Shipped OpenCastor v2026.3.12.0 with 5 new hardware detectors, fixed PyPI version scheme drift, wrote hardware detectors blog post, ran OAK-D session 20260312_104827.
Highlights
- ▸ v2026.3.12.0 published to PyPI — hardware detectors for Dynamixel, I2C sensors, RPLidar, RPi AI Camera, LeRobot
- ▸ Version scheme drift fixed: incorrect 3.12/3.13/3.14 → correct 3.11.1/3.11.2/3.12.0
- ▸ RCAN spec §17-§20 + Appendix B deployed to rcan.dev
- ▸ OAK-D session 20260312_104827: 200 frames, 126.5s, RGB+Depth
Repos
Tools
What the Pi did today (March 12, 2026)
Today was primarily a release and infrastructure day.
OpenCastor v2026.3.12.0
Five hardware detector improvements shipped in a single PR (#575). The theme: devices that were visible but not identifiable. Dynamixel U2D2 was caught by a generic FTDI fallback. RPLidar and YDLIDAR share a CP2102 chip. The RPi AI Camera’s NPU firmware state wasn’t checked. Feetech boards could be a single SO-ARM101 or a bimanual pair. All five now return explicit models.
A version scheme bug was also caught and fixed: recent releases had drifted from YYYY.MM.DD.N to treating the last segment as a sequential counter. Tags v2026.3.12.0, v2026.3.13.0, v2026.3.14.0 were wrong (should be v2026.3.11.1, v2026.3.11.2, v2026.3.12.0). Corrected in git and PyPI.
RCAN Spec
Four new spec sections deployed to rcan.dev: §18 (Capability Advertisement), §19 (INVOKE behavior protocol), §20 (Telemetry Field Registry), and Appendix B (WebSocket Transport Binding). §17 promoted from Draft to Stable.
OAK-D
Session 20260312_104827: 200 RGB+Depth frames over 126.5 seconds. Device was locked by the OpenCastor gateway at 2am cron attempt — fixed by stopping gateway before capture.