SLP-Improved DDPG Path-Planning Algorithm

project lapse time: 2022-2-2 - 2023-3-28

Introduction

For my computer science undergraduate research project, I explored the use of Sequential Linear Paths(SLP) global planning to improve Deep Deterministic Policy Gradient(DDPG) reinforcement learning for robot path planning. The goal of the project was to develop a novel path planning algorithm that could efficiently navigate a robot in complex environments. Our experiments were conducted using ROS, Turtlebot3 Burger, and Gazebo simulation environments.

Project Details

The project consisted of two main parts: global planning using SLP and local planning using DDPG. For the global planning, we used the SLP algorithm proposed in the paper "Enhancing Path Quality of Real-Time Path Planning Algorithms for Mobile Robots: A Sequential Linear Paths Approach" by Fareh et al. The SLP algorithm generated a global path for the robot based on a given map and goal location.

For the local planning, we used the DDPG reinforcement learning algorithm, which is a model-free, off-policy algorithm that can learn to control a robot's motion by directly mapping sensor readings to actions. We used the DDPG algorithm to refine the global path generated by the SLP algorithm, allowing the robot to navigate around obstacles and reach its goal. We evaluated our novel path planning algorithm using a simulated robot in various environments, including open spaces and cluttered environaments. We compared our algorithm's performance with that of other state-of-the-art path planning algorithms.

Results

Our novel path planning algorithm achieved promising results in various environments, outperforming other state-of-the-art algorithms in some scenarios. We found that the combination of SLP global planning and DDPG reinforcement learning allowed the robot to navigate efficiently and effectively in complex environments while avoiding collisions with obstacles.

Conclusion

Our research project demonstrated the effectiveness of using SLP global planning to improve DDPG reinforcement learning for robot path planning. Our novel path planning algorithm can efficiently navigate a robot in complex environments, making it a valuable tool for various applications, such as autonomous vehicles, mobile robots, and unmanned aerial vehicles. Our research paper has been published in the Sensors journal, and we hope that our work will inspire further research in this area.