Design and Implementation of a Vision Integrated Autonomous Research Rover

Abstract

This paper presents the design and implementation of a low-cost, vision-integrated autonomous rover for automated soil data collection. Traditional soil sampling methods are labor-intensive, and existing robotic solutions frequently depend on costly hardware and Global Positioning System (GPS) navigation, which can easily degrade in obstructed environments. To address these issues, a GPS-denied rover was developed using a Raspberry Pi 4B for high-level computer vision and a Robot Operating System 2 (ROS2) framework, paired with an Arduino Nano for real-time hardware control. The platform utilizes a linear actuator to autonomously deploy sensing probes for localized soil analysis. For stable navigation, an Extended Kalman Filter (EKF) is implemented to fuse wheel encoder odometry with inertial data. An OpenCV-based vision pipeline directs autonomous path following and marker detection. To mitigate tracking errors caused by processing and mechanical latency, a custom delay-compensation filter is introduced to actively reverse the rover and correct overshoot. This is paired with a history-dependent sweep prioritization algorithm that intelligently re-acquires lost paths based on prior trajectory data. The resulting system provides a reliable, accessible framework for independent environmental monitoring.