Experimental Evaluation of SLAM Performance Under Computational Constraints for Agricultural Indoor Facilities

Abstract

The paper considers the problem of implementing a simultaneous localization and mapping (SLAM) system for mobile robots with limited computing resources in an agro-industrial complex. The high cost of industrial navigation solutions makes them economically impractical for automating monitoring tasks in greenhouses, warehouses, and other agricultural facilities characterized by extended geometry and limited lighting. An architecture based on a Raspberry Pi 4 single-board computer, an ESP32 microcontroller and, an LDROBOT LiDAR integrated into the ROS 2 ecosystem is proposed. A comparison of the SLAM Toolbox and Cartographer algorithms was performed, taking into account the computational and thermal limitations of the platform. Experiments in conditions simulating the configuration of extended indoor spaces have shown that the SLAM Toolbox ensures the metric consistency of the map, while Cartographer demonstrates systematic small-scale drift. The resulting system operates in real time with CPU utilization reaching up to 23% and temperatures peaking at 51°C, demonstrating its potential for autonomous monitoring of agricultural facilities without requiring expensive equipment.