evaluate it directly on the robot.
Along the way, you’ll see how transformer-based policies predict robot actions and how VLA models connect visual observations with language instructions.
What You’ll Learn
- Demonstration data collection
- Dataset preprocessing
- ACT and VLA model training
- Policy evaluation on SO-ARM101
- Vision-based robotic manipulation
Required Equipment:
| Parts | Quantity |
| ×1 | |
| × 1 | |
| PC | ×1 |
3. Dual-Arm Imitation Learning with SO-ARM101
Difficulty: ★★★★☆
Wiki Tutorial: https://wiki.seeedstudio.com/lerobot_double_arm_so_arm_training
Video Tutorial:
Single-arm manipulation is only the beginning. With two SO-ARM101 arms and LeRobot, you can explore synchronized dual-arm teleoperation and bimanual imitation learning.
The project focuses on collecting synchronized demonstrations from two arms and training policies for tasks that require coordinated manipulation. Think folding clothes, washing dishes, or assembling objects—tasks that are difficult to perform with a single arm.
A multi-camera setup provides a third-person view, while leader-follower teleoperation captures coordinated movements from both arms.
What You’ll Learn
- Dual-arm teleoperation
- Synchronized data collection
- Bimanual coordination
- Multi-camera setups
- Dual-arm imitation learning
Required Equipment:
| Parts | Quantity |
| ×2 | |
| × 2（根据任务难度选择Quantity） | |
| PC | ×1 |
| Third-Person Camera Mount | ×1 |
| ×1 (optional, for improved data collection environment) |
4. Wireless SO-ARM101 Teleoperation with XIAO ESP32-C3 and micro-ROS
Difficulty: ★★★☆☆
Wiki Tutorial: https://wiki.seeedstudio.com/soarm101_xiao_microros_wireless_teleoperation
Want to remove the cables? This project turns the SO-ARM101 into a wirelessly controlled robotic arm using the Seeed Studio XIAO ESP32-C3 Bus Servo Adapter and micro-ROS.
The leader arm connects to an Ubuntu PC, while the follower receives commands over Wi-Fi using ROS 2 and micro-ROS.
This makes the setup more flexible for applications where the arm is mounted on a mobile robot, robotic dog, or other moving platform.
What You’ll Learn
- Wireless robotic arm control
- XIAO ESP32-C3
- micro-ROS
- ROS 2 communication
- Wi-Fi-based teleoperation
Required Equipment
| Parts | Quantity |
| ×1 | |
| ×1 | |
| PC（Ubuntu 22.04 ） | ×1 |
| 2.4 GHz Wi-Fi network | ×1 |
5. Build a Household Dual-Arm Mobile Robot
Difficulty: ★★★★★
Opensource Project — Bring Embodied AI to Every Family Around the World!
GitHub: https://github.com/Vector-Wangel/XLeRobot
What if your SO-ARM101 could leave the desktop and move around your home?
The XLeRobot project combines two SO-ARM101 arms with a mobile base to create a household dual-arm robot. The platform integrates a Lekiwi mobile base, an Orbbec depth camera, and two robotic arms for mobile manipulation.
The result is a much more ambitious embodied AI platform capable of combining navigation, perception, and bimanual manipulation. Example applications include fetching drinks, watering plants, and cleaning surfaces.
What You’ll Learn
- Mobile manipulation
- Dual-arm control
- SLAM navigation
- 3D vision
- Embodied AI on mobile robots
Required Equipment
The original project provides a complete BOM for building the robot from individual components. Pre-assembled kits can also be used to reduce assembly time.
| Parts | Quantity | |
| Body Parts | ×2 | |
| ×1 | ||
| Motors：STS3215 Servo (12V) | ×5（3 for base + 2 for head.） | |
| ×1 | ||
| ×1 | ||
| ×1 | ||
| Assembly Tools | M3 Screws and Nuts Set | ×1 |
| Flush cutter | ×1 | |
| Screwdriver Set | ×1 | |
| Electronics | PC / Raspberry Pi 5 | ×1 |
6. Reinforcement Learning with SO-ARM100 / SO-ARM101 in Isaac Lab
Difficulty: ★★★★★
Github：https://github.com/MuammerBay/isaac_so_arm101
Ready to move from imitation learning to reinforcement learning (RL)?
This open-source project brings SO-ARM100 and SO-ARM101 into NVIDIA Isaac Lab, allowing you to train robot policies entirely in simulation.
Instead of collecting large amounts of real-world demonstration data, you can define tasks such as reaching and grasping, design reward functions, and train policies across thousands of simulated environments.
The project also introduces key robotics concepts such as PPO, domain randomization, and sim-to-real transfer. GPU-accelerated simulation makes it possible to run large-scale RL experiments on a local workstation.
Hardware note: An NVIDIA RTX GPU with 16–24 GB VRAM is recommended, along with 32 GB of system RAM.
Required Equipment:
| Equipment | Quantity |
| PC with NVIDIA RTX GPU | ×1 |
Which SO-ARM101 Project Should You Start With?
These projects cover a wide range of robot learning workflows:
| Project | Difficulty | Main Focus |
| Teleoperation & Dataset Recording | ★★☆☆☆ | Teleoperation + LeRobot |
| Imitation Learning | ★★★☆☆ | ACT + VLA |
| Dual-Arm Imitation Learning | ★★★★☆ | Bimanual manipulation |
| Wireless Teleoperation | ★★★☆☆ | ROS 2 + micro-ROS |
| Household Mobile Robot | ★★★★★ | Mobile manipulation |
| Reinforcement Learning | ★★★★★ | Isaac Lab + RL |
About Author
