Multi-sensor autonomous mobile robot with finite state machine architecture. Fuses Pixy2 vision, IR proximity, ultrasonic ranging, and quadrature encoders for obstacle avoidance and cue-based navigation on a custom differential-drive platform.

Overview

This project is a ground-up autonomous mobile robot built on an Arduino Mega with a custom differential-drive chassis. The robot navigates unknown environments by fusing data from four distinct sensor modalities: a Pixy2 camera for color-based cue recognition, IR proximity sensors on each flank, an ultrasonic ping sensor for frontal ranging, and quadrature wheel encoders for odometry and heading control. A finite state machine coordinates all behaviors — from straight-line driving with encoder-based correction to obstacle avoidance and precise encoder-metered turns.

The system demonstrates core mobile robotics principles at the hardware level: real-time multi-sensor polling, interrupt-driven encoder counting, kinematic modeling for differential-drive turning, and reactive behavior arbitration — all running on bare-metal embedded C++ with no operating system.

Demo

System Architecture

The robot’s control flow follows a sense-plan-act loop running on the Arduino’s main loop():

stateDiagram-v2
    [*] --> move_forward
    move_forward --> obstacle_avoidance : front + side blocked
    move_forward --> left_turn : Pixy2 sig 1 detected
    move_forward --> right_turn : Pixy2 sig 2 detected
    move_forward --> u_turn : Pixy2 sig 3 detected
    obstacle_avoidance --> left_turn : left clearer
    obstacle_avoidance --> right_turn : right clearer
    left_turn --> move_forward : turn complete
    right_turn --> move_forward : turn complete
    u_turn --> move_forward : 180° reached (encoder)
    move_forward --> idle : stop command
    idle --> move_forward : 4s timeout

Sensor Fusion

The robot polls four sensor types every loop iteration, each serving a distinct role in the navigation pipeline:

Ultrasonic Ping Sensor (front) — Measures forward clearance using time-of-flight. The ping pin toggles between output (trigger) and input (echo), and the round-trip duration converts to centimeters via $d = \frac{t \times 0.034}{2}$. This provides the primary “stop or go” signal.

IR Proximity Sensors (left & right) — Analog Sharp-type IR sensors on A0 and A1 measure lateral clearance. Raw ADC values convert to distance through the inverse-power model $d = 27.86 \cdot V^{-1.15}$, derived from the sensor’s characteristic voltage-distance curve. When either side reads below the 12.5 cm threshold simultaneously with a frontal obstruction, the FSM transitions to obstacle_avoidance.

Pixy2 Camera — Runs the Color Connected Components (CCC) algorithm to detect trained color signatures in the field of view. Each signature maps to a navigation command: signature 1 triggers a left turn, signature 2 a right turn, and signature 3 a U-turn. The block width threshold (pixyDist = 50 pixels) ensures the robot only responds when close enough to the cue, preventing premature reactions to distant markers.

Quadrature Encoders — Two Pololu 25D gearmotors with 48 CPR encoders provide wheel odometry through hardware interrupts on pins 2, 3, 5, and 6. The ISRs decode the A/B channel phase relationship to determine rotation direction: if channel A leads channel B, the wheel rotates clockwise (decrement); if B leads A, counterclockwise (increment). Encoder counts feed two control loops: differential drive correction for straight-line tracking, and precise angular turns via kinematic conversion.

Differential Drive Kinematics

Precise turning relies on converting a desired heading change to encoder counts through the robot’s kinematic model. For a differential-drive robot with wheel base $L$ and wheel circumference $C$:

where $L = 21.7$ cm (track width), $C = \pi \times 6.5$ cm (wheel circumference), CPR $= 48$, and $\theta$ is the desired turn angle in degrees. For a 180° U-turn, this yields approximately 155 encoder counts on the pivot wheel.

The adjustWheelSpeeds() function implements a proportional correction that compares left and right encoder counts. When the difference exceeds a 15-count threshold, the faster wheel decelerates until the counts converge — a simple but effective approach to straight-line tracking without a full PID controller.

Hardware Platform

Results

The robot successfully navigates structured environments with color-coded turn markers and dynamic obstacles. Key outcomes:

Links

• Source Code