The SO-101 is an open-source 6-DOF robotic arm developed by The Robot Studio in collaboration with Hugging Face, designed as a low-cost teleoperation platform for imitation learning and autonomous manipulation research. I built a full leader-follower pair (~$230 USD total) from 3D-printed PLA+ structural components and Feetech STS3215 bus servos, then brought it up from scratch through assembly, wiring, calibration, and software integration.

Each arm runs 6 Feetech STS3215 servos (16.5 kg·cm stall torque at 7.4V) driven over a Waveshare serial bus board via USB-C. The leader arm is held by hand to record demonstrations; the follower mirrors those movements in real time. This teleoperation pipeline feeds directly into Hugging Face's LeRobot library for dataset collection and end-to-end policy training.

Current work: mechanical assembly and cable routing, joint homing and safe range calibration, and real-time Python control via LeRobot. Next phase: collecting manipulation demonstrations, training an ACT or diffusion policy, and evaluating autonomous task execution on pick-and-place objectives.

• 6x Feetech STS3215 bus servos per arm (7.4V, 16.5 kg·cm stall torque)
• Waveshare serial bus servo driver board, USB-C host interface
• 3D-printed PLA+ structure (0.2mm layer height, 15% infill)
• Leader + follower configuration for teleoperation data collection

• LeRobot library (Hugging Face) for teleoperation, dataset recording, and policy training
• Python control loop for real-time servo commanding and joint state feedback
• Joint calibration pipeline establishing homing offsets and safe motion envelopes

Making it autonomous by training it to do tasks. Using the leader arm to record teleoperation demonstrations, then training a learned policy on that data so the follower can execute the task on its own.