SLAYO-RL: A Target-Driven Deep Reinforcement Learning Approach with SLAM and YoLo for an Enhanced Autonomous Agent

This article presents an innovative approach for training an agent to reach a specific and predetermined target in an unknown environment. It uses reinforcement learning for an agent with a Lidar sensor and a camera. Given the difficulty of using raw high-dimensional information to train any reinforcement learning agent, the Lidar sensor data was processed using Simultaneous Localization and Mapping to provide the agent's location in space. To identify the agent's target of interest, the camera image was processed using the YoLo object detection model to provide the coordinates of the target in the image. In addition to processing the agent's state, the two technologies were used as a composition of the reward obtained by the agent, causing it to develop the behavior of exploring an unknown environment and, after locating the target, moving towards it until the agent collides with the target. The proposed approach differs from the state of the art because it unites the two technologies as a composition of the agent's state and reward.