Simulation Analysis of Multi-Objective Functions in Mobile Robot Navigation Based on Enhanced Deep Q-Network Algorithm

2024-01-5110

To be published on 12/24/2024

Features
Event
Automotive Technical Papers
Authors Abstract
Content
This research, path planning optimization of the deep Q-network (DQN) algorithm is enhanced through integration with the enhanced deep Q-network (EDQN) for mobile robot (MR) navigation in specific scenarios. This approach involves multiple objectives, such as minimizing path distance, energy consumption, and obstacle avoidance. The proposed algorithm has been adapted to operate MRs in both 10 × 10 and 15 × 15 grid-mapped environments, accommodating both static and dynamic settings. The main objective of the algorithm is to determine the most efficient, optimized path to the target destination. A learning-based MR was utilized to experimentally validate the EDQN methodology, confirming its effectiveness. For robot trajectory tasks, this research demonstrates that the EDQN approach enables collision avoidance, optimizes path efficiency, and achieves practical applicability. Training episodes were implemented over 3000 iterations. In comparison to traditional algorithms such as A*, GA, and ACO, as well as deep learning algorithms (IDQN and D3QN), the simulation and real-time experimental results showed improved performance in both static and dynamic environments. The results indicated a travel time reduction to 9 s, a 14.6% decrease in total path distance, and a training duration reduction of 1657 iterations compared to IDQN and D3QN.
Meta TagsDetails
Pages
11
Citation
Arumugam, V., Alagumalai, V., and Rajendran, S., "Simulation Analysis of Multi-Objective Functions in Mobile Robot Navigation Based on Enhanced Deep Q-Network Algorithm," SAE Technical Paper 2024-01-5110, 2024, .
Additional Details
Publisher
Published
To be published on Dec 24, 2024
Product Code
2024-01-5110
Content Type
Technical Paper
Language
English