Self-supervised Learning Approach to Local Trajectory Planning for Mobile Robots Using Optimization of Trajectories

Abstract:

In Industry 4.0, various control methods have been developed for autonomous navigation of robots. Some investigations are based on the use of SLAMs or routing systems for route tracking, but there are some limitations when it comes to obstacle avoidance and real-time parameter changes. Current work shows an algorithm based on the use of DQN and reinforcement learning. The model maximizes rewards and extracts information about the robot’s position and obstacles within the simulated environment as the robot performs its actions. A series of experiments have been conducted to build the algorithm, and the results show that the robot learns through exploration and uses the knowledge gained in the previous scenarios. Using a simulated environment, the DQN network computes complex functions due to randomness, resulting in better learning performance than other control methods.

Año de publicación:

2023

Keywords:

• robotics
• reinforcement learning
• obstacle avoidance

Fuente:

scopus