History

2024–2025

• 2024-04: repository started.
• 2025-02: the Julia package implements the stock ATOM protocol from scratch. It shows that pymycobot is not the bottleneck.
• 2025-02: TORA.jl branch hf/mycobot solves a circle and plays it at 20 Hz through the stock ATOM.
• 2025-04: elephantrobotics/myCobot#53 reports 20 ms per get_angles (50 Hz). No replies.
• 2025-04 to 11: a byte-by-byte frame reader and retries (workarounds for the latency). Never committed; kept in a git stash.
• 2025-11: transponder mode is not supported. A plan to find the servo bus on the base pins; no results recorded.

2026-10-03

• The FT232R latency timer (16 ms) causes most of the 20 ms. 1 ms gives 8.5 ms.
• The servos are Feetech STS. The laptop can read and write the servo bus directly.
• Velocity mode on J1 works.
• The goal-position rule is corrected: goal = present, not present + offset. Goal speed 0 means no motion.
• A 300 Hz closed loop from the laptop. Smooth multi-joint motion. Lag of 30–120 ms per joint.
• Kinematics moves from hand-written Python to RigidBodyDynamics.jl in Julia (the hand-written version was deleted).
• The circle: 12.5 mm → 5.1 mm RMS with lag compensation.
• Offline analysis: the circle’s error peaks are J2/J3 sticking after reversals; a delay + first-order model per joint; the load register looks like PWM duty.
• The servo bus layer and the player move to Julia, with tests on a simulated bus.
• Iterative learning control is implemented.

2026-10-04

• The Julia player at full rate loses SYNC WRITEs. A 1 ms gap fixes it: 5.0 mm at 284 Hz.
• ILC on the robot: 5.0 → 2.5 → 1.3 → 0.8 mm RMS in three runs.
• With the ATOM removed, the servos still answer: the base bridges the bus.
• The stock ATOM firmware is backed up (4 MB, verified).
• The bus probe firmware finds the ATOM’s bus pins (G19 RX, G22 TX) and measures 1.26 ms per read on the ATOM.
• The controller firmware plays the circle at 500 Hz onboard: 5.0 mm, then 1.0 mm after three ILC runs. The IMU shows vibration bursts at joint reversals that ILC does not change.
• This documentation site replaces the Markdown notes in docs/.

2026-10-06

• One robot description: the URDF (xacro), joint_limits.yaml (ros2_control format) and servos.yaml in mycobot_description/. tools/gen_robot.py writes the copies for the firmware, the Control page, the simulator and Julia. See Robot description.
• J5 encoder correction in the Julia package: 3.7° → 0.18° peak to peak (gyro sweeps over ±148°). The IMU fit of the static poses improves from 0.84° to 0.48° RMS with it.
• IMU calibration from static poses and gyro sweeps: the pitch zero offsets are good (0.2–0.3°), the table tilts 0.54°, J5’s encoder is off by up to 4.4°, and J2–J4 have 1–1.9° of play outside the encoders. See IMU calibration and encoder errors.

Decisions

Decision	Reason
Kinematics only through RigidBodyDynamics.jl	No hand-written kinematics. One model (the URDF) for planning, tracing and the viewer.
Planning stays on the laptop; the ATOM plays and records	The ATOM has the timing; the laptop has the compute.
Plans are stored on the ATOM, not streamed	A WiFi drop cannot interrupt a run. Streaming is on the roadmap for MPC.
The ATOM holds the pose at power-up with goal speed 0	No motion until a plan sets a speed.
The button is not an emergency stop	It moves with the end effector.
Firmware backups and WiFi credentials stay outside the repository	Proprietary firmware and secrets.
One robot description in ROS formats (URDF + xacro, ros2_control joint_limits), with our own servos.yaml only for what no format has	One source for every copy; ROS tools can read it (2026-10-06).