Home > Published Issues > 2020 > Volume 9, No. 8, August 2020 >

Autonomous Navigation for Omnidirectional Robot Based on Deep Reinforcement Learning

Van Nguyen Thi Thanh 1, Tien Ngo Manh 2, Cuong Nguyen Manh 2, Dung Pham Tien 2, Manh Tran Van 2, Duyen Ha Thi Kim 3, and Duy Nguyen Duc 3
1. Hanoi Vocational College of High Technology, Hanoi, Vietnam
2. Institute of Physics, Vietnam Academy of Science and Technology, Hanoi, Vietnam
3. Hanoi University of Industry, Hanoi, Vietnam

Abstract—This paper presents navigation for four-wheel Omni Robot using architecture deep reinforcement learning in an unknown environment. A deep reinforcement learning algorithm combines with effectively training data using stochastic gradient updates in order to connect a goal. The approach is simulated and visualized using Gazebo and is updated via policies trained by deep Q learning network. Using recent deep-learning techniques as the basis of the framework, our results indicate that it is capable of providing smooth navigation for the Omni robot in exploring unpredicted surroundings. Once extended to real-world operation, this framework could enable the Omni Robot to gain achievement for self-driving tasks. 

Index Terms—deep reinforcement learning, omni robot, Robot operating system, navigation

Cite: Van Nguyen Thi Thanh, Tien Ngo Manh, Cuong Nguyen Manh, Dung Pham Tien, Manh Tran Van, Duyen Ha Thi Kim, and Duy Nguyen Duc, "Autonomous Navigation for Omnidirectional Robot Based on Deep Reinforcement Learning," International Journal of Mechanical Engineering and Robotics Research, Vol. 9, No. 8, pp. 1134-1139, August 2020. DOI: 10.18178/ijmerr.9.8.1134-1139