Implement Ziegler-Nichols relay feedback auto-tuning with flash persistence:
Firmware (STM32F722):
- pid_flash.c/h: erase+write Kp/Ki/Kd to flash sector 7 (0x0807FFC0),
magic-validated; load on boot to restore saved tune
- jlink.h: add JLINK_CMD_PID_SAVE (0x0A) and JLINK_TLM_PID_RESULT (0x83)
with jlink_tlm_pid_result_t struct and pid_save_req flag in JLinkState
- jlink.c: dispatch JLINK_CMD_PID_SAVE -> pid_save_req; add
jlink_send_pid_result() to confirm flash write outcome over USART1
- main.c: load saved PID from flash after balance_init(); handle
pid_save_req in main loop (disarmed-only, erase stalls CPU ~1s)
Jetson ROS2 (saltybot_pid_autotune):
- pid_autotune_node.py: add Ki to Ziegler-Nichols formula (ZN PID:
Kp=0.6Ku, Ki=1.2Ku/Tu, Kd=0.075KuTu); add JLink serial client that
sends JLINK_CMD_PID_SET + JLINK_CMD_PID_SAVE after tuning completes
- autotune_config.yaml: add jlink_serial_port and jlink_baud_rate params
Trigger: ros2 service call /saltybot/autotune_pid std_srvs/srv/Trigger
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Detect if MCU was reset by IWDG watchdog timeout at startup
- Log watchdog reset events to debug terminal (USB CDC)
- Store watchdog reset flag for status reporting to Jetson
- Watchdog timer configured with 2-second timeout in safety_init()
- Main loop calls safety_refresh() to kick the watchdog every iteration
The IWDG (Independent Watchdog) resets the MCU if the main loop
hangs and fails to call safety_refresh() within the timeout window.
This provides hardware-enforced detection of software failures.
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
Adds #include "bno055.h" to src/main.c to resolve implicit declaration
warnings for bno055_read(), bno055_calib_status(), and bno055_temperature().
Functions were properly implemented but header was missing from includes.
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
Add coulomb counter for accurate SoC estimation independent of load:
- New coulomb_counter module: integrate current over time to track Ah consumed
* coulomb_counter_init(capacity_mah) initializes with battery capacity
* coulomb_counter_accumulate(current_ma) integrates current at 100 Hz
* coulomb_counter_get_soc_pct() returns SoC 0-100% (255 = invalid)
* coulomb_counter_reset() for charge-complete reset
- Battery module integration:
* battery_accumulate_coulombs() reads motor INA219 currents and accumulates
* battery_get_soc_coulomb() returns coulomb-based SoC with fallback to voltage
* Initialize coulomb counter at startup with DEFAULT_BATTERY_CAPACITY_MAH
- Telemetry updates:
* JLink STATUS: use coulomb SoC if available, fallback to voltage-based
* CRSF battery frame: now includes remaining capacity in mAh (from coulomb counter)
* CRSF capacity field was always 0; now reflects actual remaining mAh
- Mainloop integration:
* Call battery_accumulate_coulombs() every tick for continuous integration
* INA219 motor currents + 200 mA subsystem baseline = total battery draw
Motor current sources (INA219 addresses 0x40/0x41) provide most power draw;
Jetson ROS2 battery_node already prioritizes coulomb-based soc_pct from STATUS frame.
Default capacity: 2200 mAh (typical lab 3S LiPo); configurable via firmware parameter.
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
Implements complete I2C1 driver for TI INA219 power monitoring IC supporting:
- Dual sensors on I2C1 (left motor @ 0x40, right motor @ 0x41)
- Auto-calibration for 5A max current, 0.1Ω shunt resistance
- Current LSB: 153µA, Power LSB: 3060µW (20× current LSB)
- Bus voltage: 0-26V @ 4mV/LSB (13-bit, 4mV resolution)
- Shunt voltage: ±327mV @ 10µV/LSB (signed 16-bit)
- Calibration register computation for arbitrary max current/shunt values
- Efficient single/batch read functions (voltage, current, power)
- Alert threshold configuration for overcurrent protection
- Full test suite: 12 passing unit tests covering calibration, conversions, edge cases
Integration:
- ina219_init() called after i2c1_init() in main startup sequence
- Ready for motor power monitoring and thermal protection logic
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
Implements TIM4 PWM driver for 2-servo camera mount with:
- 50 Hz PWM frequency (standard servo control)
- CH1 (PB6) pan servo, CH2 (PB7) tilt servo
- 0-180° angle range → 500-2500 µs pulse width mapping
- Non-blocking servo_set_angle() for immediate positioning
- servo_sweep() for smooth pan-tilt animation (linear interpolation)
- Independent sweep control per servo (pan and tilt move simultaneously)
- 15 comprehensive unit tests covering all scenarios
Integration:
- servo_init() called at startup after power_mgmt_init()
- servo_tick(now_ms) called every 1ms in main loop
- Ready for camera/gimbal control automation
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
Adds STM32F7 STOP-mode power management with <10ms wake latency:
- power_mgmt.c: state machine (ACTIVE→SLEEP_PENDING→SLEEPING→WAKING),
30s idle timeout (PM_IDLE_TIMEOUT_MS), 3s LED fade before STOP,
gate SPI3/I2S3+SPI2+USART6+UART5 on sleep (clock-only, state preserved),
EXTI1(PA1/CRSF)+EXTI7(PB7/JLink)+EXTI4(PC4/IMU) wake sources,
PLL restore after STOP (PLLM=8/N=216/P=2 → 216MHz), uwTick save/restore
- Peripheral gating: I2S3, SPI2(OSD), USART6, UART5 disabled during STOP;
SPI1(IMU), UART4(CRSF), USART1(JLink), I2C1 remain active as wake sources
- Sleep LED: triangle-wave pulse (2s period) on LED1 during SLEEP_PENDING,
software PWM in main loop (1-bit, pm_pwm_phase vs brightness)
- IWDG: fed just before WFI; <10ms wake << 50ms WATCHDOG_TIMEOUT_MS
- JLink: JLINK_CMD_SLEEP=0x09, JLINK_TLM_POWER=0x81 (11-byte power frame
at 1Hz: power_state, est_total_ma, est_audio_ma, est_osd_ma, idle_ms)
- main.c: power_mgmt_init(), activity() on CRSF/JLink/armed, tick() when
disarmed, sleep_req handler, LED PWM, JLINK_TLM_POWER telemetry
- config.h: PM_* constants, PM_CURRENT_*_MA estimates, PM_TLM_HZ
- test_power_mgmt.py: 72 tests passing (state machine, LED, gating,
current estimates, JLink protocol, wake latency, hardware constants)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add 'G' CDC command that disarms and re-runs gyro bias calibration.
safety_refresh() added to calibration loop (every 40ms) so IWDG
does not trip during the 1s blocking re-cal when watchdog is running.
GYRO CAL button in ui/index.html sends 'G' and shows status feedback.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- bmp280.c: detect BME280 (chip_id 0x60) vs BMP280 (0x58) at init
- bmp280.c: read humidity calibration (dig_H1–H6) from 0xA1 and 0xE1–0xE7
- bmp280.c: set ctrl_hum (0xF2, osrs_h=×16) before ctrl_meas — hardware req
- bmp280.c: add bmp280_read_humidity() — float compensation (FPv5-SP FPU),
returns %RH × 10; -1 if chip is BMP280 or not initialised
- bmp280.h: add bmp280_read_humidity() declaration + timeout note
- main.c: baro_ok → baro_chip (stores chip_id for BME280 detection)
- main.c: telemetry adds t (°C×10), pa (hPa×10) for all barometers;
adds h (%RH×10) for BME280 only; alt unchanged
- ui/index.html: hidden TEMP/HUMIDITY/PRESSURE rows, revealed on first
packet containing t/h/pa fields; values shown with 1 dp
I2C hang safety: all HAL_I2C_Mem_Read/Write use 100ms timeouts, so
missing hardware (NAK) returns in <1ms, not after a hang.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Shared I2C1 bus (i2c1.c/h, PB8=SCL PB9=SDA 100kHz):
- i2c1_init() called once in main() before sensor probes.
- hi2c1 exported globally; baro and mag drivers use it directly.
Barometer (bmp280.c):
- Probes I2C1 at 0x76 then 0x77 (covers both SDO options).
- bmp280_init() returns chip_id (0x58/0x60) on success, neg if absent.
- Added bmp280_pressure_to_alt_cm() — ISA barometric formula.
- Added bmp280.h (was missing).
Magnetometer (mag.c / mag.h):
- Auto-detects QMC5883L (0x0D, id=0xFF), HMC5883L (0x1E, id='H43'),
IST8310 (0x0E, id=0x10) in that order.
- mag_read_heading() returns degrees×10 (0–3599) or -1 if not ready.
- HMC5883L: correct XZY byte order applied.
- IST8310: single-measurement trigger mode.
main.c:
- i2c1_init() + bmp280_init() + mag_init() after all other inits.
- Both skip gracefully (baro_ok=0, mag_type=MAG_NONE) if not present.
- Telemetry JSON: incremental builder appends ",\"hd\":<n>" when mag
found and ",\"alt\":<n>" when baro found. No extra bytes when absent.
UI (index.html):
- HEADING and ALT rows hidden until first packet with that field.
- Heading shown in degrees, alt in metres (firmware sends cm).
Closes#24.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
New LED behavior (active-low, PC15=LED1, PC14=LED2):
Disarmed, IMU OK : LED1 solid ON + LED2 off
Armed : LED1 solid ON + LED2 solid ON
Tilt fault : LED1 blink 1Hz + LED2 blink 1Hz
IMU error : LED1 blink 1Hz + LED2 solid ON
Rule: solid = good, slow blink (~1Hz) = needs attention.
Removed the confusing fast-blink-at-5Hz-for-error and the
baro-flash-every-5s patterns.
status_update() signature changed: baro_ok → armed + tilt_fault
so the LED pattern can reflect arm state directly.
Closes#22.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
On boot, before the main loop, sample 1000 gyro readings (~1s) while
board is held still. Compute per-axis mean offset (sensor-frame raw LSBs)
and subtract from all subsequent readings in mpu6000_read().
- mpu6000_calibrate(): LED1+LED2 solid ON during 1s sample window,
resets filter state to zero once bias is known
- mpu6000_is_calibrated(): gate; main loop blocks arming and USB
streaming until calibration completes
- Bias subtracted in sensor frame before CW270 axis transform + scale,
so all three axes (pitch/roll/yaw rate) benefit
- config.h: GYRO_CAL_SAMPLES=1000
- No flash storage — recalibrate fresh each boot (bias varies with temp)
Closes#21 (3.5°/s yaw drift), #23 (gyro bias calibration on boot).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Adds motor_driver.c/h between the balance PID and the raw
hoverboard UART driver:
- Differential drive: balance_cmd → speed, steer_cmd → steer
- Steer-only ramping at MOTOR_STEER_RAMP_RATE (balance PID keeps
full immediate authority — no ramp on speed channel)
- Headroom clamp: reduces steer so |speed|+|steer|<=MOTOR_CMD_MAX
ensuring ESC never clips the balance command
- Emergency stop: latches on TILT_FAULT, clears on BALANCE_DISARMED;
send path stays in 50Hz ESC tick to avoid flooding UART
main.c: replace bare hoverboard_send() with motor_driver_update();
config.h: MOTOR_CMD_MAX=1000, MOTOR_STEER_RAMP_RATE=20 counts/ms
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Issue #12 — Roll displayed as pitch:
- Firmware was sending r=pitch_rate (wrong). Changed to r=roll_deg*10.
- mpu6000.c: add roll complementary filter (accel atan2(ay,az) +
gyro gy integration, same COMP_ALPHA=0.98 as pitch).
- IMUData: add roll and yaw fields.
- UI: updateIMU() now uses data.r/10 for roll (not client-side filter
that computed from ax/ay/az which firmware never sent).
- Three.js: roll -> rotation.z (banking), pitch -> rotation.x (tipping)
— axes were already correct, fix was the firmware data.
Issue #13 — Add yaw telemetry:
- Firmware: gyro Z integration (gz * dt) → s_yaw, sent as y=yaw_deg*10.
Gyro-only, drifts — no magnetometer.
- IMUData: yaw field added.
- UI: yaw -> rotation.y (spinning on vertical axis). Displayed in HUD.
- UI: YAW RESET button captures current yaw as new zero reference
(client-side offset, no new firmware command needed).
Closes#12, #13.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Root cause 1 (IWDG reset loop): safety_init() was called before
mpu6000_init() — IWDG 50ms timeout fires during ~510ms IMU init,
causing infinite MCU reset. Moved safety_init() to after all
peripheral inits (IMU, hoverboard, balance).
Root cause 2 (DCache coherency): USB TX/RX buffers merged into a
single 512B-aligned struct in usbd_cdc_if.c. MPU Region 0 configured
non-cacheable (TEX=1, C=0, B=0) in usbd_conf.c USBD_LL_Init() before
HAL_PCD_Init(). DCache stays ON globally — MPU handles coherency.
Removed SCB_DisableDCache() from main.c (caused boot crash).
Also: fix safety.c IWDG_RELOAD macro (float literals not valid in
#if); add crsf.c stub so crsf_state links (UART not yet wired).
Fixes issue #9.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Safety systems implementation:
IWDG Hardware Watchdog (50ms timeout, config.h WATCHDOG_TIMEOUT_MS):
- safety_init() configures IWDG at PSC/32 (0.8ms tick), reload=62
- safety_refresh() must be called every loop iteration
- Cannot be disabled once started — MCU resets if loop hangs
- Started after 3s USB init delay (avoids spurious startup reset)
Arm Hold Interlock (3s, config.h ARMING_HOLD_MS):
- Arm command starts a hold timer, not immediate motor enable
- Motors only enable after ARMING_HOLD_MS consecutive hold
- Disarm or tilt > 10° cancels pending arm
- Prevents accidental arm from single keypress
Tilt Fault Alert:
- safety_alert_tilt_fault() fires one-shot buzzer on TILT_FAULT edge
- Rider hears alarm when tilt cutoff triggers
- Edge-detected (buzzer only fires once per fault event)
RC Timeout (infrastructure):
- safety_rc_alive() checks crsf_state.last_rx_ms vs RC_TIMEOUT_MS
- RC disarm wired but guarded (no CRSF yet) — remove guard when wired
- Compatible with future CRSF implementation
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Add USB command interface for live PID gain adjustment without reflashing:
P<kp> I<ki> D<kd> T<setpoint_deg> M<max_speed> ?
Command parsing runs in main loop (sscanf-safe), not in USB IRQ.
USB IRQ copies command to shared volatile buffer (cdc_cmd_buf), sets flag.
Acknowledgement echoes current gains: {"kp":...,"ki":...,"kd":...}
Bounds checking: kp 0-500, ki/kd 0-50, setpoint ±20°, max_speed 0-1000.
Gains validated before write — silently ignored if out of range.
Telemetry updated from raw counts to physical tuning signals:
pitch (°x10), pitch_rate (°/s x10), error (°x10),
integral (x10 for windup monitoring), motor_cmd, state
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Bug 1 (PRIMARY — DCache/USB coherency):
SCB_DisableDCache() was buried inside icm42688_init(), called ~3.5s
after USB starts. STM32F7 DCache/USB coherency issue: when DCache is
on (enabled by SystemInit()), CPU writes to TX buffers stay in cache
and the USB hardware reads stale SRAM data. Moved SCB_DisableDCache()
to main() before HAL_Init(), ensuring coherency for all USB transfers.
Bug 2 (USB TX corruption):
CDC_Transmit() passed the caller's stack-allocated buf pointer directly
to the USB stack. The USB TXFE interrupt fires asynchronously; by then
the stack buffer may have been modified by the next loop iteration.
CDC_Transmit() now copies into the static UserTxBuffer before handing
off to the USB hardware, ensuring the buffer is stable for the transfer.
Bug 3 (IMU type mismatch → wrong data to balance):
main.c called icm42688_init()/icm42688_read() directly, passing
icm42688_data_t* (raw int16 ax/ay/az/gx/gy/gz) to balance_update()
which expects IMUData* (float pitch/pitch_rate from complementary
filter). Type mismatch produced garbage balance values. Fixed by using
mpu6000_init()/mpu6000_read() which wraps icm42688 + sensor fusion.
Telemetry updated to report fused pitch/rate instead of raw ADC counts.
Also fix icm42688_init() returning 0 on who==0 (no SPI response),
which falsely indicated IMU success.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
