Trajectory Planning of Differential Drive Robots for Auto-Docking Stations Using LSTM-Based Learning Models

This paper presents the development of an auto-docking system using LSTM-based learning for trajectory planning on a differential drive robot, using data from ArUco markers to predict linear and angular velocities. The model relies on x-axis, y-axis, and yaw orientation data from ArUco to maintain stable navigation towards the auto-docking station. The experimental results show that the training accuracy for linear velocity is 87.2% and for angular velocity is 92.48%, with a Mean Squared Error (MSE) of 0.0003145 m/s for linear velocity and 0.1206 rad/s for angular velocity. In addition, this study also shows the power consumption obtained when the car is automatically docking, with an average battery power consumption of 83.05 watts and an average consumption on all motors of 30.59 watts. The results of this study indicate that the auto-docking system using LSTM-based learning for trajectory planning with ArUco marker position input can be applied to real-world conditions, thus contributing to the advancement of trajectory planning in mobile robots.