Build Log
Pool sessions, hardware failures, software milestones. What we built, what broke, and what we learned. 11 entries.
Pool session 8 complete. 2h 40m runtime. No leaks. Gate detection held above 88% confidence for the full session.
Replaced O-ring on the rear endcap before the weekend pool session. No signs of prior failure, but it had been 11 dives since last swap.
HARDWARE Update
IMU Calibration Complete
48 calibration passes over three days. Drift is now within 2.1% across all axes under simulated load.
We were seeing a consistent 4-6% yaw drift during hard thruster transitions. Traced it to magnetic interference from the ESC wiring bundle running too close to the IMU board. Rerouted the bundle 40mm to port and the drift dropped immediately.
Ready for water trials.
Milestone: Thruster Array Commissioned
All eight T200 thrusters calibrated and within 1.8% of target thrust across the array. This closes the propulsion milestone for RoboSub 2026.
Calibration took eleven days across five pool sessions. The main hold-up was the port-aft thruster, which was showing intermittent dropout at throttle above 60%. Connector issue, not the thruster itself. Replaced the JST connector and it has been clean since session 6.
From here we move to integrated navigation testing. The state machine can now command thrust with known response curves. Depth hold and heading hold will be tuned against real thrust data rather than theoretical values.
Port-aft thruster showing intermittent dropout above 60% throttle. Connector, not the motor. Flagged for replacement before next session.
SOFTWARE Update
Gate Detection at 91%
YOLOv11 gate detection model hit 91% confidence consistency in static pool conditions. That is above our 88% threshold for the competition run.
The improvement came from retraining on 340 additional frames captured during session 5, where lighting conditions were more variable. The previous model was overtrained on controlled indoor conditions and degraded in outdoor light.
Next: test against the orange buoy task. The buoy model is currently at 79% and needs another training pass before we can rely on it.
Dispatch: Pool Trial 3 - Field Notes
Milestone: Frame Assembly Complete
Sub-Epsilon is physically built. All PVC tube sections bonded, thruster mounts installed, endcap assemblies fitted.
The frame came in at 4.2 kg dry, 1.1 kg under our weight ceiling. That margin goes straight into ballast flexibility for final depth tuning at the venue.
The 3D-printed thruster junction pieces took three iterations. The first print cracked at the mounting boss during torque testing. Increased wall thickness from 3mm to 5mm and switched infill from 15% gyroid to 40% honeycomb. The third print has held through two days of handling and dry-spin testing without complaint.
Next step: seal the electronics capsule and begin wet testing.
SOFTWARE Update
Migrated to ROS 2 Jazzy
Completed the migration from ROS 2 Humble to Jazzy on both Pi 5 boards. Build time dropped from 14 minutes to 9 minutes. No breaking changes to our node interfaces.
One dependency issue: the BlueRobotics Bar30 pressure sensor driver had an undocumented breaking change in its Jazzy package. The topic type changed from Float64 to FluidPressure. Updated our depth node to consume the correct message type. Two hours of debugging that could have been one sentence in the changelog.
First wet test in a stock tank. The electronics capsule held. No ingress after 45 minutes submerged at 0.6m depth.
Dispatch: Electronics Capsule Design Notes
End of log