A Performance Evaluation of A∗ and Dijkstra Algorithms with Visual-Based Goal Selection

Abstract

Successful operation of autonomous robots in complex environments require both reliable perception and efficient path planing. Reliable perception provide the foundation of scene understanding, which enables robots to understand their environment correctly and operate effectively. This study focuses on two main objectives: generation of a spatial map of the environment and extract the coordinates of bussinesses using LiDAR and visual perception and evaluation-comparison the performance of A∗ and Dijkstra path planning algorithms based on these perception-driven destination points. As the part of visual perception pipeline, we perform text detection and recognition where the coordinates of bussinesses were extracted. These extracted coordinates then are used as destination points for path planning algorithms. The experiments were conducted in the Gazebo simulation environment using a TurtleBot3 Waffle robot platform. The performance of path planning algorithms is evaluated based on following metrics; distance to the target, time to reach the destination, energy consumption, CPU load, and RAM usage. The experimental results shows that A∗ outperforms Dijkstra algorithm in terms of shorter travel time, lower energy consumption and CPU usage.