PD Line-Follower Robot

CONTROL DESIGN

Weighted-centroid position estimate. Each of the 5 channels is normalized against its own calibrated min/max span into a 0-1000 “on-line” value, then combined into a single 0-4000-style position using fixed QTR-style weights (0, 1000, 2000, 3000, 4000) — 2000 is dead-center:

uint32_t weightedSum = 0, valueSum = 0;
for (uint8_t i = 0; i < NUM_SENSORS; i++) {
  int32_t norm = (raw[i] - sensorMin[i]) * 1000L / (sensorMax[i] - sensorMin[i]);
  uint16_t onLine = 1000 - constrain(norm, 0, 1000);   // invert: dark line -> high weight
  weightedSum += (uint32_t)onLine * SENSOR_WEIGHTS[i];
  valueSum    += onLine;
}
bool lineSeen = valueSum >= LOST_THRESHOLD;
uint16_t position = weightedSum / valueSum;             // 0..4000, 2000 = centered

PD steering + slew-limited mixing. Position error (position - 2000) drives a PD term; the term steers a fixed base speed differentially, saturates to the PWM range, then is slew-limited so the actual motor command only moves SLEW_STEP counts per 10 ms loop tick — no instant jumps even on a sharp error step:

float steer = KP * error + KD * (error - lastError) / dt;
int16_t leftTarget  = constrain(BASE_SPEED - steer, -MAX_SPEED, MAX_SPEED);
int16_t rightTarget = constrain(BASE_SPEED + steer, -MAX_SPEED, MAX_SPEED);
leftCmd  = slewToward(leftCmd,  leftTarget,  SLEW_STEP);
rightCmd = slewToward(rightCmd, rightTarget, SLEW_STEP);

Lost-line recovery. When valueSum falls below LOST_THRESHOLD (no channel sees the line), the sign of the last nonzero error is remembered and the robot spin-searches toward that side at a fixed PWM until the line reappears or LOST_TIMEOUT_MS elapses, at which point it slews back to a stop instead of spinning indefinitely.

SIMULATION

Runs in Wokwi against sketch.ino + diagram.json (sources/electronics/line-follower/). Hold the calibration button at boot to run the sweep, then watch mode / pos / err / L / R on the 115200-baud serial monitor. Wokwi has no IR-reflectance, DC-motor, or L298N part, so the diagram substitutes honestly:

• TCRT5000 module → wokwi-photoresistor-sensor — same VCC/GND/DO/AO 4-pin footprint, but it’s a light sensor, not an IR-reflectance sensor. Each channel’s light level is set by hand (a slider in the Wokwi UI) to fake a line passing under the array — there’s no simulated floor to reflect off.
• L298N + 2 DC motors → 6 LEDs on ENA/IN1/IN2/IN3/IN4/ENB — Wokwi ships no H-bridge or DC-motor part. The LED bank verifies the logic only: which direction bit is set per motor and the PWM brightness standing in for commanded speed. No current, stall, or back-EMF is modeled.

What the sim actually proves: the calibration state machine, the weighted-centroid and normalization math, the PD law and slew-limited mixing, and the lost-line spin-search/timeout — all exercised with synthetic sensor input. What it can’t prove: real line tracking (no simulated floor/line exists), motor torque/current/stall behavior, sensor height/spacing effects, or real PWM-to-wheel-speed mapping. Those need the physical chassis. Full substitution table and step-by-step instructions are in sources/electronics/line-follower/README.md.

STATUS

Design and firmware are complete and exercised in simulation: calibration, centroid math, PD steering, slew-limited mixing, and lost-line recovery all run and produce sane telemetry against synthetic sensor input in Wokwi. Not done: the chassis has not been built, no sensor array has been mounted on a real robot, no motor has been driven by an actual L298N, and no gains have been tuned against a real step response. KP, KD, BASE_SPEED, and SLEW_STEP are starting points carried over from the control design, not measured/tuned values — this is a simulation-verified control design awaiting hardware, not a working robot.

USE CASES & APPLICATIONS

The same weighted-centroid + PD structure underlies factory-floor AGVs and warehouse tow-tractors that follow painted or taped guide lines, and it’s the standard entry-level platform for robotics courses and competitions (line-following, then maze-solving). More broadly, it’s a small, tunable example of the sensor-fusion pattern — combine several noisy single-axis readings into one continuous estimate before closing a loop on it — that scales up to lane-centering and multi-sensor localization on larger mobile platforms.

FILES

• sources/electronics/line-follower/sketch.ino — full firmware: calibration sweep, weighted- centroid position, PD steering, slew-limited mixing, lost-line spin-search, serial telemetry. Pinout documented in the header comment.
• sources/electronics/line-follower/diagram.json — Wokwi circuit diagram (Arduino Nano, 5× photoresistor-sensor stand-ins, pushbutton, 6-LED H-bridge/motor stand-in bank).
• sources/electronics/line-follower/README.md — Wokwi run steps, full substitution table, and what the simulation can/can’t validate, plus real-build notes (power jumper, grounding, sensor mount height, re-calibration).