Architecture

imu_sdk and the vendored SO-ARM100 robot description feed into m5teleop, the integration point that runs the real-time teleoperation loop — the five-stage pipeline broken down step by step in Teleoperation control loop, step by step. Recording is opt-in: pass --record and m5teleop hands frames to soarm_lerobot, which buffers episodes into a LeRobotDataset for ACT / Diffusion Policy training — with no import-level dependency back the other way.

soarm_sdk and camera_calibration are otherwise standalone: the former is a parallel, self-contained Feetech servo SDK (used by its own calibration/dashboard tools, not currently wired into the teleop loop), the latter a Rerun-based camera intrinsics/ArUco toolkit with no shared dependencies at all.

soarm_tamp is the third way of moving the arm — deliberative, plan-then-execute task-and-motion planning on top of long_tamp/HPP, executing through soarm_sdk's own ServoRobot and calibration layer rather than lerobot's follower, with planning (inside an HPP container) and execution (on the host) deliberately kept in separate processes joined only by a waypoint manifest on disk.

soarm_mjlab is the newest branch — RL training of SO-ARM100 in MuJoCo via mjlab, still in progress, planned to deploy trained policies through the same soarm_sdk.RobotInterface that drives real hardware.