Path Planning of a Mobile Robot for a Dynamic Indoor Environment Based on an SAC-LSTM Algorithm

Area of Science:

• Robotics
• Artificial Intelligence
• Machine Learning

Background:

• Mobile robot navigation in dynamic environments presents challenges due to unpredictable changes.
• Existing path planning algorithms may struggle with continuous motion and memory retention.
• Soft Actor-Critic (SAC) is a reinforcement learning algorithm suitable for continuous control tasks.

Purpose of the Study:

• To propose an improved Soft Actor-Critic Long Short-Term Memory (SAC-LSTM) algorithm for fast and efficient path planning.
• To enhance decision-making capabilities by integrating historical and current states using LSTM.
• To address memory depreciation issues and improve training efficiency.

Main Methods:

• Integration of Long Short-Term Memory (LSTM) network into the Soft Actor-Critic (SAC) algorithm.
• Implementation of a burn-in training method to mitigate LSTM hidden state resetting issues.
• Adoption of a prioritized experience replay mechanism for enhanced sampling efficiency and faster convergence.
• Development of a motion model for Turtlebot3, including state/action space, reward function, and planning process.
• Simulation experiments in obstacle-free, static, and dynamic environments using ROS and Gazebo9.

Main Results:

• The SAC-LSTM algorithm demonstrated a faster convergence rate compared to the standard SAC algorithm across all tested environments.
• A significant 10.5 percentage point improvement in path planning success rate was observed in the dynamic obstacle environment.
• The SAC-LSTM algorithm resulted in shorter path planning times and more concise planned paths.
• Improved performance in handling continuous motion and incorporating historical state information.

Conclusions:

• The proposed SAC-LSTM algorithm offers superior performance for mobile robot path planning, particularly in complex and dynamic environments.
• The integration of LSTM, burn-in training, and prioritized experience replay effectively enhances path planning efficiency and success rates.
• This approach provides a robust solution for real-time navigation challenges faced by mobile robots.