Skip to content

calibration.md

Latest commit

 

History

History
63 lines (49 loc) · 3.2 KB

calibration.md

File metadata and controls

63 lines (49 loc) · 3.2 KB
Raw

Calibration

Camera (critical)

  1. Mechanically lock the Raspberry Pi Camera facing forward. Do not command the SG90 pan/tilt while localization is running.

  2. Print a rigid checkerboard with accurately measured squares.

  3. Install ros-jazzy-camera-calibration.

  4. Mono example:

    ros2 run camera_calibration cameracalibrator \
  --size 8x6 --square 0.025 image:=/camera/image_raw

  5. Fill the field of view with near/far and tilted samples. Target reprojection error below 0.5 px, save YAML under config/, and use the same intrinsics in the camera driver and ORB-SLAM3 adapter.

  6. Measure the rigid base_link → camera_link translation/rotation and update ros2_ws/src/navigen_description/config/vehicle.yaml.

Changing focus, resolution, lens, mount, or camera orientation invalidates calibration.

MPU6050

Connect the MPU6050 to the Pi's 3.3 V I2C bus and mount it rigidly with known axes.

  1. Enable I2C and verify the device address with i2cdetect -y 1.
  2. Keep the complete UGV stationary and level; record at least 60 s of /imu/data.
  3. Gyro bias is mean angular velocity. Acceleration bias is mean measured acceleration minus the gravity vector in the mounted orientation.
  4. Enter driver bias/axis settings and verify stationary angular rates are near zero. Measure, rather than promise, yaw drift.
  5. For ORB-SLAM3 mono-inertial mode, estimate noise density/random walk (preferably Allan variance) and verify camera-to-IMU time alignment and rigid extrinsics.

Encoderless wheel/PWM calibration

There are no wheel encoders, so there is no encoder calibration or wheel-speed PID. /motor/telemetry setpoints are requested wheel-surface speed; PWM is applied effort, not measured velocity.

  1. With wheels on stands and a 0.05 m/s command, increase MIN_EFFECTIVE_PWM from zero only until all motors reliably start. Keep it as low as practical.
  2. Verify forward/reverse polarity. Use MOTOR_LEFT_INVERTED and MOTOR_RIGHT_INVERTED; never alter kinematic signs in the bridge.
  3. On a clear floor with a safety operator, mark a measured straight distance and time several low-speed runs. Adjust MAX_WHEEL_VELOCITY_MPS so command-to-PWM scaling is approximately useful at the current battery voltage.
  4. If one side consistently runs faster, reduce only that side using LEFT_PWM_SCALE or RIGHT_PWM_SCALE (valid range >0 to 1). Repeat forward and reverse tests.
  5. Measure wheel radius and effective skid-steer track width and keep the ROS description/hardware YAML consistent. These values convert Twist to targets; they do not create odometry.

Open-loop calibration changes with battery charge, terrain, payload, temperature, and motor wear. It cannot provide accurate dead reckoning. The real autonomous stack must use visual-inertial pose feedback and conservative closed-loop Nav2 body motion.

If encoders are added later

Encoders require a different reviewed controller/pin profile and firmware path. Do not set fake ticks-per-revolution values or derive /wheel/odom from commanded PWM. A later encoder profile must measure direction-aware ticks, validate distance/yaw, and reintroduce closed-loop speed PID with dedicated tests before its odometry is fused.