Obstacles Avoidance Using Reinforcement Learning for Safe Autonomous Navigation

Autonomous agents can safely navigate environments around them when equipped with advanced hardware such as sensors and controlled with advanced Artificial Intelligence (AI). AI is a powerful science that can be employed to provide the highest safety for agents. AI safety is essential to provide reliable services to consumers in various fields such as military, education, health care, and automotive. This paper presents an AI safety algorithm for safe autonomous exploring in a chosen environment. The Reinforcement Learning was used to design the proposed AI safety algorithm. The designed algorithm was tested in virtual reality using the Unity environment. A 0.62% goal collision ratio was achieved, and the collision incidents were minimized from 134 to 54 in the Unity environment within 30 minutes.