Implementation of a Long Short-Term Memory Neural Network-Based Algorithm for Dynamic Obstacle Avoidance

Area of Science:

• Robotics
• Artificial Intelligence
• Machine Learning

Background:

• Autonomous mobile robots are crucial in industrial settings, necessitating safe human-robot interaction.
• Safe navigation through environments with static and dynamic obstacles is a key challenge for these robots.

Purpose of the Study:

• To present a physical implementation of a dynamic obstacle avoidance method for mobile robots.
• To utilize a long short-term memory (LSTM) neural network for real-time collision avoidance.

Main Methods:

• A TurtleBot3 robot equipped with LiDAR was used within an OptiTrack motion capture system.
• LiDAR data, target point, robot position, and velocities were collected as input for the LSTM network.
• The LSTM model was trained on diverse user-operated trajectories across multiple scenarios.

Main Results:

• The implemented LSTM model enabled the mobile robot to successfully reach target points in all tested scenarios.
• The system demonstrated effective avoidance of dynamic obstacles.
• A validation accuracy of 98.02% was achieved in physical experiments.

Conclusions:

• The physical implementation of the LSTM-based dynamic obstacle avoidance method is successful.
• This approach enhances the safety and efficiency of autonomous mobile robots in complex environments.
• The model shows high reliability in real-world navigation tasks.