feat(turret): per-slot calibration, backlash compensation, software homing, factory parameter set

[veerwang]

Summary

Two commits improving the NiMotion 4-position objective turret controller:

1. Per-slot calibration and gear backlash compensation (6918f4f)

• OBJECTIVE_TURRET_CALIBRATED_PULSES: per-slot calibrated absolute pulse targets (slot -> pulses from homing zero). Calibrated slots are used verbatim; others fall back to the theoretical (slot-1)*pulses_per_position + OBJECTIVE_TURRET_OFFSET_PULSES.
• OBJECTIVE_TURRET_BACKLASH_DEG: when > 0, every slot change overshoots below the target and approaches from below, so the final approach direction is constant and gear backlash cancels out.
• Both validated at init so a bad machine .ini fails fast; defaults keep behavior identical.

2. Software homing, DI1 origin switch, factory parameters (fffd0a5)

• Software homing replaces the driver's built-in homing modes: velocity-mode sweep toward the sensor polling the DI level → decel-stop on trigger → back off until the switch releases → fine-step back to the trigger edge → SET_ZERO there. Repeatability ±5 pulses (hardware-measured 0); ends clamped at home with holding torque. Home timeout raised to 120 s.
• DI1 permanently "origin switch" (3) — the previous scheme (DI1 temporarily remapped to negative limit + homing method 17) faults FF0E on current drive firmware whenever a normal move passes the sensor.
• Factory parameter set auto-calibrated on connect: accel/decel 1000, max speed 200, min speed 16 (written before max — fixes a min/max write-order bug), motor currents sized so a fully loaded turret does not lose steps, microstep pinned to 16 (fail fast asking for a power cycle on mismatch).
• modbus_rtu: new read_input_registers() batch read so the homing sweep sees DI + position + alarm in one frame per poll (the ~50-pulse sensor window must not be crossable between polls).

Deployment notes

• First connect on a real unit rewrites speed/current/DI parameters and saves EEPROM (one-time).
• Machines with per-slot calibration measured under the old driver-homing zero must re-measure (the zero reference moved slightly).

Test plan

• 66 tests pass (tests/control/test_objective_turret_controller.py, tests/control/test_modbus_rtu.py): calibration/backlash targeting, software-homing edge/sweep/timeout paths, factory-param write order, microstep-mismatch fail-fast
• Homing repeatability verified on hardware in the source project (0-pulse deviation between runs)
• Bench test on production hardware: startup homing, slot-switching accuracy, slot re-calibration

🤖 Generated with Claude Code

[veerwang]

Pushed fd6ee50, syncing the tuning work done on the SingleMotor bench project over 07-27/07-28 (all manager-specified values):

Homing parameter revisions

• Fine-search step 5 → 2 pulses: home repeatability tightens from ±5 to ±2 pulses
• Backoff step 150 → 60: worst-case overshoot past the trigger edge is ~49 pulses (29 stop distance + 20 detection lag), so a single backoff jog clears the sensor window; the loop remains as a fallback
• Sweep speed 50 → 60 Step/s with the DI poll period shortened 25 → 20 ms — the anti-aliasing constraint still holds (~50-pulse window crossed in 52 ms ≥ 2.6 poll periods)
• Fine-search travel limit 400 → 200 (backoff 60 + window 50 + margin)

New: temporarily lowered acceleration during fine search
home() now reads the original acceleration (0x005F), lowers it to 50 Step/s² for the fine-step approach to soften the microstep landing on the trigger edge, and restores it in the cleanup path alongside max speed. Covered by a new regression test.

Max speed 200 → 150 Step/s
The 0.95 A driver current cap leaves no torque margin at speed 200 and the loaded turret was observed losing steps. 150 × 16 microsteps = 2400 pulses/s ≈ 98°/s at the turret. Homing is unaffected (sweep 60 / jog 60 are well below).

Current parameters

• Idle current 60 → 21 (displays as 0.69 A)
• Dropped the decel-current (0x15) calibration and its "volatile" mechanism: the SDM42 drive has no such parameter — the firmware silently drops writes and always reads back 0

Not ported: SingleMotor's reconnect-polling fix is specific to its Qt status panel; this controller polls synchronously per move and doesn't have that failure mode.

These values are written on connect and persisted to EEPROM by the existing calibration path. Bench-verified on SingleMotor; re-verification on the turret hardware is still pending.