An Intelligent Planning Method For The Multi-Rotor Manipulation Robot With Reinforcement Learning

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

In this paper, an intelligent planning method based on proximal policy optimization algorithm for the multi-rotor aerial manipulation robot is presented. This method can not only avoid the complexity of the dynamic analysis and modeling but also the large disturbance of the manipulator to the robot's body produced by the independent kinematic planning method. The disadvantage of kinematic planning method in aerial manipulation is given. The detailed structure of the adopted training and simulation environment is introduced. A deep reinforcement learning formulation is proposed to deal with the aerial manipulation of the robot. The particulars of setup in training and simulation are illustrated and the practical training and a series tests are carried out. The results of simulation proved the feasibility of this intelligent planning method and its advantages in real-time planning or replanning to enhance the stability of the robot in manipulation.

Original languageEnglish
Title of host publication2022 IEEE International Conference on Mechatronics and Automation, ICMA 2022
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages1028-1033
Number of pages6
ISBN (Electronic)9781665408523
DOIs
StatePublished - 2022
Event19th IEEE International Conference on Mechatronics and Automation, ICMA 2022 - Guilin, Guangxi, China
Duration: 7 Aug 202210 Aug 2022

Publication series

Name2022 IEEE International Conference on Mechatronics and Automation, ICMA 2022

Conference

Conference19th IEEE International Conference on Mechatronics and Automation, ICMA 2022
Country/TerritoryChina
CityGuilin, Guangxi
Period7/08/2210/08/22

Keywords

  • Aerial Manipulation
  • Reinforcement Learning
  • Robot Planning
  • UAV

Fingerprint

Dive into the research topics of 'An Intelligent Planning Method For The Multi-Rotor Manipulation Robot With Reinforcement Learning'. Together they form a unique fingerprint.

Cite this