DP / PATEL
All Projects

Team Capstone Project — Robotics & Mechatronics

Autonomous Delivery and Retrieval Rover

A team-built 4WD skid-steer rover for autonomous package delivery and retrieval, integrating ToF and IMU sensing with staged failure analysis.

RoboticsEmbeddedSensor FusionControl SystemsValidation

ROVER — FULL PLATFORM, CHASSIS + SENSOR MOUNTS

01

Overview

A three-person capstone project spanning two mission phases: a 4WD skid-steer rover that crosses varied terrain, descends a ramp, delivers a package into a tray, retrieves a second package, and returns — guided by four Time-of-Flight sensors and a 6-DOF IMU on an Arduino Due control stack.

02

Team Project & My Contribution

This was a three-person capstone project built across two development phases. Mechanical design — the chassis, gripper, and drivetrain mounts — was a team effort in CAD and DFM review. My work focused on electrical system integration, embedded software, and the control and sensor-related software running on the rover.

  • Electrical system integration: wiring the Arduino Due, dual TB6612 H-bridge motor drivers, encoder motors, servos, four VL53L1X ToF sensors, and the 6-DOF IMU.
  • Embedded software: the zone-based state machine (Navigate → Deliver → Navigate → Retrieve → Return) and its state re-entry and timeout-recovery logic.
  • Sensor and control software: sequential I2C polling and median filtering across the four ToF sensors, IMU-based heading correction, and encoder-feedback control for wall-following and ramp traversal.
03

System Architecture

Arduino Due

Primary control hardware

Dual TB6612 H-Bridge Drivers

Motor drive interface

4× Micro Gear Encoder Motors (150:1)

4WD skid-steer drivetrain with feedback

Metal-Gear Servos

Gripper actuation, upgraded from SG90 servos after Phase B

Four VL53L1X ToF Sensors

Sequential I2C ranging with median filtering

6-DOF IMU

Heading correction and ramp-tilt monitoring

Custom PLA Chassis (15% infill)

Three-level structure, team CAD design, ~1.5 kg final mass

04

Software / Controls

The rover ran a hierarchical state machine across four zones — terrain navigation, ramp descent, maze delivery/retrieval, and return. Sequential I2C polling and median filtering stabilized the four ToF sensors on uneven terrain, while IMU feedback corrected heading after obstacle avoidance and during ramp climbs. State re-entry logic and timeout-based fallbacks were added after Phase B testing showed the rover could lose track of its position following a manual save-move.

05

Testing

  • Staged testing in three steps: individual sensor/subsystem checks, obstacle-avoidance trials, then full-mission runs varying starting angle, lighting, and obstacle placement.
  • Terrain navigation and obstacle avoidance performed consistently — the rover crossed the terrain zone without drifting.
  • Ramp descent was stable and repeatable, confirming the IMU-assisted ramp-traversal logic carried over correctly from the previous phase.
  • Drivetrain load testing showed stable electrical operation with no excessive heat or current draw; the theoretical torque margin for the 30° incline was not validated as a full climb during final testing.
06

Engineering Challenges

Multi-Sensor Integration

Sequential I2C polling across four ToF sensors to avoid crosstalk on uneven terrain.

Rigid Drivetrain Mounting

Replaced a brass-coupler wheel mount with a direct-screw connection to eliminate slip and improve encoder-based odometry.

State-Machine Recovery

Added re-entry conditions so the state machine could recover position after a manual save-move, addressing a desynchronization issue from the prior test phase.

07

Failure Analysis

The final mission run did not complete the full delivery-and-retrieval sequence, and the causes were mechanical and navigational rather than software:

  • Gripper: two hinge supports were missing from the arm assembly, so it lacked the rigidity to grip the package — a structural build gap, not a servo or control-logic fault.
  • Terrain transition: the rover's wheels caught in the gap between the terrain section and an adjoining metal plate, requiring three manual save-moves to free it.
  • State-machine desynchronization: those save-moves shifted the rover's position enough that its sensor readings no longer matched its expected state, and the state machine halted as a safety fallback before reaching the maze.
  • What did work: sensor initialization, terrain-zone obstacle avoidance, and both ramp descents completed reliably, and FMEA had already flagged navigation, sensor initialization, and drivetrain activation as the highest-risk areas going in.