SLAM-DRLnav: A SLAM-Enhanced Deep Reinforcement Learning Navigation Framework for Indoor Self-driving

Abstract

This work proposes a new framework for mobile robot navigation that combines two existing methods - SLAM and DRL - to enhance performance. Instead of relying solely on SLAM for mapping and location, or solely on DRL for path planning, the proposed framework utilizes both techniques in conjunction. Specifically, it utilizes SLAM for creating maps and determining the robot’s location, while using an ACO algorithm to generate a static path plan. In dynamic environments with moving obstacles, the framework employs DRL-based local path planning. Additionally, the proposed framework compares and evaluates the performance of three different DRL-based navigation algorithms - DGN, TD3, and PPO. Overall, the goal of this framework is to improve the localization and path planning capabilities of mobile robots in dynamic environments.