An Embarrassingly Pragmatic Introduction to Vision-based Autonomous Robots

Abstract: Autonomous robots are currently one of the most popular Artificial Intelligence problems, having experienced significant advances in the last decade, from Self-driving cars and humanoids to delivery robots and drones. Part of the problem is to get a robot to emulate the perception of human beings, our sense of sight, replacing the eyes with cameras and the brain with mathematical models such as Neural Networks. Developing an AI able to drive a car without human intervention and a small robot to deliver packages in the city may seem like different problems, nevertheless from the point of view of perception and vision, both problems have several similarities. The main solutions we currently find focus on the environment perception through visual information using Computer Vision techniques, Machine Learning, and various algorithms to make the robot understand the environment or scene, move, adapt its trajectory and perform its tasks (maintenance, exploration, etc.) without the need for human intervention. In this work, we develop a small-scale autonomous vehicle from scratch, capable of understanding the scene using only visual information, navigating through industrial environments, detecting people and obstacles, or performing simple maintenance tasks. We review the state-of-the-art of fundamental problems and demonstrate that many methods employed at small-scale are similar to the ones employed in real Self-driving cars from companies like Tesla or Lyft. Finally, we discuss the current state of Robotics and autonomous driving and the technological and ethical limitations that we can find in this field.