Evaluating the Efficacy of Depth Estimation in Obstacle Detection Systems in Simulated Environment

This paper explores an approach to utilize state of the art (SOTA) depth estimation models for obstacle Detection. Detecting obstacles is a fundamental challenge in the field of autonomous robotics, where accurate depth perception is crucial for safe and reliable navigation. Traditional depth estimation methods often rely on stereo cameras that calculates depth through the disparity between the two captured images, or LiDAR (Light Detection and Ranging) sensors that employs pulsed laser light to measure distances through time-of-flight calculations. Despite their accuracy, both methods can be expensive and have weight limitations. In contrast, a predictive approach using deep learning to infer depth information from monocular images can offer a cost-effective and versatile solution for obstacle detection and safety navigation for autonomous robots. This research aims to validate the efficacy of such models in simulated scenarios, with the intention of offering a more accessible approach to navigation in the field of robotics.