Login| Sign Up| Help| Contact|

Patent Searching and Data


Title:
DEEP REINFORCEMENT LEARNING-BASED INTELLIGENT COLLISION-AVOIDANCE METHOD FOR SWARM OF UNMANNED SURFACE VEHICLES
Document Type and Number:
WIPO Patent Application WO/2021/082864
Kind Code:
A1
Abstract:
A deep reinforcement learning-based intelligent collision-avoidance method for a swarm of unmanned surface vehicles (USVs), which belongs to the technical fields of deep reinforcement learning systems and USV intelligent collision-avoidance. Provided is an intelligent collision-avoidance method for an intelligent USV system. First, a deep reinforcement learning-based autonomous learning collision-avoidance theoretical framework for a swarm of USVs is proposed, and memory capabilities of an LSTM neural network are merged to achieve the continuity of collision-avoidance motions. Then, a design representation method for a USV environment in the framework is obtained, such as an environmental observation value, and a USV collision-avoidance reward and punishment function is proposed to determine a collision-avoidance effect. Finally, an intelligent collision-avoidance deep reinforcement learning training system for the swarm of USVs is formed. By means of simulation and verification, it is shown that trained USVs can safely navigate and implement intelligent collision-avoidance in a swarm of USV collision-avoidance environment.

Inventors:
MA YONG (CN)
ZHAO YUJIAO (CN)
WANG YULONG (CN)
Application Number:
PCT/CN2020/119188
Publication Date:
May 06, 2021
Filing Date:
September 30, 2020
Export Citation:
Click for automatic bibliography generation   Help
Assignee:
UNIV WUHAN TECH (CN)
International Classes:
G05D1/02; G05B11/42; G05B13/02; G05B13/04
Foreign References:
CN110658829A2020-01-07
CN108820157A2018-11-16
CN109540151A2019-03-29
CN108710372A2018-10-26
CN109540136A2019-03-29
CN110196605A2019-09-03
US10279474B22019-05-07
Attorney, Agent or Firm:
WUHAN ZHI JIA JOINT INTELLECTUAL PROPERTY AGENCY (GENERAL PARTNERSHIP) (CN)
Download PDF: