Proximal Policy Optimization Based Autonomous Navigation in Dynamic Environment Using LiDAR-Camera Fusion Technique
Abstract
Smart robots are being deployed to autonomously navigate complex and dynamic indoor environments. Autonomous navigation in unknown and dynamic environments is a major challenge for robots, especially when it comes to making safe decisions in complex environment. In this research, we use the Proximal Policy Optimization (PPO) algorithm combined with LiDAR-camera sensor fusion to address this problem. While using only 3D LiDAR or a camera often leads to failure in complex scenes, fusing the two sensors provides a much clearer and more reliable understanding of the environment. This improved perception helps the robot avoid dynamic obstacles and make safer navigation choices. This research results show a clear improvement in both training performance and safety: the robot achieves a 75% average success rate across episodes of training and identifies important environmental features with 80% probability. Overall, this research offers a practical and effective solution for safe autonomous navigation in challenging, complex environments.