Steering Servo

Steering is done by a single servo that turns the Ackermann linkage. The Raspberry Pi 5 does not generate the servo pulse directly; it sends an angle command over I2C to a PCA9685 Servo Controller, which generates the PWM signal for the steering servo.

Parts (Amazon)

Wiring and Settings

From config.py / hardware.py:

Setting Value
Bus I2C, address 0x40
Servo channel 0
PWM frequency 50 Hz
Pulse range 1,000-2,000 microseconds
Steering-command range 180 degrees

How It Works

  • Logical steering is 0 = left, 90 = center, 180 = right. The runtime thinks and logs in these logical/reference degrees; the model and training labels never use the trimmed physical servo command.
  • hardware.py first maps the logical 0..180 range onto the current reference endpoints (48.812..131.188), then applies the current +17 degree center trim before writing the physical command through adafruit_servokit. Center preload is present as a code path but both preload constants are currently zero.
  • The runtime clamps the logical target, stores it for logs/dashboard use, and snaps commands within 0.5 degrees of logical center to exactly 90 before the hardware mapping. With the current settings, logical center 90 produces a physical command near 107 degrees.
  • Bench work observed direction-dependent return behavior: the wheels did not always return to the same physical center after approaching from opposite directions. The runtime now includes an experimental IMU yaw-rate correction path in default straight mode; field claims still require telemetry showing that it engaged. No exact hysteresis angle is claimed without the corresponding measurement record.

Hysteresis and Pull Diagnosis

The same logical center can settle differently after a left approach versus a right approach. Possible contributors include joint play, linkage flex, servo return behavior, vehicle load, surface conditions, or unequal motor force. The software keeps these controls separate:

  • Servo trim changes front-wheel angle.
  • Left and right motor scale changes drive force; both current scales are 1.0.
  • Experimental yaw feedback changes steering from measured rotation.

When the car pulls, repeat the test on the same surface, payload, tire state, battery state, and throttle. Verify physical wheel center from both approach directions before changing a motor scale. Change one control at a time so one fault is not hidden by compensating for it elsewhere. A planned pass-through-versus-flick-to-center test remains useful because joystick history may affect how the linkage returns even when the final command is 90.

Why This Choice

  • The PCA9685 Servo Controller generates the 50 Hz servo pulse in dedicated hardware and offloads pulse timing from the Raspberry Pi 5. This reduces dependence on control-loop timing, but does not remove power, I2C, calibration, servo, or linkage sources of jitter.
  • Keeping all servo-specific compensation in the mapping layer keeps the logical steering labels clean for training and for the dashboard.

Verify Before a Run

  • Bench tests: code/test_files/steering/pca9685_servo_test.py sweeps the servo, and code/test_files/steering/calibrate_servo.py helps set the center trim.
  • Failure symptoms: the servo not moving at all can come from I2C, servo power, pulse configuration, wiring, or the servo/linkage itself (confirm whether 0x40 is visible). An off-center rest position can come from trim, linkage, installation, or directional hysteresis; it should not be assigned to one cause without a bench check. Do not finalize trim constants unless the exact number has been characterized.